Split (1)

train · 521k rows

text stringlengths 254 1.16M
--- title: Rescue of High Glucose Impairment of Cultured Human Osteoblasts Using Cinacalcet and Parathyroid Hormone authors: - V. A. Shahen - A. Schindeler - M. S. Rybchyn - C. M. Girgis - B. Mulholland - R. S. Mason - I. Levinger - T. C. Brennan-Speranza journal: Calcified Tissue International year: 2023 pmcid: PMC10025212 doi: 10.1007/s00223-023-01062-7 license: CC BY 4.0 --- # Rescue of High Glucose Impairment of Cultured Human Osteoblasts Using Cinacalcet and Parathyroid Hormone ## Abstract Patients with type 2 diabetes mellitus (T2DM) experience a higher risk of fractures despite paradoxically exhibiting normal to high bone mineral density (BMD). This has drawn into question the applicability to T2DM of conventional fracture reduction treatments that aim to retain BMD. In a primary human osteoblast culture system, high glucose levels (25 mM) impaired cell proliferation and matrix mineralization compared to physiological glucose levels (5 mM). Treatment with parathyroid hormone (PTH, 10 nM), a bone anabolic agent, and cinacalcet (CN, 1 µM), a calcimimetic able to target the Ca2+-sensing receptor (CaSR), were tested for their effects on proliferation and differentiation. Strikingly, CN+PTH co-treatment was shown to promote cell growth and matrix mineralization under both physiological and high glucose conditions. CN+PTH reduced apoptosis by 0.9-fold/0.4-fold as measured by Caspase-3 activity assay, increased alkaline phosphatase (ALP) expression by 1.5-fold/twofold, increased the ratio of nuclear factor κ-B ligand (RANKL) to osteoprotegerin (OPG) by 2.1-fold/1.6-fold, and increased CaSR expression by 1.7-fold/4.6-fold (physiological glucose/high glucose). Collectively, these findings indicate a potential for CN+PTH combination therapy as a method to ameliorate the negative impact of chronic high blood glucose on bone remodeling. ## Introduction Type 2 diabetes mellitus (T2DM) is in the top ten causes of death globally [1], being associated with an increased risk of cardiovascular disease, stroke, cancer, and chronic kidney disease. The effects of T2DM on bone health are often underestimated or overlooked. Regardless of obesity or altered bone mineral density (BMD) as independent risk factors [2], bone microarchitecture and bone quality can deteriorate in patients with T2DM, leading to adverse skeletal events [3–6]. T2DM is associated with an increased fracture risk, with the relative impact reported to range from + $20\%$ to + $300\%$ [7]. While bone complications and fracture do not represent a primary health concern in T2DM, they add to the constellation of factors that need to be considered in terms of clinical management and healthcare costs. The effects of T2DM on bone are often paradoxical, with the apparent bone fragility not associated with common markers of fracture risk such as BMD or the fracture risk assessment tool, FRAX, score [8–10]. Thus, while the mechanism(s) of impaired diabetic bone health are unclear, they are likely to result from the complex hyperglycemic and proinflammatory states that manifest in T2DM and can adversely influence the bone microenvironment [11]. Hyperglycemic conditions can facilitate an excess of glycosylation reactions, leading to the formation of advanced glycation end products (AGEs) [12]. This has the potential to inhibit osteoblast differentiation, suppress remodeling, and lead to a stiffening of the bone matrix [13, 14]. This can be exacerbated by oxidative stress and increased levels of proinflammatory cytokines that can accompany T2DM [15, 16]. There is no established best practice for managing the diabetic bone phenotype. Radiographs, BMD metrics, and circulating bone markers are typically only assessed following spontaneous and/or low-impact fracture(s). Even if these measures are atypical, there is limited evidence to support intervention using classical pharmacotherapies. Many anti-osteoporotic agents are bone antiresorptives, and it is unclear whether these would be beneficial in T2DM where bone turnover is already suppressed [4, 6] and BMD can be in the normal range. A 2021 interdisciplinary expert panel highlighted the need for clinical trials to examine the efficacy and safety of available anti-osteoporotic drugs in patients with diabetes [17]. The panel also considered the use of osteoanabolic agents, such as parathyroid hormone (PTH1–84) and teriparatide (PTH1–34). These systemic interventions for osteoporosis can promote new bone formation [18, 19]; however, the treatment window for PTH is limited, with most courses lasting 18–24 months. Whether PTH and teriparatide is clinically useful in T2DM remains to be determined. Cinacalcet (CN) is a calcimimetic drug that can act as an allosteric modulator of the calcium-sensing receptor (CaSR) [20]. Although CN is mostly used to treat patients with hypercalcemia [21, 22], it has potential to influence osteoblastic differentiation and activity. The effects on bone may be multifaceted, as CaSR modulation can also affect other bone cells such as osteoclasts [23, 24]. Thus, there are mechanistic reasons to examine the effects of CN in monoculture systems, including osteoblasts and osteoclasts [25]. CN has been reported to reduce serum intact PTH levels in hemodialysis patients with secondary hyperparathyroidism and increase BMD [26]. It has recently been used in a trial in combination with denosumab for primary hyperparathyroidism [27]. Due to the complexity of cellular interactions in the bone microenvironment, there are clear advantages to assaying the effects of bone drugs in monoculture systems. Primary human osteoblast cultures represent the gold-standard for examining the effects of pharmacotherapy on bone cells. It has been previously reported that culture under hyperglycemic conditions (high glucose) can negatively affect primary osteoblasts [28]. It was speculated that treatment with CN+PTH could lead to improved outcomes for osteoblasts grown under high glucose conditions. As such, the aim of this study was to assess the impact of CN, PTH, CN+PTH versus vehicle controls on primary human osteoblasts under physiological and high glucose conditions and examine the impact on cell growth and osteogenic differentiation. CN has been previously shown to directly reduce parathyroid hormone stimulation in vitro, even in cells with pathologically reduced expression of the CaSR [29]. For this reason, we chose 1uM CN. A concentration of 10 nM of PTH was chosen for the current study as this is a common concentration used in in vitro studies. For example, Choudhary et al. has reported that 10 nM of PTH increased mineralization in vitro using marrow stromal cell (MSC) and calvarial osteoblast (COB) cultures from COX-2 knockout (KO) and wild type (WT) mice, as well as inhibiting SOST mRNA expression [30]. 10 nM PTH was also the lowest concentration of PTH found to inhibit the transcription of FGF23 in chicken bone marrow mesenchymal stem cells in vitro by Lyu et al.[31]. ## Primary Cell Culture Primary human osteoblasts (HOBs) were previously isolated from the minced trabecular ends of long bones of 18–20-week-old fetuses using a well-established model [28, 32–36], according to the National Health and Medical Research Council (NHMRC) guidelines and the project had the approval of the University of Sydney (USYD) human ethics committee (HEC) (approval number $\frac{01}{02}$/40). These primary human osteoblasts are very well characterized with a mesenchymal appearance in monolayer culture as shown in Fig. 1A. They have high alkaline phosphatase activity, produce both osteoprotegerin (OPG) and receptor-activator of NFκB-Ligand (RANKL), and express the calcium-sensing receptor and Homer proteins [34]. When cultured on an appropriate substrate for prolonged periods, these cells form multilayers which secrete the osteocyte marker sclerostin [35]. All cultures used here are between passage 2 and 5. The effects of high glucose concentrations and osmolar controls (mannitol) on the HOBs used here have also been previously assessed [28]. Cultured primary cells were maintained in $10\%$ fetal bovine serum (FBS) and low glucose DMEM (5 mM glucose with 3.5 g/L NaCHO3, 0.57 g/L Na2HPO4, and 0.006 g/L NaH2PO4, pH 7.4) (Invitrogen, Waltham, MA, USA).Fig. 1A Phase contrast microscopy of primary human osteoblasts (HOBs) in monolayer culture at 10X. B Confocal microscopy images of a monolayer culture of primary human osteoblasts taken with × 10 objective (scale bar = 10 μm) stained for CaSR (green) or Homer (red). Nuclei (blue) were stained with DAPI Cells were treated in multi-well plates with 1 µL/mL DMSO vehicle, 1 µM CN (MedChemExpress, Monmouth Junction, NJ, USA), 10 nM PTH 1–34 (Sigma-Aldrich, St Louise, MO, USA), or 1 µM CN+10 nM PTH 1–34 in either ‘physiological glucose’ (5 mM) or ‘high glucose’ (25 mM) DMEM. The treatment period was 7 days, with media changes being performed on alternate days. Studies were performed on triplicate samples (i.e., from three separate donors) and repeated $$n = 3$$ or $$n = 4$$ times. See Fig. 1A. ## Immunofluorescence (Human Osteoblasts) Primary human osteoblasts (HOBs) were grown on poly-l-lysine-coated coverslips for 7 days and were then fixed with $4\%$ (w/v) paraformaldehyde followed by $100\%$ ice-cold methanol and were processed with the following antibodies: anti-CaSR (Sigma, mouse monoclonal, clone HL1499), anti-Homer1 (Santa Cruz Biotechnology, rabbit polyclonal, clone H-174), and isotype controls. Coverslips were then washed and incubated with anti-rabbit Alexa Fluor 488 (1:750; Santa Cruz Biotechnology) and anti-mouse Cy3 (1:750; Life Technologies, Inc.) at room temperature for 60 min. To visualize the nuclei, coverslips were mounted with UltraCruz™ mounting medium containing DAPI. Slides were observed under a LSM510 Meta confocal laser microscope (Zeiss) at the Advanced Microscopy Facility (Bosch Research Institute, University of Sydney). See Fig. 1B. ## Live-Cell Imaging The effects of the treatments on the rate of osteoblast proliferation were examined using the IncuCyte® Live-Cell Analysis System (Essen Bioscience, Ann Arbor, MI, USA). HOBs were seeded at a density of 0.03 × 106 cells/well in 24-well plates before treatments were added. The cells were then incubated at 37° C and $4\%$ CO2 in the IncuCyte Analysis instrument. Throughout the 7-day treatment period, phase scans were automatically taken at 2-h intervals using a 10 × objective. Phase object confluency was then automatically calculated and recorded by the machine and used to plot a time course for confluency. To normalize for possible differences in the initial confluence percentage between the different samples, the time required to double the initial confluence value was used for subsequent statistical analysis. ## Cell Viability At treatment day 7, HOBs were detached, incubated with trypan blue (Sigma-Aldrich), centrifuged, resuspended, and counted using a Neubauer Hemocytometer (Livingstone). ## Caspase-3 Assay At treatment day 7, a Caspase-3 assay was performed as previously described [35]. In brief, HOBs were seeded at 1 × 103 in 96-well plates in physiological glucose media and allowed to attach. After 24 h, media were changed to either physiological or high glucose containing treatments as indicated. After 7 days, oxidative stress was then induced by the addition of 50 µM hydrogen peroxide and caspase activity assessed by a previously reported protocol using the caspase-3 substrate, Ac-DEVD-AFC [37]. Caspase-3 activity was used as a measure for apoptosis [38]. Cell numbers were controlled for by BCA assay (Thermo Scientific). ## ALP Activity At treatment day 7, HOBs in 96-well plates were assayed for alkaline phosphatase activity using a colorimetric assay as previously published [32]. Briefly, cells were lysed in 1 vol of PBS pH 7.2 containing $0.1\%$ v/v TX-100. To the lysate, 1 vol of 1 mg/mL p-Nitrophenyl Phosphate (Sigma) in 0.2 M glycine pH 9 was added and the plate incubated at RT for 15 min. A standard curve was reacted simultaneously on the same plate using calf intestinal ALP (New England Biolabs) under the same conditions. The formation of the yellow reaction product was measured at 405 nm using a Clariostar spectrophotometer (BMG Labtech). Values were corrected for total cell protein as measured by BCA assay. ## Alizarin Red S Staining At treatment day 7, HOBs were washed, fixed with $4\%$ paraformaldehyde, then stained using $2\%$ Alizarin Red made up freshly in distilled water, pH 4.2. Cell monolayers were washed with $10\%$ acetic acid in distilled water for 30 min at room temperature. To quantify the stain, cells were transferred into tubes, heated at 95° C, immediately placed on ice for 5 min, and centrifuged for 15 min at 20,000 g at 4° C. The absorbance of each sample was quantified using a standard curve and corrected for total cellular protein as determined by BCA assay (Thermo Fisher Scientific). ## Western Blotting HOBs lysates from 6-well plates were separated by SDS-PAGE (Bio-Rad Laboratories, Hercules CA, USA) and were transferred to methanol-activated PVDF membranes (Thermo Fisher Scientific, Waltham, MA USA) by wet tank transfer. Membranes were blocked with BSA (Thermo Fisher Scientific) then incubated with primary antibody targeting RUNX2 (rabbit polyclonal, 1:200), OPG (mouse monoclonal, 1:200), RANKL (goat polyclonal, 1:200), OCN (mouse monoclonal, 1:200), or CaSR (mouse monoclonal, 1:2000) (all Santa Cruz Biotechnology, Dallas, TX, USA). A primary antibody targeting the housekeeping protein, β-Tubulin (mouse monoclonal, 1:200, Santa Cruz Biotech), was also used to provide means for normalization. The membranes were then incubated with the appropriate HRP-conjugated secondary antibody (Santa Cruz Biotech). Bands were detected using a ChemiDoc Imaging System (Bio-Rad) following exposure of the membrane to a chemiluminescent HRP substrate (Merck). ## Statistical Analysis Graphing and statistical analysis were performed using GraphPad Prism software version 7.02 (GraphPad Software, California USA). A non-parametric, two-way, 2 × 4 analysis of variance (ANOVA) and post hoc Tukey’s Multiple Comparison tests were conducted compare all of the treatment groups. Graphical representations of Western blot data show the mean fold change from vehicle treatment in physiological glucose conditions ± standard error of the mean (SEM). ## CN+PTH Promotes the Growth of Cultured Human Osteoblasts The effect of physiological versus high glucose was compared using the IncuCyte to determine longitudinal cell division and calculate relative doubling times. High doubling times indicated slower growth. Under high glucose conditions, this time was significantly higher for cultured human osteoblasts (Fig. 2A). While treatment with CN or PTH did not significantly rescue this slower cell growth, the combination of CN+PTH reduced the doubling times of osteoblasts grown in high glucose to that of physiological glucose (Fig. 2A). Additionally, cell numbers were counted at the conclusion of treatment at day 7 (Fig. 2B). Similar results were found, with CN+PTH increasing cell numbers (1.9-fold in physiological glucose, twofold in high glucose). CN and PTH led to increases in cell number not reflected by the IncuCyte data; however, the former assay examines growth/proliferation rates, whereas cell counts represent the product of cell proliferation at a single time point. Fig. 2Osteoblastic proliferation and cell number. The phase object confluency (%) of primary human osteoblasts cultured in physiological (5 mM) or high (25 mM) glucose conditions and treated with a vehicle (1 µL DMSO/mL), cinacalcet (CN, 1 µM), parathyroid hormone (PTH, 10 nM), or CN+PTH for 7 days. A The time required for primary human osteoblasts to double their initial confluency. B Osteoblast number after 7 days of treatments. Data are presented as mean fold change from vehicle ± SEM. * $p \leq 0.05$, $p \leq 0.01$, and $p \leq 0.001$ ## The Effects of CN+PTH on Osteoblast Survival and Differentiation Biochemical assays were performed for caspase-3 and alkaline phosphatase activity. A significant 1.3-fold increase in caspase-3 activity ($$p \leq 0.04$$) was observed under high glucose conditions (Fig. 3A). This was significantly reduced by CN+PTH treatment, both compared to the elevated high glucose values as well as the basal physiological glucose levels ($p \leq 0.001$).Fig. 3Caspase-3 and alkaline phosphatase (ALP) activity assays. Osteoblasts were treated with vehicle, CN, PTH, or CN+PTH as previously described under physiological or high glucose conditions. A Caspase-3 activity and B ALP activity in cells after 7 days of treatment. Data are presented as mean fold change from vehicle ± SEM. Note that vehicle is 5 mM (physiological) glucose. $p \leq 0.05$ and **$p \leq 0.001$ Similarly, ALP activity was significantly impaired when cells were grown under high glucose conditions ($$p \leq 0.02$$, Fig. 3B). CN+PTH increased ALP activity in both physiological (1.5-fold, $$p \leq 0.001$$) and high glucose (twofold, $$p \leq 0.001$$) environments. ## CN+PTH Treatment Rescues the Poor Mineralization Seen in High Glucose Media Cultured human osteoblasts will form a monolayer, differentiate, and express osteoblastic markers and create a mineralized matrix. This mineralized matrix is a chief feature of osteoblast maturation and thus a key functional outcome measure. In the absence of treatment, primary osteoblasts grown in standard physiological glucose media formed mineralized nodules; however, this was impaired under high glucose conditions (Fig. 4A). Staining was enhanced under both physiological glucose by CN and PTH; however, the greatest enhancement was seen with CN+PTH. This latter treatment also rescued the lack of mineralization seen in high glucose media. Quantification of staining using a dye elution method illustrated that these differences were statistically significant. For CN+PTH treatment, the increase was 1.4-fold in physiological glucose and 3.6-fold in high glucose ($$p \leq 0.04$$, Fig. 4B).Fig. 4Matrix mineralization in CN, PTH, and CN+PTH-treated human osteoblasts. Osteoblasts were treated with vehicle, CN, PTH, or CN+PTH as previously described under physiological or high glucose conditions for 7 days. A Alizarin Red S staining at day 7 of treatment B Elution of stain showing quantitative Ca.2+ deposition levels. Data are presented as mean fold change from vehicle ± SEM. $p \leq 0.05$ and **$p \leq 0.001$ ## CN+PTH Upregulates RUNX2 and OCN Expression in Human Osteoblasts Runt-related transcription factor 2 (RUNX2) and Osteocalcin (OCN) are key markers of early- and late-stage osteogenic differentiation. Expression of these proteins was measured using Western blotting under different glucose conditions and drug treatments. CN and PTH, used together or separately at physiological glucose levels, produced an increase in RUNX2 (Fig. 5A). CN increased RUNX2 expression by 2.5-fold ($p \leq 0.01$) and PTH by 2.2-fold ($$p \leq 0.02$$) when compared to the vehicle. CN+PTH increased the expression levels of RUNX2 by 3.3-fold ($p \leq 0.001$). When exposed to high levels of glucose, RUNX2 was reduced; however, this effect was rescued by CN + PTH ($p \leq 0.001$). OCN expression was similarly enhanced by CN+PTH treatment (Fig. 5B).Fig. 5The expression levels of RUNX2 and OCN. Gene expression from immunoblots was quantified for A RUNX2 and B OCN protein expression under different glucose conditions and treatment with vehicle, CN, PTH, and CN+PTH. In panels, blots are shown above, with quantitation below as mean-fold change from physiological glucose vehicle ± SEM after normalization to the expression of the housekeeping gene, β-tubulin. $p \leq 0.05$, $p \leq 0.01$, $p \leq 0.001$ ## CN+PTH Improves the RANKL:OPG Ratio in Human Osteoblasts Under physiological glucose conditions, CN, PTH, and CN+PTH increased RANKL expression by 1.6- to 2.2-fold, with the greatest effect with CN+PTH (Fig. 6A, B). Under high glucose conditions, RANKL expression was 0.4-fold decreased in vehicle-treated samples, but treatment with CN, PTH, and CN+PTH again increased RANKL levels by 1.9- to 2.5-fold. While OPG levels were halved under high glucose conditions, treatment with CN, PTH, and CN+PTH did not significantly change OPG (Fig. 6A, B).Fig. 6The expression levels of RANKL and OPG. Gene expression from immunoblots was quantified for A RANKL and B OPG protein expression under different glucose conditions and treatment with vehicle, CN, PTH, and CN+PTH. In panels, blots are shown above, with quantitation below as mean-fold change from vehicle ± SEM after normalization to the expression of the housekeeping gene, β-tubulin. $p \leq 0.05$, $p \leq 0.01$, and $p \leq 0.001$ Calculating RANKL:OPG ratios relative to vehicle in physiological glucose showed a 1.9-fold increase with CN, 1.4-fold increase with PTH, and 2.1-fold increase with CN+PTH (physiological glucose), and a 1.1-fold increase with vehicle, 1.8-fold increase with CN, 1.7-fold increase with PTH, and a 1.8-fold increase with CN+PTH (high glucose). ## High Glucose Levels Downregulate the CaSR, but this is Rescued Using CN+PTH In high glucose media, CN and PTH were found to significantly elevate CaSR expression by 1.5-fold and 1.7-fold, respectively (Fig. 7). CN+PTH showed a greater effect on CaSR expression that was significantly elevated vs no treatment vehicle controls under physiological (1.7-fold increase) and high glucose (3.6-fold increase) conditions (Fig. 7).Fig. 7The expression levels of CaSR. Gene expression from immunoblots was quantified for CaSR under different glucose conditions and treatment with vehicle, CN, PTH, and CN+PTH. In panels, blots are shown above, with quantitation below as mean-fold change from vehicle ± SEM after normalization to the expression of the housekeeping gene, β-tubulin. $p \leq 0.001$ ## Discussion This study provides a comprehensive analysis of the impact of intervention with PTH, CN, and PTH+CN on primary osteoblasts cultured under physiological and high glucose conditions. We have previously shown that even 2 h of varying glucose concentrations affects these HOBs with higher concentrations (up to 20 mM) reducing ALP activity, cell viability, and osteocalcin RNA expression, and increasing apoptosis (via the same Caspase assay we have used in the current study) [28]. The treatment period chosen for the current study was 7 days to allow the high glucose environment to elicit an impact that is more likely to mimic that of chronic hyperglycemia in T2DM within the time constraints of an in vitro cell culture model. Our findings support the concept of hyperglycemia negatively impacting on osteoblasts. Drug treatments promoted a range of benefits, including increased proliferation, differentiation, and function (mineralization). In many cases, the improvements produced by PTH+CN were greater under the problematic high glucose conditions. The pathways affected by pharmaceutical intervention are shown schematically in Fig. 8. One of the most interesting findings to come from the current study was the upregulation of the CaSR in HOBs exposed to either PTH alone, CN alone, and even more markedly with the PTH+CN combination treatment. The body requires that circulating Ca2+ concentrations to be kept within a very narrow range, requiring that any cells that need to detect changes in this near-consistent parameter have a remarkable sensitivity to fluctuations in extracellular Ca2+ concentrations. It is thus very likely that calciotropic hormones, such as PTH, and pharmaceutical agents that have been developed to work with Ca2+ homeostasis, indeed upregulate the expression of the very receptor that detects these changes—the CaSR. Another calcimimetic, NPS R-568 has been shown to upregulate both mRNA and protein levels of the CaSR in the parathyroid gland in rats [39]. This co-operation underpins the sophistication of the Ca2+ homeostatic system. Fig. 8The effects of cinacalcet (CN) and parathyroid hormone (PTH) on osteoblasts. CN acts on and increases the sensitivity of the Ca2+ sensing receptor (CaSR) to Ca2+. Ca2+ activates the CaSR and PTH activates the PTH receptor (PTH-R). The activation of the receptors leads to several transcriptional effects. Firstly, receptor activation upregulates runt-related transcription factor 2 (RUNX2). This increases the expression of osteocalcin (OCN) and enhances alkaline phosphatase (ALP) activity, which are hallmarks of osteoblastic activity. The increase in activity then results in enhanced mineralization and hence bone formation. RUNX2 also stimulates differentiation and maturation and inhibits apoptosis by reducing caspase-3 activity. This stimulates osteoblastic proliferation, leading to more cells. Secondly, receptor activation upregulates receptor activator of nuclear factor κ-B ligand (RANKL). This increases the ratio of RANKL to its inhibitor, osteoprotegerin (OPG), which can then stimulate osteoclastic activity and bone resorption. Finally, activating the CaSR and PTH-R also leads to the upregulation of the CaSR, creating a positive feedback loop PTH+CN may also indirectly improve bone turnover via improving the RANKL:OPG ratio, as bone turnover can be impaired in T2DM [28]. While the impacts of CN and PTH were not redundant, for most outcomes, the combination produced an additive rather than overtly synergistic effect. The mechanisms by which high glucose leads to poor bone health is not well understood [40]. Published studies have used simulated models of diabetes (in vitro and in vivo) to examine how hyperglycemia affects the behavior of bone cells [41, 42]. The effects of high glucose on osteoblasts do not appear to be related to osmolarity, as prior studies found no impact with a comparable concentration of D-mannitol [28]. Most prior studies have employed immortalized osteoblasts or primary rodent cells, rather than primary human osteoblasts. While cell lines are a convenient and accessible research tool, they should not be considered appropriate replacements for primary osteoblasts [43]. Our findings of the effects of high glucose are consistent with those of Garcia-Hernandez et al., who treated primary human osteoblasts with up to 24 mM glucose and reported impaired matrix mineralization, osteogenic protein markers, and RANKL:OPG ratio [44]. Future directions for this research will focus on mechanism and clinical translation. The effects of high glucose on osteoblasts are potentially mediated by elevated cytokine expression [44]. However, the correlation and relevance of in vitro cytokine expression and in vivo physiology are unclear. Other factors may also play a role such as the glycoprotein Fetuin-A, which can be upregulated by cinacalcet [45], and are associated with impaired insulin sensitivity and glucose tolerance. While there is utility in candidate approaches, unbiased approaches such as transcriptome sequencing may uncover novel mechanisms and druggable targets. Our data support further investigation into the clinical use of PTH/teriparatide and cinacalcet in patients with diabetes and osteoporosis. Further data supporting prospective studies could be potentially gained from subgroup analysis of prior trials. For example, reduced fractures were seen with cinacalcet treatment in end-stage renal disease patients [46]; presumably, a proportion of these patients were diabetic as it is one of the most common causes of renal impairment. ## References 1. Lin X. Global, regional, and national burden and trend of diabetes in 195 countries and territories: an analysis from 1990 to 2025*. Sci Rep* (2020) 10 14790. DOI: 10.1038/s41598-020-71908-9 2. Fruhbeck G. Obesity: the gateway to ill health - an EASO position statement on a rising public health, clinical and scientific challenge in Europe. Obes Facts (2013) 6 117-120. DOI: 10.1159/000350627 3. Warming L, Hassager C, Christiansen C. Changes in bone mineral density with age in men and women: a longitudinal study. Osteoporos Int (2002) 13 105-112. DOI: 10.1007/s001980200001 4. Reyes-Garcia R. Serum levels of bone resorption markers are decreased in patients with type 2 diabetes. Acta Diabetol (2013) 50 47-52. DOI: 10.1007/s00592-011-0347-0 5. Lee BK. Comparison of age of onset and frequency of diabetic complications in the very elderly patients with type 2 diabetes. Endocrinol Metab (Seoul) (2016) 31 416-423. DOI: 10.3803/EnM.2016.31.3.416 6. Purnamasari D. Low bone turnover in premenopausal women with type 2 diabetes mellitus as an early process of diabetes-associated bone alterations: a cross-sectional study. BMC Endocr Disord (2017) 17 72. DOI: 10.1186/s12902-017-0224-0 7. Dede AD. Type 2 diabetes mellitus and fracture risk. Metabolism (2014) 63 1480-1490. DOI: 10.1016/j.metabol.2014.09.002 8. Tuominen JT. Bone mineral density in patients with type 1 and type 2 diabetes. Diabetes Care (1999) 22 1196-1200. DOI: 10.2337/diacare.22.7.1196 9. Schwartz AV. Association of BMD and FRAX score with risk of fracture in older adults with type 2 diabetes. JAMA (2011) 305 2184-2192. DOI: 10.1001/jama.2011.715 10. Giangregorio LM. FRAX underestimates fracture risk in patients with diabetes. J Bone Miner Res (2012) 27 301-308. DOI: 10.1002/jbmr.556 11. Shahen VA. Multifactorial effects of hyperglycaemia, hyperinsulinemia and inflammation on bone remodelling in type 2 diabetes mellitus. Cytokine Growth Factor Rev (2020) 55 109-118. DOI: 10.1016/j.cytogfr.2020.04.001 12. Malik P. Role of receptor for advanced glycation end products in the complication and progression of various types of cancers. Biochim Biophys Acta (2015) 1850 1898-1904. DOI: 10.1016/j.bbagen.2015.05.020 13. McCarthy AD. Advanced glycation endproducts interefere with integrin-mediated osteoblastic attachment to a type-I collagen matrix. Int J Biochem Cell Biol (2004) 36 840-848. DOI: 10.1016/j.biocel.2003.09.006 14. Snedeker JG, Gautieri A. The role of collagen crosslinks in ageing and diabetes - the good, the bad, and the ugly. Muscles Ligaments Tendons J (2014) 4 303-308. DOI: 10.32098/mltj.03.2014.07 15. Bai XC. Oxidative stress inhibits osteoblastic differentiation of bone cells by ERK and NF-kappaB. Biochem Biophys Res Commun (2004) 314 197-207. DOI: 10.1016/j.bbrc.2003.12.073 16. Lam J. TNF-alpha induces osteoclastogenesis by direct stimulation of macrophages exposed to permissive levels of RANK ligand. J Clin Invest (2000) 106 1481-1488. DOI: 10.1172/JCI11176 17. Chiodini I. Management of bone fragility in type 2 diabetes: perspective from an interdisciplinary expert panel. Nutr Metab Cardiovasc Dis (2021) 31 2210-2233. DOI: 10.1016/j.numecd.2021.04.014 18. Hodsman AB. Efficacy and safety of human parathyroid hormone-(1–84) in increasing bone mineral density in postmenopausal osteoporosis. J Clin Endocrinol Metab (2003) 88 5212-5220. DOI: 10.1210/jc.2003-030768 19. Aslan D. Mechanisms for the bone anabolic effect of parathyroid hormone treatment in humans. Scand J Clin Lab Invest (2012) 72 14-22. DOI: 10.3109/00365513.2011.624631 20. Hebert SC. Therapeutic use of calcimimetics. Annu Rev Med (2006) 57 349-364. DOI: 10.1146/annurev.med.57.121304.131328 21. Block GA. Cinacalcet for secondary hyperparathyroidism in patients receiving hemodialysis. N Engl J Med (2004) 350 1516-1525. DOI: 10.1056/NEJMoa031633 22. Torres PU. Cinacalcet HCl: a novel treatment for secondary hyperparathyroidism caused by chronic kidney disease. J Ren Nutr (2006) 16 253-258. DOI: 10.1053/j.jrn.2006.04.010 23. Dvorak MM. Constitutive activity of the osteoblast Ca2+-sensing receptor promotes loss of cancellous bone. Endocrinology (2007) 148 3156-3163. DOI: 10.1210/en.2007-0147 24. Dvorak MM. Physiological changes in extracellular calcium concentration directly control osteoblast function in the absence of calciotropic hormones. Proc Natl Acad Sci USA (2004) 101 5140-5145. DOI: 10.1073/pnas.0306141101 25. Shalhoub V. In vitro studies with the calcimimetic, cinacalcet HCl, on normal human adult osteoblastic and osteoclastic cells. Crit Rev Eukaryot Gene Expr (2003) 13 89-106. PMID: 14696958 26. Tsuruta Y. Effects of cinacalcet on bone mineral density and bone markers in hemodialysis patients with secondary hyperparathyroidism. Clin Exp Nephrol (2013) 17 120-126. DOI: 10.1007/s10157-012-0665-8 27. Leere JS. Denosumab and cinacalcet for primary hyperparathyroidism (DENOCINA): a randomised, double-blind, placebo-controlled, phase 3 trial. Lancet Diabetes Endocrinol (2020) 8 407-417. DOI: 10.1016/S2213-8587(20)30063-2 28. Levinger I. Glucose-loading reduces bone remodeling in women and osteoblast function in vitro. Physiol Rep (2016) 4 e12700. DOI: 10.14814/phy2.12700 29. Kawata T. Direct in vitro evidence of the suppressive effect of cinacalcet HCl on parathyroid hormone secretion in human parathyroid cells with pathologically reduced calcium-sensing receptor levels. J Bone Miner Metab (2006) 24 300-306. DOI: 10.1007/s00774-006-0687-y 30. Choudhary S. Anabolic effects of PTH in cyclooxygenase-2 knockout osteoblasts in vitro. Biochem Biophys Res Commun (2008) 372 536-541. DOI: 10.1016/j.bbrc.2008.05.050 31. Lyu Z. Fibroblast growth factor 23 inhibits osteogenic differentiation and mineralization of chicken bone marrow mesenchymal stem cells. Poult Sci (2022) 102 102287. DOI: 10.1016/j.psj.2022.102287 32. Brennan TC, Rybchyn MS, Green W, Atwa S, Conigrave AD, Mason RS. Osteoblasts play key roles in the mechanisms of action of strontium ranelate. Br J Pharmacol (2009) 157 1291-1300. DOI: 10.1111/j.1476-5381.2009.00305.x 33. Rybchyn MS, Green WL, Conigrave AD, Mason RS. Involvement of both GPRC6A and the calcium-sensing receptor in strontium ranelate-induced osteoclastogenic signal expression and replication in primary human osteoblasts. Bone (2009) 44 S317. DOI: 10.1016/j.bone.2009.03.595 34. Rybchyn MS. Homer1 mediates CaSR-dependent activation of mTOR complex 2 and initiates a novel pathway for AKT-dependent beta-catenin stabilization in osteoblasts. J Biol Chem (2019) 294 16337-16350. DOI: 10.1074/jbc.RA118.006587 35. Rybchyn MS. An Akt-dependent Increase in canonical Wnt signaling and a decrease in sclerostin protein levels are involved in strontium ranelate-induced osteogenic effects in human osteoblasts. J Biol Chem (2011) 286 23771-23779. DOI: 10.1074/jbc.M111.251116 36. Slater M. Modulation of growth factor incorporation into ECM of human osteoblast-like cells in vitro by 17 beta-estradiol. Am J Physiol (1994) 267 E990-1001. PMID: 7810645 37. Grebenová D. Mitochondrial and endoplasmic reticulum stress-induced apoptotic pathways are activated by 5-aminolevulinic acid-based photodynamic therapy in HL60 leukemia cells. J Photochem Photobiol B (2003) 69 71-85. DOI: 10.1016/S1011-1344(02)00410-4 38. McIlwain DR, Berger T, Mak TW. Caspase functions in cell death and disease. Cold Spring Harb Perspect Biol (2015). DOI: 10.1101/cshperspect.a026716 39. Mizobuchi M. Calcimimetic compound upregulates decreased calcium-sensing receptor expression level in parathyroid glands of rats with chronic renal insufficiency. J Am Soc Nephrol (2004) 15 2579-2587. DOI: 10.1097/01.ASN.0000141016.20133.33 40. Rizzoli R, Biver E, Brennan-Speranza TC. Nutritional intake and bone health. Lancet Diabetes Endocrinol (2021) 9 606-621. DOI: 10.1016/S2213-8587(21)00119-4 41. Li Y. Impact of diabetes mellitus simulations on bone cell behavior through in vitro models. J Bone Miner Metab (2020) 38 607-619. DOI: 10.1007/s00774-020-01101-5 42. Aswamenakul K. Proteomic study of in vitro osteogenic differentiation of mesenchymal stem cells in high glucose condition. Mol Biol Rep (2020) 47 7505-7516. DOI: 10.1007/s11033-020-05811-x 43. Czekanska EM. A phenotypic comparison of osteoblast cell lines versus human primary osteoblasts for biomaterials testing. J Biomed Mater Res A (2014) 102 2636-2643. DOI: 10.1002/jbm.a.34937 44. Garcia-Hernandez A. High glucose concentrations alter the biomineralization process in human osteoblastic cells. Bone (2012) 50 276-288. DOI: 10.1016/j.bone.2011.10.032 45. Messa P. Calcimimetic increases osteoprotegerin and decreases fetuin-A levels in dialysis patients. Nephrol Dial Transplant (2007) 22 2724-2725. DOI: 10.1093/ndt/gfm312 46. Investigators ET. Effect of cinacalcet on cardiovascular disease in patients undergoing dialysis. N Engl J Med (2012) 367 2482-2494. DOI: 10.1056/NEJMoa1205624
--- title: Guidelines on lung adenocarcinoma prognosis based on immuno-glycolysis-related genes authors: - Yuting Zhang - Wen Qin - Wenhui Zhang - Yi Qin - You Lang Zhou journal: Clinical & Translational Oncology year: 2022 pmcid: PMC10025218 doi: 10.1007/s12094-022-03000-9 license: CC BY 4.0 --- # Guidelines on lung adenocarcinoma prognosis based on immuno-glycolysis-related genes ## Abstract ### Objectives This study developed a new model for risk assessment of immuno-glycolysis-related genes for lung adenocarcinoma (LUAD) patients to predict prognosis and immunotherapy efficacy. ### Methods LUAD samples and data obtained from the Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) databases are used as training and test columns, respectively. Twenty-two [22] immuno-glycolysis-related genes were screened, the patients diagnosed with LUAD were divided into two molecular subtypes by consensus clustering of these genes. The initial prognosis model was developed using the multiple regression analysis method and Receiver Operating characteristic (ROC) analysis was used to verify its predictive potential. Gene set enrichment analysis (GSEA) showed the immune activities and pathways in different risk populations, we calculated immune checkpoints, immune escape, immune phenomena (IPS), and tumor mutation burden (TMB) based on TCGA datasets. Finally, the relationship between the model and drug sensitivity was analyzed. ### Results Fifteen [15] key differentially expressed genes (DEGs) with prognostic value were screened and a new prognostic model was constructed. Four hundred and forty-three [443] samples were grouped into two different risk cohorts based on median model risk values. It was observed that survival rates in high-risk groups were significantly low. ROC curves were used to evaluate the model’s accuracy in determining the survival time and clinical outcome of LUAD patients. Cox analysis of various clinical factors proved that the risk score has great potential as an independent prognostic factor. The results of immunological analysis can reveal the immune infiltration and the activity of related functions in different pathways in the two risk groups, and immunotherapy was more effective in low-risk patients. Most chemotherapeutic agents are more sensitive to low-risk patients, making them more likely to benefit. ### Conclusion A novel prognostic model for LUAD patients was established based on IGRG, which could more accurately predict the prognosis and an effective immunotherapy approach for patients. ### Supplementary Information The online version contains supplementary material available at 10.1007/s12094-022-03000-9. ## Introduction At present, Lung cancer has become the second most common cancer worldwide, and its incidence and deaths are increasing over the years with a high level of invasiveness. In 2020, Lung cancer accounted for $11.4\%$ of all new cancer cases, $18.0\%$ (1.8 million) of deaths from malignant tumors worldwide [1]. Non-small cell lung cancer (NSCLC) accounts for about 80–$85\%$ of primary lung cancers [2], with $70\%$ of patients having progressed to intermediate or advanced/late stages by the time of diagnosis [3]. Lung adenocarcinoma (LUAD) accounts for approximately $40\%$ of all newly diagnosed lung cancer cases and is the most common histologic form of NSCLC [4, 5]. Despite significant advances in non-invasive surgery and immunotherapy in recent decades, 5-year overall survival (OS) rate is still as low as $17.4\%$ [6]. An accurate prognosis of LUAD patients is a prerequisite for more effective treatment, improved survival rates and reduced mortality, which remains a great clinical challenge to the current medical healthcare system. Cellular metabolism involves the synthesis, maintenance or decomposition of biomolecules, which not only provide material and energy for cell activities but also function as signaling transduction agents or transducers. Abnormal metabolic reprogramming promotes cell growth and division, leading to uncontrolled and sustained malignant proliferation [7]. When metabolic reprogramming occurs, it can cause various diseases, such as glycolysis disorders, which can lead to diabetes and various cancers [8, 9]. Numerous specific metabolic reprogramming occur during precancerous lesions, for example, the oncogenic KRas gene causes metabolic reprogramming, which increases mitochondrial reactive oxygen species (mROS) and promotes acinar ductal metaplasia (ADM), which contributes to the development of pancreatic cancer [10]. Therefore, metabolic reprogramming can be used to predict the occurrence of cancer under certain conditions, and focusing on metabolic markers in tumor metabolic reprogramming has functional and meaningful implications for targeted therapy. Abnormal glucose metabolism is an important part of tumor metabolic reprogramming. Tumor cells alter metabolic fluxes to maintain their normal survival and progression in the microenvironment. Tumor cells do not metabolize energy through oxidative phosphorylation (OXPHOS), which is significantly different from normal cells. One of the most common and important changes to metabolism is aerobic glycolysis, widely referred to as the “Warburg effect”. Warburg effect represents a shift in glucose utilization by tumor cells from oxidative phosphorylation to glycolysis, characterized by increased glucose uptake and lactate secretion, occurs even under normal oxygen content, and is now considered as one of the hallmarks of tumors [11–13]. Many glycolysis-related enzymes, such as hexokinase 2, phosphofructokinase, and pyruvate kinase, are overexpressed in lung cancer cells compared to normal cells [14–16]. The reprogramming of metabolic pathways facilitates the malignant proliferation of tumor cells and the ability to adapt to the harsh living environment, providing energy and conditions for the proliferation and invasion of cancer cells in LUAD [17]. Tumor microenvironment (TME) is an integrated cellular environment surrounding a tumor cell in which various innate immune cells. When the tumor microenvironment is associated with the function and signal transduction of these immune cells, it can also be refered to as the tumor immune microenvironment (TIME). Dysregulation of the tumor immune microenvironment and alterations of metabolic pathways are two unique markers of tumor cells [18]. The tumor microenvironment, especially the immune microenvironment, is a key factor in evaluating the clinical survival of cancer patients and can effectively reflect the ability of immune response [19, 20]. In TIME, tumorigenesis and evolution are important as crosstalk between immune cells and tumor cells generates an environment that promotes tumor proliferation and metastasis. For example, PD-1 on the surface of T cells interacts with PD-L1 on the surface of tumor cells, inhibiting T cell immune function and protecting tumor cells from immune attack, resulting in an immune evasion [21]. Thus, the state of the immune microenvironment can determine tumor cell progression and anti-tumor immune response. In this study, a new prognostic feature based on glycolysis-related genes (GRGs) and immune-related genes (IRGs) was developed and characterized by multiple statistical methods showing their reliability. It improves the ability to accurately determine the prognosis of LUAD and provides assistance for the rescheduling of clinical management strategies. ## Data acquisition and collection Data collected for lung adenocarcinoma mRNA expression and clinical data were obtained from TCGA (https://www.portal.gdc.cancer.gov/)—LUAD dataset, and microarrays obtained at GEO (https://www.ncbi.nlm.nih.gov/geo/) were used for validation. The following analysis was performed using the expression profile of 594 LUAD samples (535 tumors and 59 normal). Clinical information of LUAD patients was downloaded from TCGA-LUAD dataset, including survival time and status, clinical grade, gender, age, TMN classification. The information from 488 LUAD patients was later used for model development and accuracy validation, excluding patients with 0 survival time and incomplete information. The expression matrix file (GSE68465) from the GEO database was then used for external validation. Three hundred and two [302] GRGs were obtained by accessing GSEA (http://www.gsea-msigdb.org/gsea/index.jsp), with 2483 IRGS available in ImmPort (https://www.immport.org/). These data sources are publicly accessible, so the study has no ethical or conflict of interest and does not require review approval from a local council. ## Acquisition of intersecting genes The obtained three hundred and two [302] glycolysis-related genes and 2483 immune-related genes were used to draw a Venn diagram using an online tool (http://www.bioinformatics.psb.ugent.be/webtools/Venn/). Twenty-two [22] overlapping genes were identified from the two sets of data, overlapping genes were identified as candidate genes for subsequent analysis. ## Screening for immune- and glycolysis-related DEGs Differentially expressed genes (DEGs) were calculated using the “Limma” R package (version 4.1.2) to identify which immuno-glycolytic related genes were differentially expressed in normal and tumor tissues. Genes with $P \leq 0.05$ were identified as DEGs by the Wilcoxon rank sum test. The “pheatmap” package in R language was used to visualize the DEGs and draw the heatmap. The protein–protein interactions (PPI) of DEGs were calculated using the online public database STRING database (version11.5, https://www.string-db.org), setting the confidence score to ≥ 0.4 and removing free nodes. The PPI network was drawn next to elucidate the protein–protein interactions. The correlation coefficients between DEGs were calculated after removing samples from normal tissues via the “igraph” package in R language, and a co-expression network graph was created, which finally showed the interrelationships between 11 DEGs. ## Consensus clustering The “ConsensusClusterPlus” tool in R was used to implement an unsupervised clustering method that divides the LUAD samples in the TCGA dataset into two groups based on 22 prognostic candidate genes. The ideal number of clusters between $k = 2$ and 9 was then evaluated and 1000 replications/repetitions were performed to determine the most reliable classification. Based on the “survival” and “survminer” packages in R language, survival differences between the different clusters were analyzed, and $P \leq 0.05$ was considered as the difference in patient survival between the two clusters. This was then visualized using the “ggsurvplot” package to plot Kaplan–Meier (K–M) survival curves, DEGs between different clusters were identified using the “ggplot2” package in R language. Twenty-two [22] candidate genes were simultaneously observed for DEGs between different clusters (FDR < 0.05, \|logFC\|> 1) and heat maps were created for 7 DEGs screened by pheatmap to visualize their differential expression and clinical traits between clusters. ## Risk model construction and validation For genotyping differential genes, univariate Cox regression analysis was performed using the “survival” package to screen out genes associated with OS ($P \leq 0.05$) and identified as prognosis-related genes. Then “glmnet” was applied to process the above-mentioned genes to identify key genes and build a prognostic model, thereby selecting the optimal number of genes and candidate genes by the obtained least absolute shrinkage and selection operator (LASSO) results. The formula of risk score obtained according to LASSO regression results was as follows: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm{risk} \mathrm{score}=\sum_{$i = 1$}^{n}\left({\mathrm{Coef}}_{\mathrm{i}}\times {\mathrm{Exp}}_{\mathrm{i}}\right)$$\end{document}riskscore=∑$i = 1$nCoefi×Expi Where n is the number of prognosis-related genes in the model, Coefi the related gene coefficient, and Expi represents gene expression. All patients included in the analysis were grouped into high or low risk according to the cut-off point of the best risk score. Kaplan–Meier(K–M) curves were plotted using the “survminer” R package to detect and demonstrate differences in survival rates. Receiver operating characteristic (ROC) curves were also plotted at 1, 3, and 5 years to validate the accuracy of the prognostic model developed. A dot plot was created with “pheatmap” in R to determine the association between risk score and survival status. Principal Component Analysis (PCA) and t-distributed Stochastic Neighbor Embedding (t-SNE) analysis were performed using “ggplot2” and “Rtsne” to explore whether the model could accurately distinguish between different risk groups and visualize the results. This was followed by univariate and multivariate analyses combining risk scores with clinical characteristics to explore the correlation between this index and patient OS, and those with significant correlation were identified as independent prognostic factors ($P \leq 0.05$). ## Pathway enrichment and immune function analysis of differential genes Gene ontology (GO) annotation and immune infiltration were evaluated according to DEGs (\|logFC\|≥ 1 and FDR < 0.05) in both risk groups using the “clusterProfiler” and “gsva” R software packages. Filtered with P value < 0.05 and q value < 0.05 as thresholds to identify significant enrichment pathways. The Single sample gene set enrichment analysis (ssGSEA) was assessed for potential immunological function and active pathways for relevant biological values. ## Immune response and tumor mutation burden analysis Potential immune checkpoints were extracted from previous literature reviews, and the “ggpubr” R package evaluated and compared the expression levels of 22 immune checkpoint genes in the high and low-risk groups. The correlations between immune-related genes and risk scores were then assessed using spearman correlation analysis, and correlation analysis was performed using the R package “limma”. The TME of both groups was also analyzed. The Cancer Immunome Atlas (TCIA, https://www.tcia.at) provided an Immunophenoscore (IPS) for LUAD patients. In combination with the Tumor Immune Dysfunction and Exclusion (TIDE) algorithm, the predictive efficiency of the model for immunotherapy response was analyzed in the high-risk and low-risk groups. Immune evasion and correlation analyses were performed using the “ggpubr” and “corpplot” software packages, respectively. The differences in methylation expression between the two risk groups were further compared and the immune score, stromal score, estimated score and tumor purity were analyzed for both groups. Based on the downloaded nucleotide variant data from LUAD, the mutational load of the samples was calculated using perl software (version 5.32.1), thus comparing the differences between the two risk groups and evaluating the correlation between mutational load and risk. Next, the mutational burden and the risk value were combined and evaluated for survival analysis using “survminer” R. ## Sensitivity analysis of chemotherapy Treatment response to known common chemotherapeutics was assessed using the “pRRophetic” package. The half-maximal inhibitory concentrations (IC50) were calculated from the TCGA-LUAD dataset to investigate the difference in sensitivity to commonly used chemotherapeutic drugs between the high and low expression groups, and thus estimate the relationship between the model and drug response. ## Statistical analysis All statistical and graphing work was done by R software (4.1.2). Perl was used for all data processing and collation of the data matrix. The K–M method and Log-rank test were used to analyze survival curves and differences. Univariate and multivariate Cox analyses were performed to determine whether the prognostic model could be used as an independent prognostic factors. Differences between the two groups were compared using the Wilcoxon rank sum test, Spearman’s correlation analysis method was used to assess the correlation and all heatmaps were generated by the pheatmap parameter in R software. Statistical tests were two-sided, and $P \leq 0.05$ was considered a criterion to distinguish differences. ## Analysis of genes related to immunity and glycolysis A Venn diagram was drawn for 302 GRGs and 2483 IRGs, 22 candidate genes (MET, GPI, SDC2, PPARG, PSMC4, PPIA, VEGFA, ANGPTL4, SOD1, SDC1, HSPA5, ISG20, TGFA, MIF, ECD, ARTN) were obtained as shown in Fig. 1A. Twenty-two [22] IGRGs were differentially analyzed between tumor and normal tissues, among them, 16 genes showed differential expression, with SDC2 and PPARG significantly down-regulated in tumor samples. MET, GPI, ANGPTL4, HSPA5, TGFA, MIF, and ARTN were significantly up-regulated in tumor samples (Fig. 1B). A protein–protein interaction (PPI) network analysis was established for these 16 differential genes using the STRING database to identify their interactions. Four free nodes (ARTN, IAG20, ECD and GPI) were removed to obtain the protein interaction between the remaining 12 genes (Fig. 1C),co-expression networks of 16 DEGs were subsequently constructed by weighted gene co-expression network analysis, and the results showed that co-expression relationships existed between 11 genes, and all were positively correlated as shown in Fig. 1D.Fig. 1Analysis of genes related to immunity and glucose metabolism and classification of clusters. A A Venn diagram shows the intersection of immune-glycolysis-related genes. B Heatmap of 16 differentially expressed genes in normal samples and LUAD. * $P \leq 0.05$; $P \leq 0.01$; $P \leq 0.001.$ C The network is made up of 12 interconnected, differentiated genes. D The co-expression network consisted of 11 positively correlated genes (red: positively correlated; Blue: negative correlation). E Consensus clustering matrix, CDF curves and the relative changes of different clusters under the CDF curve in TCGA cohort. F Kaplan–meier curves of OS of two groups of LUAD patients and the number of surviving patients in cluster1 and cluster2 at different time periods. G Heatmap and clinicopathological characteristics of differential genes in two clusters (cluster1 and cluster2) ($P \leq 0.05$) ## Consensus clustering was used to identify two molecular subtypes The clinical samples in the TCGA database were divided into different clusters ($k = 2$–9) based on the expression of 22 genes, and the consensus matrix, the consensus CDF curve, and the relative change in area under the curve (Fig. 1E) showed that $k = 2$ was the optimal partition. Four hundred and forty-three [443] clinical data of lung adenocarcinoma were rationally assigned to two different subtypes named Cluster1 (C1, $$n = 261$$) and Cluster2 (C2, $$n = 182$$). K–M survival analysis subsequently showed that Cluster1 had an inferior survival rate compared to Cluster2 ($P \leq 0.001$) (Fig. 1F). This also implied that the median clinical survival was higher in Cluster2 patients. Moreover, 7 DEGs (STC1, VEGFA, ANGPTL4, TGFA, MIF, STC2 and ARTN) were obtained by differential genetic analysis of 22 IRGs in the two genotypes. The clinicopathological characteristics of these seven differential genes between the two subtypes were then investigated. Heatmap results showed that all 7 genes were down-regulated in Cluster2, with different subtypes distributed differently in the N stage ($P \leq 0.05$) and the clinical stage ($P \leq 0.05$) (Fig. 1G). ## Prognostic model construction and validation Differential expression analysis was first performed on Cluster1 and Cluster2, and a total of 1567 DEGs were identified. These DEGs were used for batch correction and expression extraction in TCGA and GEO databases. The TCGA and GEO expression data were combined with the survival data after excluding normal samples. The TCGA cohort was used as the training group and GES68465 from the GEO database was used as the test group. Three hundred and ninety-three genes [393] genes were then selected as prognostic genes from the combined TCGA survival and expression data (Supplementary Table 1). To avoid the risk of over-fitting and subsequent bias, the Cox regression model of the lasso method was optimized for the above 393 genes, of which 15 genes were identified as optimal variables (Fig. 2A, B). Finally, these 15 genes were used to construct the prognosis model, and the following equation was obtained: Risk score = 0.0717 × FLNC + 0.0025 × FBN2 + 0.0054 × CCL20 + 0.1881 × NTSR1 + 0.0265 × KRT6A + 0.0869 × DKK1 + 0.0004 × KYNU + 0.0605 × TENM3 + 0.0011 × ANGPTL4 + (− 0.0952) × STAP1 + 0.0606 × HMMR + 0.0282 × IGFBP1 + (− 0.0013) × C11orf16 + 0.0623 × LDHA + 0.0209 × PLEK2. LUAD patients in the training group were divided into high ($$n = 221$$) low ($$n = 222$$) risk groups based on optimal cutoff values. Fig. 2Establishment and validation of IGRGs prognosis model based on the training set. A–B Regression coefficients and partial likelihood deviations of 393 prognostic DEGs. C–D K–M analysis of two risk subgroups of the training group and validation group and the number of patients in the two groups who survived in different time periods. E–F The model predicts AUC for patient survival. G–H Scatter plot of the score distribution, survival time and status of patients in the training set. I–J Risk score distribution, survival time, and status in the validation set K–M analysis showed a significant reduction in survival rate for all high-risk patients in the training and test groups. There was a significant difference in overall survival between the two risk cohorts in the training group ($P \leq 0.001$), while the low-risk cohort had significantly better survival outcome. This conclusion was also supported by the test group ($$P \leq 0.002$$) (Fig. 2C, D). The effectiveness of the risk score in predicting OS was assessed by ROC analysis, the area under the curve (AUC) in the training group were 0.796, 0.709, and 0.677, in the test group, they were 0.702, 0.650, and 0.603, at 1, 3 and 5 years, respectively. These two sets of results showed that the model was highly specific and had the good predictive ability (Fig. 2E, F). The accuracy was then validated, and a survival distribution was plotted for the training and test groups to explore the relationship between risk values and survival prognosis (Fig. 2G–J). These results showed that as risk increased, mortality increased and survival decreased. ## Analysis of independent prognostic factors in training and test groups Each patient's clinicopathological features were analyzed and validated. The model had reliable clustering ability in both groups as revealed by principal component analysis (PCA) and t-distributed stochastic neighbor embedding (t-SNE) (Fig. 3A–D). Univariate analysis revealed clinical parameters, such as T ($P \leq 0.001$) and N ($P \leq 0.001$), were closely related to the OS of the training group, the risk score ($P \leq 0.001$) was an independent prognostic factor for LUAD, with a hazard ratio (HR) of 5.675 (Fig. 3E). Multivariate Cox regression results obtained revealed that the risk score ($P \leq 0.001$, HR = 4.601) had the ability to independently predict patients’ OS (Fig. 3F). Also, similar results were obtained from the GSE68465 test data, where the risk score ($P \leq 0.001$, HR = 2.094) was also proven to be an independent predictor of poor outcome. ( Fig. 3G, H). The expression changes of 15 key prognostic genes in different parameters were then compared in the training group, the heatmap of clinical characteristics showed significant differences in gender ($P \leq 0.05$), grade ($P \leq 0.001$), N ($P \leq 0.001$), and T ($P \leq 0.001$) between the two groups. Thirteen [13] genes (FLNC, FBN2, CCL20, NTSR1, KRT6A, DKK1, KYNU, TENM3, ANGPTL4, HMMR, IGFBP1, LDHA and PLEK2) were high-risk genes, and two genes (STAP1 and C11orf16) were low-risk genes as shown in (Fig. 3I).Fig. 3Independent prognostic value of gene characteristics in training and test cohorts. A–D PCA and t-SNE analysis of the model. C–D PCA and T-SNE analyses based on prognostic genes in the validation cohort. E–F Univariate and multivariate analyses of clinical characteristics associated with survival in the training cohort. G–H Univariate and multivariate analysis of OS-related factors in the validation cohort. I Risk heatmaps of 15 prognostic genes. * $P \leq 0.05$; $P \leq 0.01$; $P \leq 0.001$ ## Functional analysis and immune cell infiltration GO analysis was performed on related genes in the training group to further understand the biological functions of risk differential genes in TCGA-LUAD samples. Using “limma” R, DEGs meeting the filtering conditions of FDR < 0.05 and \|logFC\|> 1 were extracted. Ninety-one [91] DEGs were screened out from 443 genes in the training risk group, among which 44 genes were considered to be pro-oncogenes and 47 anti-oncogenes (Supplementary Table 2). The GO enrichment results showed that DEGs were mainly involved in the biological process of mitosis, nuclear division and organelle fission, and in cytological components mainly associated with cell–cell junctions (Fig. 4A, B).Fig. 4Immune cell infiltration and functional analysis. A–B GO enrichment analysis based on TCGA queue. C–D Box plot of immune infiltration and function scores in the TCGA cohort. E–F Differences in immune infiltration and functional scores between the two risk groups in the validation set The relationship between immune environment and risk score was further discussed from an immunological perspective, single-sample gene set enrichment analysis (ssGSEA) was used to assess differences in the functional activity of immune cells and pathways. In the TCGA cohort, immature dendritic cells (iDCs), B-cells, activated dendritic cells (aDCs), neutrophils, mast_cells, tumor-infiltrating lymphocytes (TIL), human leukocyte antigen (HLA), T cell co-inhibition, Type_II interferons(IFN) response were infiltrated at high levels in a low-risk cohort, but MHC_class_I and Parainflammation decreased significantly (Fig. 4C, D). In the GEO cohort, iDCs, ADCs, mast_cells, B-cells, neutrophils, regulatory T cells (Treg), HLA, Type_II_IFN response, T cell co-stimulation were higher in low-risk patients, while Th2_cells and MHC class I scores decreased (Fig. 4E, F). The results were similar to those of TCGA. ## Tumor immune response and mutation burden analysis To better understand the status of the immune microenvironment associated with the newly developed risk model, ESTAMATE was performed to calculate Immune scores, stromal scores, ESTIMATE scores, and tumor purity for each risk group, all scores of the two groups compared had no significant differences (Supplementary Fig. S1A–D). In clinical practice, immune checkpoint inhibition is an important approach to cancer treatment. It is an inhibitory molecule that modulates immune activation and kills tumor cells through co-inhibition or co-stimulation. Forty-seven[47] immune checkpoint-associated genes were selected to investigate the relationship between risk models and these genes, the expression of 22 immune checkpoints between different risk groups was studied and observed that most of these had higher levels in the low-score groups (Fig. 5A), correlation analysis showed that there were significant positive correlations among many genes, while the negative correlation between CD40LG and risk scores was most pronounced (Fig. 5B). The risk scores of two different subgroups Cluster1 and Cluster2 were compared, and Cluster1 patients had significantly higher scores than Cluster2 patients ($P \leq 2.22$E−16) (Fig. 5C).Fig. 5Comprehensive assessment of immune characteristics in TCGA-LUAD dataset. A The expression levels of 22 differentially expressed checkpoint genes in two groups of patients. $P \leq 0.05$; $P \leq 0.01$; $P \leq 0.001.$ B Association of risk scores with immune checkpoint related genes. C Comparison of risk scores between clusters. D–F The expression levels of tumor immune dysfunction, rejection and microsatellite instability in the two groups were evaluated. G Correlation analysis* between TIDE fraction and model. H Differential expression of mRNAsi TIDE tool was used to predict the response of the model to immunotherapy, tumor immune dysfunction was significantly higher in the low-score group (Fig. 5D), whereas oncological rejection was significantly lower in the high-risk group (Fig. 5E). However, there was no significant difference in tumor microsatellite instability (MSI) between the two groups (Supplementary Fig. S2A). Moreover, the TIDE score was significantly lower in the high-risk group (Fig. 5F), and there was a significant inverse correlation between TIDE and risk score (Fig. 5G), suggesting patients in the high-risk group would be benefited more from immunotherapy than patients in the low-risk group. There was an unexpected correlation between mRNA expression based-stemness index (mRNAsi), immune cell infiltration and immune checkpoints. Figure 5H shows the high risk group had stronger stem cell characteristics. The immunogenicity of the models was subsequently analyzed using IPS, with higher scores for ips_ctla4_neg_pd1_neg and ips_ctla4_pos_pd1_neg in the low-risk group (Fig. 6A, B), which indicated that low-risk patients had a better response to immunotherapy. However, when PD-1 blockade or its combination with CTLA4 blockade was used, there was no significant difference between the two groups (Supplementary Fig. S2B, C). The TMB expression was then investigated in both clusters and both risk groups, TMB was significantly different between Cluster1 and Cluster2, with Cluster1 having a significantly higher TMB score than Cluster2 (Fig. 6C). The high mutational burden favored the group with the higher score, which was positively and significantly associated with TMB (Fig. 6D). The mutation data in the Training group were used to evaluate the status of TMB in both groups. The top five mutated genes were TP53, TTN, MUC16, RYR2 and CSMD3, and the mutation rate in the high-risk group ($91.28\%$) was higher than that in the low-risk group ($85.78\%$) (Fig. 6E, F). Subsequently, it was observed that higher risk score reflected higher TMB (Fig. 6G). In addition, patients with high tumor mutation burden and low risk had the highest 5-year survival rates (Fig. 6H), while there was no significant difference in 5-year survival between patients with high and low mutations (Supplementary Fig. S2D).Fig. 6Immunotherapy and evaluation of tumor mutational burden. A–B Correlation between IPS of two subtypes and risk characteristics. C Relationship between the two clusters and TMB score. D Boxplot of TMB expression of the training cohort. E–F Comparison of mutations in the top 20 common genes. G Relationship between TMB score and risk score in the training cohort. H Survival analyses for patients stratified by both TMB and Riskscore ## Analysis of drug susceptibility in two risk groups To further enhance the clinical effect of the risk model, its ability to predict drug sensitivity was investigated. Five [05] common chemotherapeutic drugs (Cisplatin, Erlotinib, Vinorelbine, Docetaxel, Gemcitabine) and seven [07] other cancer chemotherapeutic agents (Doxorubicin, Tipifarnib, Bicalutamide, Imatinib, Dasatinib, Pazopanib, Methotrexate) sensitivities were studied in both groups. The results showed that the low-risk group was more sensitive to cancer chemotherapeutic agents except Methotrexate (Fig. 7A–K), implying that low-risk patients were more sensitive to chemotherapy. As a result, low-risk patients are more likely to benefit from these chemotherapeutic drugs. However, high-risk patients were more sensitive to Methotrexate (Fig. 7L) which had better therapeutic effects. Fig. 7Comparison of chemotherapy response in TCGA-LUAD. A–E Sensitivity analysis of common chemotherapeutic agents in two risk groups. F–L Sensitivity analysis of other commonly used chemotherapy drugs in cancer ## Discussion Cancerous cells show a significant increase in metabolic demands compared to normal cells. Cancerous cells produce more glucose and lactate through a transition pattern from oxidative phosphorylation to glycolysis that promotes proliferation, survival, and metastasis [22]. Glycolysis and the production of lactic acid have been paid more and more attention to tumor immune regulation. Lactic acid has been reported to benefit tumor metastasis, promote angiogenesis, and, more importantly, produce immunosuppression, all of which are associated with poor clinical outcomes [23]. The large amount of lactic acid produced by glycolysis leads to an acidic tumor microenvironment, which facilitates immune evasion [24, 25]. Studies have shown that high lactate concentration in the tumor environment not only inhibits the function of T cells but also inhibits NK and T cells activation, thus realizing tumor cell immune evasion [26, 27]. The lactic acid secreted by the tumor can also impair the cytolysis capacity of CD8 + effector T cells, but Treg cells also need to ingestion lactic acid to maintain their high inhibitory functions [28, 29]. Therefore, glycolysis and immune status may be potential biomarkers of cancer growth, invasiveness and metastasis, and identification of IGRG function may have predictive value for the survival and prognosis of LUAD patients. Recent studies have shown that cancer markers such as glycolysis and immunity significantly influence the survival prognosis of LUAD. The changes of immune cells affect the occurrence, proliferation and metastasis of tumors [30–32]. A variety of Immunotargeted drugs have also been developed for extensive clinical treatment of cancer [33]. Tumor immunotherapy is an important means, and a series of immune genes with great clinical potential have been discovered (PD-1, CTLA-4) [34, 35], longer OS for some patients. In recent years, the ability of glycolysis to mediate the immune microenvironment has become a focus of attention. In addition, activated T cells are mainly metabolized through the glycolytic pathway, making them play a stronger eradication role. The results show that the glycolysis of immune and tumor cells is not the same and there are differences between them [36]. Although there is an obvious link between glycolysis and immunity, the relationship has rarely been studied in depth. With advances in bioinformatics and genome sequencing, many models have emerged to assess the potential prognosis of LUAD patients, but most of these analyses are based on genomes or transcriptomes rather than on biological processes. There are increasing evidences that previous clinicopathological factors can no longer meet the need for accurate prediction, additional factors should be considered to synthesize the information. Glycolysis and immune microenvironment are two important biological tumor markers and have great potential in predicting the clinical prognosis of LUAD patients [37, 38]. GRGs and IRGs were included in this study, and IGRG model was constructed through expression data obtained from public databases. IGRG model showed better predictive ability in different subgroups of datasets, and could effectively evaluate the clinical outcome of LUAD patients. Thus, the accuracy of LUAD prognosis indicated by IGRG indicates great potential for clinical application. In this study, a new prognostic model of IGRG was developed using 15 genes from the TCGA database, among them, CCL20 and DKK1 are IRGs, HMMR and LDHA are GRGs, and ANGPTL4 is both immune and glycolysis-related genes. CCL20 chemokine is a powerful immunomodulatory molecule, commonly present in various mucosal tissues of human body, including liver, lymph node, lung, colon [39, 40], and participates in the regulation of structure and immune homeostasis. CCL20 is a key influencing factor in inflammation and immune response, and CCR6 is the only known chemokine receptor. Some tumors have high expression of CCL20 and its receptor, which proves that CCL20 signal transduction is related to the growth and metastasis of cancerous cells [41, 42]. CCL20 is also involved in controlling the immune response, and has been found to be overexpressed in inflammatory bowel disease (IBD), psoriasis and rheumatoid arthritis, leading to the occurrence of autoimmune diseases (AIDs) [43–46]. CCL20/CCR6 was shown to promote the growth of colorectal cancer through ERK phosphorylation in some studies [47]. A major role of CCL20 in cancer is its involvement in cancer metastasis. One study showed that IL-1β induced signaling pathway can directly stimulate the production of CCL20 in lung cancer cells, and activate MAPK and PI3K signaling pathways through its autocrine, which has a positive effect on the progression and invasion of cancer cells in lung tissue [48]. Another study showed that compared to normal tissues or cells, CCR6 was overexpressed in laryngeal cancer tissues or cell lines. P38 was significantly activated through the CCL20/CCR6 axis, and then p38 played a signal transduction function to modify the miRNA spectrum, thereby creating conditions for the metastasis of tumor cells [49]. Another important role of CCL20 is to determine resistance to treatment. For example, upregulations of CCL20 are associated with gefitinib resistance, and CCL20 can be used as a biomarker to predict gefitinib resistance [50]. Therefore, CCL20 can be used as an effective biomarker for the clinical monitoring of LUAD patients. DKK1 is a secretory glycoprotein with stronger inhibitory effects on the Wnt/β-catenin signaling pathway and is also an endogenous Wnt signaling antagonist. As a member of a typical carcinogenic signaling pathway, DKK1 plays an important anticancer role in human cancers [51, 52]. Evidence has shown that DKK1 is not only involved in osteogenesis but also plays a central role in promoting tumor bone metastasis [53, 54]. In the tumor tissue of thyroid papillary carcinoma (PTC), abnormal nuclear localization of β-catenin is associated with poor prognosis of PTC patients and thus contributes to tumor growth. DKK1-secreted protein relocates abnormal expression of β-catenin through Wnt/β-catenin signal transduction, reducing PTC cell survival [55, 56]. The dysregulation of DKK1 gene is a favorable condition for cancer cells to survive and invade. Abnormal expression of DKK1 gene has been detected in a variety of cancer models. In NSCLC, upregulation of DKK1 contributes to cancer, possibly through antagonistic Wnt signaling pathway mediating tumor inhibition of p53 [57]. One study showed that DKK1 was overexpressed in patients with lung and esophageal cancer, leading to poor prognosis in these patients and also becoming a new target for immunotherapy [58]. However, DKK1 is under-expressed in gastric cancer and colorectal cancer, in which DKK1 is regulated by miR-493 and epigenetic silencing, respectively [59]. The activity and expression of DKK1 vary in different cancers, so further exploration of its mechanism is required to verify the prognostic function of DKK1, which can serve as a potential biomarker to accurately predict poor prognosis in patients with these diseases [60]. LDHA is an important energy-metabolizing enzyme with elevated expression in most cancer cells compared to normal tissues [61]. Previous evidence suggests that LDHA mediate tumor spread, invasion, and progression and may be a promising therapeutic target [62–65]. Abnormal expression and upregulation of LDHA are closely associated with a variety of cancers and can be used as a sensitive prognostic factor for lung, liver and pancreatic cancers [66–68]. For example, in gastric cancer (GC), circ-Donson binds to Mir-149-5p, while Mir-149-5p targets LDHA in GC. Down-regulation of circ-Donson inhibits invasion, migration and angiogenesis of tumor cells. However, the high expression of LDHA eventually increased circ-Donson and reduced the inhibition of GC progression [69]. In addition, studies have shown that inhibition of LDHA expression can significantly inhibit cell proliferation, colony formation and migration in lung cancer patients, and enhance their sensitivity to conventional chemotherapy and radiotherapy [70]. Therefore, LDHA is expected to be a promising prognostic indicator for lung cancer treatment. In our study, LDHA was upregulated in patients with a high score, showing a significant association with poor outcomes in LUAD patients. HMMR, also known as RHAMM/CD168, has a relatively non-negligible role in neurodevelopment, tissue homeostasis and cancer progression [71, 72]. HMMR expression increases in many cancer types and is an important potential prognostic factor in cancers such as Glioblastoma, breast cancer, hepatocellular carcinoma (HCC) [73–75]. HMMR is highly expressed and promotes the growth of LUAD cells. Highly expressed miR-34a-5p induces LUAD cell apoptosis and inhibits cancer cell proliferation by targeting HMMR. However, HCG18 sponges Mir-34A-5p in LUAD to regulate HMMR expression, leading to the rapid development of lung adenocarcinoma and reduced clinical survival time in patients with lung cancer [76]. Previous studies have shown that MPPO-AS1 negatively regulates has-let-7b-5p in lung tumor cells, leading to an over-expression of HMMR and promoting the progression of lung adenocarcinoma [77]. HMMR expression was up-regulated in LUAD, and its high expression was correlated with tumor size and lymph node metastasis. In addition, it was significantly associated with adverse clinical features and prognosis [78], while HMMR knockdown significantly inhibited the invasion ability of LUAD tumor cells. Our results also showed that the expression level of HMMR was closely related to the clinical outcome. The higher HMMR expression, the poorer the OS and clinical prognosis of patients are, making it an important biological marker for the treatment of LUAD. ANGPTL4, a member of the ANGPTL (ANGPTL1-8) family, is highly expressed in the human vascular system, adipose tissue and intestinal tract, and is involved in the regulation of vascular permeability, angiogenesis and tumorigenesis. In contrast, ANGPTL4 is more important in tumor energy metabolism, antioxidant and metastasis [79]. Studies have shown that ANGPTL4 might have anti-angiogenic and anti-metastatic effects on gastric cancer through the down-regulation of ERK and epigenetic inhibition [80]. However, colorectal cancer (CRC) studies have identified opposite roles of ANGPTL4. DNA methylation-mediated silencing of ANGPTL4 induces the activation of cancer-associated fibroblasts (CAFs) and help CRC transfer through the ERK pathway, enhancing its invasive ability [81]. Moreover, ANGPTL4 can participate in tumor energy metabolism in different NSCLC cells and affect cell proliferation through this process [82]. High expression of ANGPTL4 predicts adverse clinical outcomes in tumors, such as renal clear cell carcinoma, cholangiocarcinoma, melanoma, bladder cancer, and oral cancer [83–87]. In this study, ANGPTL4 was a high-risk gene that increased with tumor progression, suggesting a reduced survival rate and poor prognosis in LUAD patients. ## Conclusion This study constructed and validated a new prognostic model for LUAD patients based on immune-glycolysis-related genes. The model incorporates clinical prognostic features to predict overall survival in patients diagnosed with LUAD. These findings provide a new method or approach for predicting the prognosis and developing therapeutic strategies for patients with lung adenocarcinoma. ## Supplementary Information Below is the link to the electronic supplementary material. Supplementary file1 (DOCX 198 KB)Supplementary file2 393 genes were selected from survival and expression data of TCGA as prognostic genes (XLS 65 KB)Supplementary file3 91 DEGs were screened from 443 genes in the TCGA risk group (XLS 31 KB) ## References 1. Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin (2021) 71 209-249. DOI: 10.3322/caac.21660 2. Geraci E, Chablani L. Immunotherapy as a second-line or later treatment modality for advanced non-small cell lung cancer: a review of safety and efficacy. Crit Rev Oncol Hematol (2020) 152 103009. DOI: 10.1016/j.critrevonc.2020.103009 3. Gou W, Yu X, Wu S, Wu H, Chang H, Chen L. Targeted inhibition of acidic nucleoplasmic DNA-binding protein 1 enhances radiosensitivity of non-small cell lung cancer. Cancer Lett (2022) 530 100-109. DOI: 10.1016/j.canlet.2022.01.020 4. Siegel RL, Miller KD, Fuchs HE, Jemal A. Cancer statistics, 2022. CA Cancer J Clin (2022) 72 7-33. DOI: 10.3322/caac.21708 5. Feng S, Huang C, Guo L, Wang H, Liu H. A novel epithelial-mesenchymal transition-related gene signature for prognosis prediction in patients with lung adenocarcinoma. Heliyon (2022) 8 e08713. DOI: 10.1016/j.heliyon.2022.e08713 6. Luo J, Du X. A promising prognostic signature for lung adenocarcinoma (LUAD) patients basing on 6 hypoxia-related genes. Medicine (Baltimore) (2021) 100 e28237. DOI: 10.1097/MD.0000000000028237 7. Vander Heiden MG, DeBerardinis RJ. Understanding the intersections between metabolism and cancer biology. Cell (2017) 168 657-669. DOI: 10.1016/j.cell.2016.12.039 8. Wang FF, Zhang JL, Ji Y, Yan XJ, Sun L, Zhu Y. KLF2 mediates the suppressive effect of BDNF on diabetic intimal calcification by inhibiting HK1 induced endothelial-to-mesenchymal transition. Cell Signal (2022) 94 110324. DOI: 10.1016/j.cellsig.2022.110324 9. Keating ST, El-Osta A. Metaboloepigenetics in cancer, immunity and cardiovascular disease. Cardiovasc Res (2022). DOI: 10.1093/cvr/cvac058 10. Liou GY, Döppler H, DelGiorno KE, Zhang L, Leitges M, Crawford HC. Mutant KRas-induced mitochondrial oxidative stress in acinar cells upregulates EGFR signaling to drive formation of pancreatic precancerous lesions. Cell Rep (2016) 14 2325-2336. DOI: 10.1016/j.celrep.2016.02.029 11. Stine ZE, Schug ZT, Salvino JM, Dang CV. Targeting cancer metabolism in the era of precision oncology. Nat Rev Drug Discov (2022) 21 141-162. DOI: 10.1038/s41573-021-00339-6 12. Zhang L, Li Y, Dai Y, Wang D, Wang X, Cao Y. Glycolysis-related gene expression profiling serves as a novel prognosis risk predictor for human hepatocellular carcinoma. Sci Rep (2021) 11 18875. DOI: 10.1038/s41598-021-98381-2 13. Li T, Tong H, Zhu J, Qin Z, Yin S, Sun Y. Identification of a three-glycolysis-related lncRNA signature correlated with prognosis and metastasis in clear cell renal cell carcinoma. Front Med (Lausanne) (2021) 8 777507. DOI: 10.3389/fmed.2021.777507 14. Chang L, Fang S, Gu W. The molecular mechanism of metabolic remodeling in lung cancer. J Cancer (2020) 11 1403-1411. DOI: 10.7150/jca.31406 15. Mamede M, Higashi T, Kitaichi M, Ishizu K, Ishimori T, Nakamoto Y. [18F]FDG uptake and PCNA, Glut-1, and Hexokinase-II expressions in cancers and inflammatory lesions of the lung. Neoplasia (2005) 7 369-379. DOI: 10.1593/neo.04577 16. Li XB, Gu JD, Zhou QH. Review of aerobic glycolysis and its key enzymes—new targets for lung cancer therapy. Thorac Cancer (2015) 6 17-24. DOI: 10.1111/1759-7714.12148 17. Kerr EM, Martins CP. Metabolic rewiring in mutant Kras lung cancer. Febs J (2018) 285 28-41. DOI: 10.1111/febs.14125 18. Hanahan D, Weinberg RA. Hallmarks of cancer: the next generation. Cell (2011) 144 646-674. DOI: 10.1016/j.cell.2011.02.013 19. Chen R, Yang W, Li Y, Cheng X, Nie Y, Liu D. Effect of immunotherapy on the immune microenvironment in advanced recurrent cervical cancer. Int Immunopharmacol (2022) 106 108630. DOI: 10.1016/j.intimp.2022.108630 20. Wu T, Dai Y. Tumor microenvironment and therapeutic response. Cancer Lett (2017) 387 61-68. DOI: 10.1016/j.canlet.2016.01.043 21. Lei Q, Wang D, Sun K, Wang L, Zhang Y. Resistance mechanisms of anti-PD1/PDL1 therapy in solid tumors. Front Cell Dev Biol (2020) 8 672. DOI: 10.3389/fcell.2020.00672 22. Lu J. The Warburg metabolism fuels tumor metastasis. Cancer Metastasis Rev (2019) 38 157-164. DOI: 10.1007/s10555-019-09794-5 23. de la Cruz-López KG, Castro-Muñoz LJ, Reyes-Hernández DO, García-Carrancá A, Manzo-Merino J. Lactate in the regulation of tumor microenvironment and therapeutic approaches. Front Oncol (2019) 9 1143. DOI: 10.3389/fonc.2019.01143 24. Anichini A, Perotti VE, Sgambelluri F, Mortarini R. Immune escape mechanisms in non small cell lung cancer. Cancers (Basel) (2020) 12 3605. DOI: 10.3390/cancers12123605 25. Ngwa VM, Edwards DN, Philip M, Chen J. Microenvironmental metabolism regulates antitumor immunity. Cancer Res (2019) 79 4003-4008. DOI: 10.1158/0008-5472.CAN-19-0617 26. Fischer K, Hoffmann P, Voelkl S, Meidenbauer N, Ammer J, Edinger M. Inhibitory effect of tumor cell-derived lactic acid on human T cells. Blood (2007) 109 3812-3819. DOI: 10.1182/blood-2006-07-035972 27. Brand A, Singer K, Koehl GE, Kolitzus M, Schoenhammer G, Thiel A. LDHA-associated lactic acid production blunts tumor immunosurveillance by T and NK cells. Cell Metab (2016) 24 657-671. DOI: 10.1016/j.cmet.2016.08.011 28. Watson MJ, Vignali PDA, Mullett SJ, Overacre-Delgoffe AE, Peralta RM, Grebinoski S. Metabolic support of tumour-infiltrating regulatory T cells by lactic acid. Nature (2021) 591 645-651. DOI: 10.1038/s41586-020-03045-2 29. Johnson S, Haigis MC, Dougan SK. Dangerous dynamic duo: lactic acid and PD-1 blockade. Cancer Cell (2022) 40 127-130. DOI: 10.1016/j.ccell.2022.01.008 30. Zhou Y, Fan Y, Mao Y, Lou M, Liu X, Yuan K. NCAPG is a prognostic biomarker of immune infiltration in non-small-cell lung cancer. Biomark Med (2022) 16 523-535. DOI: 10.2217/bmm-2021-1090 31. Shi Y, Xu Y, Xu Z, Wang H, Zhang J, Wu Y. TKI resistant-based prognostic immune related gene signature in LUAD, in which FSCN1 contributes to tumor progression. Cancer Lett (2022) 532 215583. DOI: 10.1016/j.canlet.2022.215583 32. Chen MJ, Wang YC, Wang L, Shen CJ, Chen CY, Lee H. PD-L1 expressed from tumor cells promotes tumor growth and invasion in lung cancer via modulating TGF-β1/SMAD4 expression. Thorac Cancer (2022) 13 1322-1332. DOI: 10.1111/1759-7714.14388 33. Ochoa-de-Olza M, Bourhis J, Coukos G, Herrera FG. Radiotherapy plus immune checkpoint blockade in PD(L)-1-resistant metastatic NSCLC. Lancet Oncol (2022) 23 e157. DOI: 10.1016/S1470-2045(22)00154-1 34. Yu H, Chen P, Cai X, Chen C, Zhang X, He L. Efficacy and safety of PD-L1 inhibitors versus PD-1 inhibitors in first-line treatment with chemotherapy for extensive-stage small-cell lung cancer. Cancer Immunol Immunother (2022) 71 637-644. DOI: 10.1007/s00262-021-03017-z 35. Lee JB, Kim HR, Ha SJ. Immune checkpoint inhibitors in 10 years: contribution of basic research and clinical application in cancer immunotherapy. Immune Netw (2022) 22 e2. DOI: 10.4110/in.2022.22.e2 36. Pollizzi KN, Powell JD. Integrating canonical and metabolic signalling programmes in the regulation of T cell responses. Nat Rev Immunol (2014) 14 435-446. DOI: 10.1038/nri3701 37. Li N, Su M, Zhu L, Wang L, Peng Y, Dong B. A prognostic signature of glycolysis-related long noncoding RNAs for molecular subtypes in the tumor immune microenvironment of lung adenocarcinoma. Int J Gen Med (2021) 14 8955-8974. DOI: 10.2147/IJGM.S340615 38. Ren C, Li J, Zhou Y, Zhang S, Wang Q. Typical tumor immune microenvironment status determine prognosis in lung adenocarcinoma. Transl Oncol (2022) 18 101367. DOI: 10.1016/j.tranon.2022.101367 39. Jia SN, Han YB, Yang R, Yang ZC. Chemokines in colon cancer progression. Semin Cancer Biol (2022) 86 400-407. DOI: 10.1016/j.semcancer.2022.02.007 40. Louis I, Dulude G, Corneau S, Brochu S, Boileau C, Meunier C. Changes in the lymph node microenvironment induced by oncostatin M. Blood (2003) 102 1397-1404. DOI: 10.1182/blood-2003-01-0316 41. Zhu H, Tan J, Zhao Y, Wang Z, Wu Z, Li M. Potential role of the chemotaxis system in formation and progression of intracranial aneurysms through weighted gene co-expression network analysis. Int J Gen Med (2022) 15 2217-2231. DOI: 10.2147/IJGM.S347420 42. Nan H, Zhou L, Liang W, Meng J, Lin K, Li M. Epigenetically associated CCL20 upregulation correlates with esophageal cancer progression and immune disorder. Pathol Res Pract (2021) 228 153683. DOI: 10.1016/j.prp.2021.153683 43. Schutyser E, Struyf S, Van Damme J. The CC chemokine CCL20 and its receptor CCR6. Cytokine Growth Factor Rev (2003) 14 409-426. DOI: 10.1016/S1359-6101(03)00049-2 44. Kaser A, Ludwiczek O, Holzmann S, Moschen AR, Weiss G, Enrich B. Increased expression of CCL20 in human inflammatory bowel disease. J Clin Immunol (2004) 24 74-85. DOI: 10.1023/B:JOCI.0000018066.46279.6b 45. Matsui T, Akahoshi T, Namai R, Hashimoto A, Kurihara Y, Rana M. Selective recruitment of CCR6-expressing cells by increased production of MIP-3 alpha in rheumatoid arthritis. Clin Exp Immunol (2001) 125 155-161. DOI: 10.1046/j.1365-2249.2001.01542.x 46. Furue K, Ito T, Tsuji G, Nakahara T, Furue M. The CCL20 and CCR6 axis in psoriasis. Scand J Immunol (2020) 91 e12846. DOI: 10.1111/sji.12846 47. Zhang XP, Hu ZJ, Meng AH, Duan GC, Zhao QT, Yang J. Role of CCL20/CCR6 and the ERK signaling pathway in lung adenocarcinoma. Oncol Lett (2017) 14 8183-8189. PMID: 29250193 48. Wang B, Shi L, Sun X, Wang L, Wang X, Chen C. Production of CCL20 from lung cancer cells induces the cell migration and proliferation through PI3K pathway. J Cell Mol Med (2016) 20 920-929. DOI: 10.1111/jcmm.12781 49. Lu E, Su J, Zhou Y, Zhang C, Wang Y. CCL20/CCR6 promotes cell proliferation and metastasis in laryngeal cancer by activating p38 pathway. Biomed Pharmacother (2017) 85 486-492. DOI: 10.1016/j.biopha.2016.11.055 50. Mao K, Lin F, Zhang Y, Zhou H. Identification of key genes and pathways in gefitinib-resistant lung adenocarcinoma using bioinformatics analysis. Evol Bioinform Online (2021) 17 11769343211023767. DOI: 10.1177/11769343211023767 51. Huang Y, Liu L, Liu A. Dickkopf-1: current knowledge and related diseases. Life Sci (2018) 209 249-254. DOI: 10.1016/j.lfs.2018.08.019 52. Fassio A, Adami G, Benini C, Vantaggiato E, Saag KG, Giollo A. Changes in Dkk-1, sclerostin, and RANKL serum levels following discontinuation of long-term denosumab treatment in postmenopausal women. Bone (2019) 123 191-195. DOI: 10.1016/j.bone.2019.03.019 53. Pinzone JJ, Hall BM, Thudi NK, Vonau M, Qiang YW, Rosol TJ. The role of Dickkopf-1 in bone development, homeostasis, and disease. Blood (2009) 113 517-525. DOI: 10.1182/blood-2008-03-145169 54. Suda T, Yamashita T, Sunagozaka H, Okada H, Nio K, Sakai Y. Dickkopf-1 promotes angiogenesis and is a biomarker for hepatic stem cell-like hepatocellular carcinoma. Int J Mol Sci (2022) 23 2801. DOI: 10.3390/ijms23052801 55. Zhang Z, Li X, Xiao Q, Wang Z. MiR-574-5p mediates the cell cycle and apoptosis in thyroid cancer cells via Wnt/β-catenin signaling by repressing the expression of quaking proteins. Oncol Lett (2018) 15 5841-5848. PMID: 29556311 56. Cho SW, Lee EJ, Kim H, Kim SH, Ahn HY, Kim YA. Dickkopf-1 inhibits thyroid cancer cell survival and migration through regulation of β-catenin/E-cadherin signaling. Mol Cell Endocrinol (2013) 366 90-98. DOI: 10.1016/j.mce.2012.12.007 57. Wang J, Shou J, Chen X. Dickkopf-1, an inhibitor of the Wnt signaling pathway, is induced by p53. Oncogene (2000) 19 1843-1848. DOI: 10.1038/sj.onc.1203503 58. Sato N, Yamabuki T, Takano A, Koinuma J, Aragaki M, Masuda K. Wnt inhibitor Dickkopf-1 as a target for passive cancer immunotherapy. Cancer Res (2010) 70 5326-5336. DOI: 10.1158/0008-5472.CAN-09-3879 59. Sato H, Suzuki H, Toyota M, Nojima M, Maruyama R, Sasaki S. Frequent epigenetic inactivation of DICKKOPF family genes in human gastrointestinal tumors. Carcinogenesis (2007) 28 2459-2466. DOI: 10.1093/carcin/bgm178 60. Yamabuki T, Takano A, Hayama S, Ishikawa N, Kato T, Miyamoto M. Dikkopf-1 as a novel serologic and prognostic biomarker for lung and esophageal carcinomas. Cancer Res (2007) 67 2517-2525. DOI: 10.1158/0008-5472.CAN-06-3369 61. Miao P, Sheng S, Sun X, Liu J, Huang G. Lactate dehydrogenase A in cancer: a promising target for diagnosis and therapy. IUBMB Life (2013) 65 904-910. DOI: 10.1002/iub.1216 62. Ye Y, Chen M, Chen X, Xiao J, Liao L, Lin F. Clinical significance and prognostic value of lactate dehydrogenase expression in cervical cancer. Genet Test Mol Biomark (2022) 26 107-117. DOI: 10.1089/gtmb.2021.0006 63. Sheng SL, Liu JJ, Dai YH, Sun XG, Xiong XP, Huang G. Knockdown of lactate dehydrogenase A suppresses tumor growth and metastasis of human hepatocellular carcinoma. Febs J (2012) 279 3898-3910. DOI: 10.1111/j.1742-4658.2012.08748.x 64. Hou XM, Yuan SQ, Zhao D, Liu XJ, Wu XA. LDH-A promotes malignant behavior via activation of epithelial-to-mesenchymal transition in lung adenocarcinoma. Biosci Rep. (2019) 39 BSR20181476. DOI: 10.1042/BSR20181476 65. Li L, Kang L, Zhao W, Feng Y, Liu W, Wang T. miR-30a-5p suppresses breast tumor growth and metastasis through inhibition of LDHA-mediated Warburg effect. Cancer Lett (2017) 400 89-98. DOI: 10.1016/j.canlet.2017.04.034 66. Yu C, Hou L, Cui H, Zhang L, Tan X, Leng X. LDHA upregulation independently predicts poor survival in lung adenocarcinoma, but not in lung squamous cell carcinoma. Future Oncol (2018) 14 2483-2492. DOI: 10.2217/fon-2018-0177 67. Shi Q, Liu Y, Lu M, Lei QY, Chen Z, Wang L. A pathway-guided strategy identifies a metabolic signature for prognosis prediction and precision therapy for hepatocellular carcinoma. Comput Biol Med (2022) 144 105376. DOI: 10.1016/j.compbiomed.2022.105376 68. Chen Q, Pu N, Yin H, Zhang J, Zhao G, Lou W. A metabolism-relevant signature as a predictor for prognosis and therapeutic response in pancreatic cancer. Exp Biol Med (Maywood) (2022) 247 120-130. DOI: 10.1177/15353702211049220 69. Shao Y, Li F, Liu H. Circ-DONSON facilitates the malignant progression of gastric cancer depending on the regulation of miR-149-5p/LDHA axis. Biochem Genet (2022) 60 640-655. DOI: 10.1007/s10528-021-10120-4 70. Yang Y, Chong Y, Chen M, Dai W, Zhou X, Ji Y. Targeting lactate dehydrogenase a improves radiotherapy efficacy in non-small cell lung cancer: from bedside to bench. J Transl Med (2021) 19 170. DOI: 10.1186/s12967-021-02825-2 71. Shi J, Chen Y, Wang Z, Guo J, Tong C, Tong J. Comprehensive bioinformatics analysis to identify the gene HMMR associated with lung adenocarcinoma prognosis and its mechanism of action in multiple cancers. Front Oncol (2021) 11 712795. DOI: 10.3389/fonc.2021.712795 72. Casini P, Nardi I, Ori M. RHAMM mRNA expression in proliferating and migrating cells of the developing central nervous system. Gene Expr Patterns (2010) 10 93-97. DOI: 10.1016/j.gep.2009.12.003 73. Tilghman J, Wu H, Sang Y, Shi X, Guerrero-Cazares H, Quinones-Hinojosa A. HMMR maintains the stemness and tumorigenicity of glioblastoma stem-like cells. Cancer Res (2014) 74 3168-3179. DOI: 10.1158/0008-5472.CAN-13-2103 74. Mateo F, He Z, Mei L, de Garibay GR, Herranz C, García N. Modification of BRCA1-associated breast cancer risk by HMMR overexpression. Nat Commun (2022) 13 1895. DOI: 10.1038/s41467-022-29335-z 75. Tang YP, Yin YX, Xie MZ, Liang XQ, Li JL, Li KZ. Systematic analysis of the clinical significance of hyaluronan-mediated motility receptor in colorectal cancer. Front Mol Biosci (2021) 8 733271. DOI: 10.3389/fmolb.2021.733271 76. Li W, Pan T, Jiang W, Zhao H. HCG18/miR-34a-5p/HMMR axis accelerates the progression of lung adenocarcinoma. Biomed Pharmacother (2020) 129 110217. DOI: 10.1016/j.biopha.2020.110217 77. Jiang X, Tang L, Yuan Y, Wang J, Zhang D, Qian K. NcRNA-mediated high expression of HMMR as a prognostic biomarker correlated with cell proliferation and cell migration in lung adenocarcinoma. Front Oncol (2022) 12 846536. DOI: 10.3389/fonc.2022.846536 78. Zhou Q, Fan G, Dong Y. Polo-like kinase 4 correlates with greater tumor size, lymph node metastasis and confers poor survival in non-small cell lung cancer. J Clin Lab Anal (2020) 34 e23152. DOI: 10.1002/jcla.23152 79. Yang X, Cheng Y, Su G. Biosci Rep (2018). DOI: 10.1042/BSR20180557 80. Okochi-Takada E, Hattori N, Tsukamoto T, Miyamoto K, Ando T, Ito S. ANGPTL4 is a secreted tumor suppressor that inhibits angiogenesis. Oncogene (2014) 33 2273-2278. DOI: 10.1038/onc.2013.174 81. Zhang K, Zhai Z, Yu S, Tao Y. DNA methylation mediated down-regulation of ANGPTL4 promotes colorectal cancer metastasis by activating the ERK pathway. J Cancer (2021) 12 5473-5485. DOI: 10.7150/jca.52338 82. Xiao S, Nai-Dong W, Jin-Xiang Y, Long T, Xiu-Rong L, Hong G. ANGPTL4 regulate glutamine metabolism and fatty acid oxidation in nonsmall cell lung cancer cells. J Cell Mol Med (2022) 26 1876-1885. DOI: 10.1111/jcmm.16879 83. Han X, Song D. Using a machine learning approach to identify key biomarkers for renal clear cell carcinoma. Int J Gen Med (2022) 15 3541-3558. DOI: 10.2147/IJGM.S351168 84. Aung TM, Ciin MN, Silsirivanit A, Jusakul A, Lert-Itthiporn W, Proungvitaya T. Serum angiopoietin-like protein 4: a potential prognostic biomarker for prediction of vascular invasion and lymph node metastasis in cholangiocarcinoma patients. Front Public Health (2022) 10 836985. DOI: 10.3389/fpubh.2022.836985 85. Izraely S, Ben-Menachem S, Sagi-Assif O, Meshel T, Marzese DM, Ohe S. ANGPTL4 promotes the progression of cutaneous melanoma to brain metastasis. Oncotarget (2017) 8 75778-75796. DOI: 10.18632/oncotarget.19018 86. Hsieh HY, Jou YC, Tung CL, Tsai YS, Wang YH, Chi CL. Epigenetic silencing of the dual-role signal mediator, ANGPTL4 in tumor tissues and its overexpression in the urothelial carcinoma microenvironment. Oncogene (2018) 37 673-686. DOI: 10.1038/onc.2017.375 87. Pérez-Valencia JA, Prosdocimi F, Cesari IM, da Costa IR, Furtado C, Agostini M. Angiogenesis and evading immune destruction are the main related transcriptomic characteristics to the invasive process of oral tongue cancer. Sci Rep (2018) 8 2007. DOI: 10.1038/s41598-017-19010-5
--- title: Impaired glymphatic drainage underlying obstructive sleep apnea is associated with cognitive dysfunction authors: - Jiuqi Wang - Yiming Tian - Chi Qin - Lin Meng - Renyi Feng - Shuqin Xu - Yanping Zhai - Dongxiao Liang - Rui Zhang - Haiyan Tian - Han Liu - Yongkang Chen - Yu Fu - Pei Chen - Qingyong Zhu - Junfang Teng - Xuejing Wang journal: Journal of Neurology year: 2023 pmcid: PMC10025229 doi: 10.1007/s00415-022-11530-z license: CC BY 4.0 --- # Impaired glymphatic drainage underlying obstructive sleep apnea is associated with cognitive dysfunction ## Abstract Obstructive sleep apnea (OSA) is highly prevalent but easily undiagnosed and is an independent risk factor for cognitive impairment. However, it remains unclear how OSA is linked to cognitive impairment. In the present study, we found the correlation between morphological changes of perivascular spaces (PVSs) and cognitive impairment in OSA patients. Moreover, we developed a novel set of dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) methods to evaluate the fluid dynamics of glymphatic drainage system. We found that the inflow and outflow parameters of the glymphatic drainage system in patients with OSA were obviously changed, indicating impairment of glymphatic drainage due to excessive perfusion accompanied with deficient drainage in OSA patients. Moreover, parameters of the outflow were associated with the degree of cognitive impairment, as well as the hypoxia level. In addition, continuous positive airway pressure (CPAP) enhances performance of the glymphatic drainage system after 1 month treatment in OSA patients. We proposed that ventilation improvement might be a new strategy to ameliorate the impaired drainage of glymphatic drainage system due to OSA-induced chronic intermittent hypoxia, and consequently improved the cognitive decline. ### Supplementary Information The online version contains supplementary material available at 10.1007/s00415-022-11530-z. ## Introduction The glymphatic drainage system is a highly organized fluid clearance pathway between cerebrospinal fluid (CSF) and interstitial fluid (ISF), which subserves the CSF influx into the brain through periarterial spaces (PASs), and ISF efflux from the brain into perivenous spaces (PVESs) that eventually reaches the subarachnoid space [1–3], which potentially drains the metabolic wastes from the brain [4]. Studies show that the dysfunction of glymphatic drainage system causes cognitive impairment[5], and it has been clarity that the dysfunction of glymphatic drainage was associated with the impaired cognitive function in Alzheimer's disease (AD) and Parkinson’s disease (PD) [6, 7]. Obstructive sleep apnea (OSA) is a common sleep related disorder with the symptoms of upper airway collapse, apnea, hypoxia and recurrent arousals, accompanied by decline in cognitive function [8, 9]. Sleep-disordered breathing was linked to an increased risk of cognitive impairment and AD in the elderly population [10–14]. However, the mechanism and prognosis of cognitive impairment in OSA patients remain unclear. Treatment with continuous positive airway pressure (CPAP) is the main treatment for OSA, which was confirmed to delay progression and delay the decline of cognitive function [15–18]. The mechanism of improvement in cognitive impairment, however, is unknown. In the present work, we used the MRI axial T2 sequence to compare morphological changes of glymphatic drainage system in patients with OSA and normal controls (NCs). The diagnostic accuracy of morphological changes was analyzed and the correlations between the glymphatic drainage system morphological changes and cognitive impairment were confirmed. In addition, we developed a set of novel DCE-MRI techniques for assessing the fluid dynamics in the glymphatic drainage system. We investigated that the dysfunction of outflow drainage of the glymphatic drainage system was correlated with the cognitive decline. Moreover, treatment with CPAP could promote the outflow drainage of the glymphatic dynamic and further improve the cognitive function in OSA patients. ## Approval and patient informed consent This study was authorized by the Institutional Ethics Committees of The First Affiliated Hospital of Zhengzhou University (2022-KY-0282-004). Informed and signed consent was obtained from all participants or their legal guardians. ## Participants 52 OSA patients were recruited based on the following criteria: [1] have no other medical illnesses, such as a history of high blood pressure, cardiopulmonary disease, neurological, kidney, or liver diseases, diabetes, or cancer, as well as no prior upper airway surgery, pulmonary surgery, or snoring therapy, [2] have no structural abnormalities on brain MRI with visual inspections, and [3] diagnosed with OSA on apnea–hypopnea index (AHI) ≥ 5 measured by polysomnography (PSG) monitoring combined with symptoms, such as sleepiness or chronic snoring. In addition, 56 healthy participants were also recruited through the Physical Examination Center of the First Affiliated Hospital of Zhengzhou University. Physical examination, ECG, laboratory testing, such as blood and urine routine, liver and kidney function, MRI, and magnetic resonance angiography scans, all revealed that they were in good conditions. All healthy participants had with no history of medical or neurological disorders, as determined by two attending neurologists and a psychiatrist. All the participants underwent a clinical assessment and were investigated by two board-certified neurologists who had experience with neurodegenerative diseases. We applied the Mini-mental State Examination (MMSE) and the Montreal Cognitive Assessment (MoCA) to assess cognitive function, the Pittsburgh Sleep Quality Index (PSQI) and the Epworth sleepiness scale (ESS) for the sleep assessment. ## PSG PSG was used to capture the electroencephalogram, bilateral electro-oculograms, surface EMG, electrocardiogram, chest and abdominal wall movements, and oxygen saturation (SaO2) using a pulse oximeter in OSA patients. According to predetermined standards, respiratory episodes were scored. [ 19, 20]. Apnea was defined as a complete halt of airflow lasting at least 10 s. Hypopneas were defined as a noticeable decrease in airflow lasting at least 10 s and accompanied by at least $3\%$ desaturation. The total number of obstructive apneas and obstructive hypopneas per hour of sleep was used to defined the apnea–hypopnea index (AHI). Total oxygen desaturation index (ODI) was outlined as the sum of all desaturations of at least $3\%$ for each hour of total sleep. According to frequently used clinical cutoffs, the following groups were established: no OSA (AHI < 5); mild-moderate OSA group (AHI ≥ 5 but < 30); and severe OSA group (AHI ≥ 30). Moreover, the hypoxemia severity groups were established: mild hypoxemia group (lowest SaO2 (LSaO2) ≥ $80\%$ but < 90); and severe hypoxemia group (LSaO2 < $80\%$)[21]. ## Transcranial doppler For the TCD examination, a transcranial doppler (TCD) analyzer (EMS-9A, Shenzhen Delica Medical Equipment Co., Ltd.) with a 1.6 MHz frequency setting was utilized. The pulsatility index of the middle cerebral artery was investigated via the temporal window. ## MRI procedures Vital signs monitoring was performed before and immediately after DCE-MRI scans, including blood pressure and heart rates. Blood flow velocities and blood vessel diameters of the internal carotid arteries and the external carotid arteries were measured in NCs ($$n = 25$$), before-CPAP-treatment OSA ($$n = 11$$) and after-CPAP-treatment OSA ($$n = 13$$) groups using color Doppler ultrasound (Vivid E95, GE Healthcare). All images were acquired using a 3 T MRI scanner (Skyra, Siemens Healthcare) with a standard 20-channel head coil for radiofrequency transmission. The MRI protocol included the following sequences:To visualize the PVSs and the ventricles, the 3D T2-weighted sequence was used with main sequence parameters as follows: TR/TE = $\frac{4200}{110}$ ms, flip angle 150°, FOV 240 mm, acquisition matrix 320 × 270, 5.0 mm thickness and acquisition time of 46 s.To detect the perfusion of the brain parenchyma from capillaries and the perivascular fluid flow dynamics patterns, and the following high-resolution MRI sequences were used.3D T1-weighted sequence was used with main sequence parameters as follows: repetition time/echo time (TR/TE) = $\frac{190}{2.6}$ ms, flip angle 70°, FOV 250 mm, acquisition matrix 288 × 230, 5.0 mm thickness and acquisition time of 44 s.T1 mapping sequence was used with main sequence parameters as follows: TR/TE = $\frac{5.3}{2.4}$ ms, flip angle 2.99999984411°, FOV 240 mm, acquisition matrix 192 × 126, 5.0 mm thickness and acquisition time of 1 min 28 s.DCE-MRI: the 3D T1-vibe sequence was used with main sequence parameters as follows: TR/TE = $\frac{2.8}{0.8}$ ms, flip angle 14.9999992205°, FOV 220 mm, acquisition matrix 128 × 128, 80 contiguous sections with 5 mm thickness and acquisition time of 9 min 59 s. For the DCE scan, the suggested dosage (0.1 mmol/kg−1) of gadobutrol (Gadovist, Bayer Pharma AG) was administered intravenously with an automatic high-pressure syringe (Spectris MRI Injector System, Medrad) for a stable injection speed. ## Motion correction During head MRI scans, a series of measures were used to avoid movement artifacts, as previously described [22]:Long-term averaging method was taken to reduce motion artifacts generated by swallowing and cerebral artery pulsation. During MRI scans, the head of participants was fixed by placing folded towels in the head coil. Behavioral interventions for reducing head motion during MRI scans. In detail, a cross logo was fixed in the middle and upper part of the machine and participants were told to stare at the fixation cross logo throughout the MRI scans. Those who could not complete the MRI scans or whose images were rated as blurred after motion correction were excluded from the analysis. ## MRI analysis Three radiologists with a combined 10 years of expertise in MRI analysis analyzed all of the MRI data using commercial image viewing software (IntelliSpace Portal v.7, Philips Healthcare), post-processing software (syngoMMWP VE40A, Siemens AG), and Syngo. VIA. They were blinded to the patients’ information. Furthermore, data were analyzed independently by a trained team consisting of three board-certified neuroradiologists, who were kept in the dark regarding the patient's name and medical background. Fiji/ImageJ software was used to analyze T2 axial pictures while obscuring patient data. Following binarization and inversion, these images were used to pinpoint the regions of interest (ROIs) of the PVSs in the frontal cortex and basal ganglia, the bilateral lateral ventricles, the fourth ventricle, the total brain parenchyma at the level of the frontal cortex, the total brain parenchyma of the bilateral basal ganglia, the total brain parenchyma at the level of the lateral ventricles, and the total cerebellum parenchyma at the level of the fourth ventricle. The relative area ratios of the PVSs in the bilateral frontal cortex and the basal ganglia were calculated from the area of PVSs in the frontal cortex or basal ganglia divided by the total brain area at the level of the frontal cortex or the total area of the bilateral basal ganglia. The relative area ratios of the lateral ventricles were calculated from the area of lateral ventricles divided by the total brain area at the level of the lateral ventricles. The relative area ratios of the fourth ventricle were calculated from the area of the fourth ventricle divided by the total cerebellum area at the level of the fourth ventricle, respectively. Importing the data of DCE-MRI scan into TISSUE 4D software for motion correction and image registration, so that signal strength was converted to gadolinium concentration. DCE-MRI pseudo-color pictures were created automatically and combined with T2 images. The modified two-compartment Tofts model was fitted to the DCE-MRI images, and the arterial input function was chosen as medium. ROIs that outline the structure of PVSs were expertly and manually defined at the bilateral frontal cortex and basal ganglia according to T2-weighted MRI images by radiologists. The PVSs had high intensity as seen in T2-weighted images, while had low intensity in T1-weighted images [23]. Using concentration–time curves (CTCs), obtained DCE-MRI data of PVSs were derived and calculated characteristic parameters including peak concentration, wash-in rate, and wash-out rate. CTCs were generated by averaging contrast concentration within each ROI at different time points using the mean curve function in Syngo. VIA software. Each average CTC was derived from the average of the curves belonging to the corresponding groups using ggplot2 R package (https://ggplot2.tidyverse.org.). Clustering was performed using the K-means clustering algorithm (https://CRAN.R-project.org/package=factoextra) based on the feature parameters of CTCs including wash-in rate and peak concentrations. ## Statistical analysis Statistical analyses were performed and Figures were illustrated using R (version 4.0.5), GraphPad Prism (version 8.0.0 for MacOS, GraphPad Software, La Jolla, CA, USA) and SPSS 26.0 (IBM). The clinical and demographic continuous data were represented by mean ± s.d. Demographic and clinical characteristics were compared using a chi-squared test and two-sided Mann–Whitney U-test. Normality testing and homogeneity tests of variance for all continuous variables were made before the analysis. The values of the relative area ratios of the bilateral lateral ventricles and the relative area ratios of the fourth ventricles in different groups were compared using a Mann–Whitney U-test, respectively. The values of the relative area ratios of the PVSs in the bilateral frontal cortex and the basal ganglia were compared using a Mann–Whitney U-test, respectively. The diagnostic accuracy was evaluated using ROC curve analysis. Spearman correlation was used to test the association among the relative area ratios of the bilateral lateral ventricles, the PVSs in the basal ganglia, the wash-out rate values of type I CTCs of frontal cortex of OSA patients, AHI and oxygen desaturation index (ODI) measured by PSG monitoring, and the MMSE scores, the MoCA scores, the PSQI scores and the ESS scores. ( https://ggplot2.tidyverse.org.) ( https://CRAN.R-project.org/package=ggridges.) DCE-MRI parameters in different groups were defined and visualized in clusters using K-means cluster analysis (https://CRAN.R-project.org/package=factoextra.) The best number of clusters was determined by using the elbow method [24] (http://www.jstatsoft.org/v61/i$\frac{06}{.}$) The values of DCE-MRI parameters from the CTCs in different groups were compared using a Mann–Whitney U-test, respectively. $P \leq 0.05$ was considered statistically significant. ## Demographics The demographic and clinical characteristics of the NCs and OSA are shown in the Consolidated Standards of Reporting Trials (CONSORT) flow diagram (Fig. 1, Table 1, Supplementary Table 1, Supplementary Table 2). 59 participants (NCs, $$n = 31$$; OSA, $$n = 28$$) completed 3D T2-weighted MRI scans to determine the relative area ratios of the perivascular spaces (PVSs) in the bilateral frontal cortex and the bilateral basal ganglia, and the relative area ratios of the lateral ventricles and the fourth ventricle. Dynamic contrast-enhanced MRI (DCE-MRI) scans were performed on 55 participants (NCs, $$n = 25$$; OSA before CPAP treatment, $$n = 11$$; OSA after CPAP treatment, $$n = 13$$).Fig. 1CONSORT diagram. Consolidated Standards of Reporting Trials flow diagram showing study participants screening, eligibility and inclusionTable 1Demographics and clinical characteristics of the participantsNCsOSANo. (% female)56 ($44.6\%$)52 ($42.3\%$)Age, years52.0 (11.7)52.6 (10.0)AHINA28.9 (18.0)ODINA30.7 (20.8)LSaO2 (%)NA79.7 (8.9)BMI (kg/m2)25.6 (2.0)28.4 (2.4)MMSENA27.7 (2.3)MoCANA27.0 (2.4)PSQINA8.1 (3.7)ESSNA6.3 (3.0)Data are given as mean (SD). Mean (standard deviation) and N (%) were reported. Demographic factors and clinical characteristics were compared using chi-square test and two-sided Mann–Whitney testsOSA obstructive sleep apnea, NCs normal controls, AHI Apnea–hypopnea index, ODI Oxygen desaturation index, LSaO2 oxygen saturation, MMSE Mini Mental State Examination, MoCA Montreal Cognitive Assessment, BMI Body Mass Index, PSQI Pittsburgh Sleep Quality Index and, ESS Epworth sleepiness scale, NA not applicable Furthermore, there were notable differences between the NCs and OSA groups in terms of systolic blood pressure (SBP), diastolic blood pressure (DBP), respiratory rate (RR) and heart rate (HR) either before or after MRI scans (Supplementary Fig. 1). We assessed the diameters of the internal and external carotid arteries, as well as the pulse index of the middle cerebral arteries, to determine the participants' hemodynamic state. The hemodynamic parameters of the OSA group were not significantly different from those of the NCs group, according to the statistical analysis. ( Supplementary Fig. 1). ## PVS and ventricular morphology changes and the correlation with cognitive impairment in OSA patients Morphological changes of PVSs in the bilateral frontal cortex and basal ganglia were first assessed using the axial T1-weighted MR images in NCs and OSA groups (Fig. 2A,B). When comparing the OSA group to the NCs group, statistical analysis revealed that the relative area ratios of PVSs in both the bilateral frontal cortex and the basal ganglia were significantly higher in the OSA group (Fig. 2C). Then, the PVSs in the bilateral frontal cortex and basal ganglia between mild-moderate OSA and severe OSA groups were compared. The statistical results showed that the relative area ratios of PVSs in both the bilateral frontal cortex and the basal ganglia were significantly higher in the severe OSA group than in the mild-moderate OSA group (Fig. 2E). Furthermore, PVSs in the bilateral frontal cortex and basal ganglia were also compared between OSA with mild hypoxemia and OSA with severe hypoxemia groups. The OSA with severe hypoxemia group had higher relative area ratios of PVSs in both the bilateral frontal cortex and the basal ganglia compared to those in OSA with mild hypoxemia group (Fig. 2G). Next, receiver operating characteristic (ROC) curve analysis was used to evaluate the diagnostic accuracy of the PVSs in the bilateral frontal cortex and basal ganglia to discriminate between OSA and NCs. The results revealed that the relative area ratios of PVSs in the bilateral frontal cortex (sensitivity of $90.32\%$, specificity of $96.43\%$, AUROC of 0.9781, $P \leq 0.0001$) and the bilateral basal ganglia (sensitivity of $93.55\%$, specificity of $92.86\%$, AUROC of 0.9677, $P \leq 0.0001$) (Fig. 2D, Supplementary Table 3) all had the adequate accuracy to distinguish OSA from NCs (Fig. 2D, Supplementary Table 3). In addition, the relative area ratios of PVSs in the bilateral frontal cortex (sensitivity of $90.91\%$, specificity of $94.12\%$, AUROC of 0.9305, $$P \leq 0.0002$$) and in the bilateral basal ganglia (sensitivity of $81.82\%$, specificity of $94.12\%$, AUROC of 0.9251, $$P \leq 0.0002$$) all had the adequate accuracy to distinguish severe OSA from mild-moderate OSA (Fig. 2F, Supplementary Table 3). Moreover, the relative area ratios of PVSs in the bilateral frontal cortex (sensitivity of $90.00\%$, specificity of $88.89\%$, AUROC of 0.8944, $$P \leq 0.0007$$) and in the bilateral basal ganglia (sensitivity of $70.00\%$, specificity of $83.33\%$, AUROC of 0.7389, $$P \leq 0.0392$$) all had the appropriate accuracy to distinguish OSA with severe hypoxemia from OSA with mild hypoxemia (Fig. 2H, Supplementary Table 3).Fig. 2Measurement of the PVSs and ventricles areas by MRI. ( A) Representative T2 axial MRI scans of PVSs in the bilateral frontal cortex and the bilateral basal ganglia of the NCs. The irregular regions within the closed yellow contour represent the frontal cortex or bilateral basal ganglia. The irregular red regions represent the PVSs. Representative T2 axial MRI scans of the bilateral lateral ventricle of the NCs. The irregular blue regions represent the bilateral lateral ventricles. The irregular regions within the closed yellow contour represent the total brain area at the level of the lateral ventricle. Representative images of the fourth ventricle of the NCs. The irregular blue regions represent the fourth ventricle. The irregular regions within the closed yellow contour represent the total cerebellum area at the level of the fourth ventricle. Scale bar, 20 mm. ( B) Representative T2 axial MRI scans of PVSs in the bilateral frontal cortex and the bilateral basal ganglia, and the bilateral lateral ventricle and the fourth ventricle of the OSA. Scale bar, 20 mm. ( C) Comparison of the relative area ratios of the bilateral frontal cortex, basal ganglia, lateral ventricle and the fourth ventricles between the NCs ($$n = 28$$) and OSA ($$n = 31$$) groups (Mann–Whitney U-test). ( D) ROC curves of the relative area ratios of the bilateral frontal cortex, basal ganglia, lateral ventricle and the fourth ventricles for distinguishing OSA ($$n = 31$$) from NCs ($$n = 28$$). ( E) Comparison of the relative area ratios of the bilateral frontal cortex, basal ganglia, lateral ventricle and the fourth ventricles between the mild-moderate OSA ($$n = 17$$) and severe OSA ($$n = 11$$) groups (Mann–Whitney U-test). ( F) ROC curves of the relative area ratios of the bilateral frontal cortex, basal ganglia, lateral ventricle and the fourth ventricles for distinguishing severe OSA ($$n = 11$$) from mild-moderate OSA ($$n = 17$$). ( G) Comparison of the relative area ratios of the bilateral frontal cortex, basal ganglia, lateral ventricle and the fourth ventricles between the OSA with mild hypoxemia ($$n = 18$$) and OSA with severe hypoxemia ($$n = 10$$) groups (Mann–Whitney U-test). ( H) ROC curves of the relative area ratios of the bilateral frontal cortex, basal ganglia, lateral ventricle and the fourth ventricles for distinguishing OSA with severe hypoxemia ($$n = 10$$) from OSA with mild hypoxemia ($$n = 18$$) To further evaluate the CSF circulation, the relative area ratios of both bilateral lateral ventricles and the fourth ventricle were measured using the axial T2 sequence (Fig. 2A, B). The findings showed that the relative area ratios of both the bilateral lateral ventricles and the fourth ventricle in the OSA group were increased compared with those in the NCs group (Fig. 2C). Additionally, the severe OSA group has higher relative area ratios of both the bilateral lateral ventricles and the fourth ventricle than mild-moderate OSA group (Fig. 2E). Besides, in the OSA with severe hypoxemia group, the relative area ratios of both the bilateral lateral ventricles and the fourth ventricle are higher than in the OSA with mild hypoxemia group (Fig. 2G). Then, ROC curve analysis showed that the relative area ratios of the bilateral lateral ventricles (sensitivity of $87.10\%$, specificity of $100.0\%$, AUROC of 0.9804, $P \leq 0.0001$) and the fourth ventricle (sensitivity of $70.97\%$, specificity of $67.86\%$, AUROC of 0.7281, $$P \leq 0.0027$$) all showed the desirable accuracy to distinguish OSA from NCs (Fig. 2D, Supplementary Table 3). In addition, the relative area ratios of the bilateral lateral ventricles (sensitivity of $100.0\%$, specificity of $70.59\%$, AUROC of 0.8663, $$P \leq 0.0013$$) and the fourth ventricle (sensitivity of $90.91\%$, specificity of $82.35\%$, AUROC of 0.9037, $$P \leq 0.004$$) all had the adequate accuracy to distinguish severe OSA from mild-moderate OSA (Fig. 2F, Supplementary Table 3). Moreover, the relative area ratios of the bilateral lateral ventricles (sensitivity of $70.00\%$, specificity of $94.44\%$, AUROC of 0.7170, $P \leq 0.0001$) and the fourth ventricle (sensitivity of $70.00\%$, specificity of $88.89\%$, AUROC of 0.8111, $$P \leq 0.0073$$) all showed appropriate accuracy for distinguishing OSA with severe hypoxemia from OSA with mild hypoxemia (Fig. 2H, Supplementary Table 3). Next, Spearman correlation was used to test the correlations among morphological changes of PVSs and cerebral ventricle, the AHI and the ODI measured by PSG monitoring, the MMSE scores and the MoCA scores. The correlation analysis indicated that the relative area ratios of PVSs in both the bilateral frontal cortex and the basal ganglia were positively correlated with the relative area ratios of both bilateral lateral ventricles and the fourth ventricle, respectively (Fig. 4A). In addition, the AHI showed a strong positive correlation with the relative area ratios of PVSs in both the bilateral frontal cortex, and moderate positive correlations with the relative area ratios of PVSs in both the bilateral basal ganglia and the relative area ratios of both bilateral lateral ventricles and the fourth ventricle (Fig. 4A). Moreover, the ODI showed moderate positive correlations with the relative area ratios of PVSs in both the bilateral frontal cortex and the basal ganglia, and the relative area ratios of the fourth ventricle (Fig. 4A). Furthermore, the MMSE score was strongly negatively correlated with the relative area ratios of PVSs in both the bilateral frontal cortex, and moderately negatively correlated with the relative area ratios of PVSs in both the bilateral basal ganglia and the relative area ratios of both bilateral lateral ventricles and the fourth ventricle (Fig. 4A), while the MoCA score was strongly negatively correlated with the relative area ratios of PVSs in both the bilateral frontal cortex, moderately negatively correlated with the relative area ratios of both bilateral lateral ventricles, and weakly negatively correlated with the relative area ratios of PVSs in both the bilateral basal ganglia and the relative area ratios of the fourth ventricle (Fig. 4A). Moreover, the PSQI score showed a moderate positive correlation with the relative area ratios of PVSs in both the bilateral frontal cortex and the bilateral basal ganglia, while the ESS score was moderate positively correlated with the relative area ratios of PVSs in both the bilateral frontal cortex and the relative area ratios of the fourth ventricle, and weakly positively correlated with the relative area ratios of PVSs in both the bilateral basal ganglia (Fig. 4A). Furthermore, the spearman correlation was also used to test the correlations among other PSG score and morphological changes of PVSs and ventricles. The correlation analysis showed that the apnea index (AI) was positively correlated with the relative area ratios of PVSs in both the bilateral frontal cortex and the relative area ratios of both bilateral lateral ventricles, while the time spent with SaO2 < $90\%$ (T90) was positively correlated with morphological changes of PVSs and the cerebral ventricle (Supplementary Fig. 2). These findings demonstrated that there were correlations among morphological changes of PVSs, ventricles, sleep habits, and cognitive impairment in OSA patients. Moreover, the MMSE score was strongly negatively correlated with the AHI and the ODI and the MoCA score was moderately negatively correlated with the AHI and the ODI, while the PSQI score was moderately positively correlated with the AHI and the ODI, and the ESS score was strongly positively correlated with the AHI and the ODI, which indicated that clinical manifestations were also correlated with the cognitive impairment in OSA patients. ## Impaired glymphatic drainage and correlation with intermittent attacks and cognitive impairment in OSA patients To detect the fluid dynamics in the glymphatic drainage system, DCE-MRI was utilized to identify the fluid flow of PVSs of bilateral frontal cortex, and CTCs were extracted by tracing the borders of all the visible PVSs. Two types of CTCs were recognized using k-means cluster analysis based on feature characteristics including wash-in rates and peak concentrations in the NCs and OSA groups. The type I CTCs showed a steeper climbing slope and greater peak concentration than the type II CTCs, which was compatible with the fluid flow patterns in the PASs. In contrast, the type II CTCs had a lower ascending slope and peak concentration, which depicted the fluid flow patterns in the PVESs. We first identified 275 CTCs of the PVSs of frontal cortex by k-means cluster in the two groups (Fig. 3A). The average curves generated by all the CTCs of type I (the top) and type II (the bottom) in the frontal cortex were shown in Fig. 3B.Fig. 3Measurement of fluid flow of PVSs by DCE-MRI. ( A) K-means cluster outcome of all CTCs in the bilateral frontal cortex of the NCs ($$n = 236$$) and before-CPAP-treatment OSA ($$n = 253$$) groups. The red drops represent type I CTCs. The blue triangles represent type II CTCs. ( B) Average CTCs based on the cluster analysis of totaling 489 CTCs from the NCs ($$n = 236$$) and before-CPAP-treatment OSA ($$n = 253$$) groups. ( C) The proportions of the type I and type II CTCs in NCs ($$n = 236$$) and before-CPAP-treatment OSA ($$n = 253$$) groups. ( D) The numbers of type I CTCs for each subject of NCs ($$n = 236$$) and before-CPAP-treatment OSA ($$n = 253$$) groups. ( E) The average CTCs of the NCs and before-CPAP-treatment OSA groups. ( F) ROC curves of peak concentration values wash-in rate values, and wash-out rate values of all CTCs in frontal cortex for distinguishing type II CTCs ($$n = 314$$) from type I CTCs ($$n = 175$$). ( G) Comparison of peak concentration values wash-in rate values, and wash-out rate values of the type I CTCs in the frontal cortex between NCs ($$n = 70$$) and before-CPAP-treatment OSA ($$n = 105$$) groups (Mann–Whitney U-test). ( H) Comparison of peak concentration values, wash-in rate values, and wash-out rate values of the type II CTCs in the frontal cortex between NCs ($$n = 166$$) and before-CPAP-treatment OSA ($$n = 148$$) groups (Mann–Whitney U-test). ( I) K-means cluster outcome of all CTCs in the bilateral frontal cortex of the NCs ($$n = 236$$) and after-CPAP-treatment OSA ($$n = 253$$) groups. The red drops represent type I CTCs. The blue triangles represent type II CTCs. ( J) Average CTCs based on the cluster analysis of totaling 489 CTCs from the NCs ($$n = 236$$) and after-CPAP-treatment OSA ($$n = 253$$) groups. ( K) The proportions of the type I and type II CTCs in NCs ($$n = 236$$) and after-CPAP-treatment OSA ($$n = 253$$) groups. ( L) The numbers of type I CTCs for each subject of NCs ($$n = 236$$) and after-CPAP-treatment OSA ($$n = 253$$) groups. ( M) The average CTCs of the NCs and after-CPAP-treatment OSA groups. ( N) ROC curves of peak concentration values wash-in rate values, and wash-out rate values of all CTCs in frontal cortex for distinguishing type II CTCs ($$n = 314$$) from type I CTCs ($$n = 175$$). ( O) Comparison of peak concentration values wash-in rate values, and wash-out rate values of the type I CTCs in the frontal cortex between NCs ($$n = 70$$) and after-CPAP-treatment OSA ($$n = 105$$) groups (Mann–Whitney U-test). ( P) Comparison of peak concentration values wash-in rate values, and wash-out rate values of the type II CTCs in the frontal cortex between NCs ($$n = 166$$) and after-CPAP-treatment OSA ($$n = 148$$) groups (Mann–Whitney U-test) The OSA group ($43.84\%$) had a higher percentage of type I CTCs than the NCs group ($35.66\%$) (Fig. 3C). In addition, the OSA group (each patient had an average of 4.923 CTCs) had considerably more type I CTCs than the NCs group (each patient had an average of 3.833 CTCs) (Fig. 3D). Furthermore, Fig. 3E depicted the average curves formed by all type I and type II CTCs in the frontal cortex in the NCs and OSA groups. The statistical results indicated that the peak concentration values of all CTCs in the frontal cortex had adequate accuracy for distinguishing type II CTCs from type I CTCs (sensitivity of $98.73\%$, specificity of $98.86\%$, AUROC of 0.9993, $P \leq 0.0144$) (Fig. 3F, Supplementary Table 4). In addition, the wash-in rate values of all CTCs in the frontal cortex showed desirable accuracy for distinguishing type II CTCs from type I CTCs (sensitivity of $81.21\%$, specificity of $84.57\%$, AUROC of 08,761, $P \leq 0.0036$) (Fig. 3F, Supplementary Table 4). The wash-out rate values of all CTCs in the frontal cortex had appropriate accuracy for distinguishing type II CTCs from type I CTCs (sensitivity of $76.11\%$, specificity of $80.00\%$, AUROC of 0.8417, $P \leq 0.0004$) (Fig. 3F, Supplementary Table 4). Moreover, the OSA group had considerably higher peak concentration and lower wash-out rate values of type I CTCs than the NCs group (Fig. 3G). Besides, there were no differences of the peak concentration, the wash-in rate, and the wash-out rate values of type II CTCs between the OSA group and the NCs group (Fig. 3H). Next, correlation analysis was performed using Spearman correlation. The correlation analysis indicated that there were strongly negative correlations among the wash-out rate values of type I CTCs of frontal cortex, AHI and ODI in OSA patients (Fig. 4B). Moreover, the MMSE scores and the MoCA scores of OSA patients were moderately positively correlated with the wash-out rate values of type I CTCs, which demonstrated that the cognitive impairment in OSA patients was associated with the dysfunction of glymphatic drainage. ( Fig. 4B).Fig. 4Spearman correlations between imaging parameters and clinical manifestation. ( A) Heatmap of Spearman correlations among morphological changes of PVSs, ventricle enlargement, the AHI, the ODI, the MMSE scores, the MoCA scores, the PSQI scores and the ESS scores. The circle size and color intensity represent the magnitude of correlation. ( B) Heatmap of Spearman correlations among wash-out rate values of type I CTCs of frontal cortex, the AHI, the ODI, the MMSE scores and the MoCA scores. The circle size and color intensity represent the magnitude of correlation ## Improvement of glymphatic drainage in OSA patients after CPAP treatment It has been indicated that treatment with CPAP could improve cognitive functions in OSA patients [17], however, the exact mechanism of this improvement is unknown. Using the k-means cluster, we identified total 184 CTCs in the PVSs of the frontal cortex in OSA patients and the NCs after 1 month CPAP treatment (Fig. 3I). The average curves generated by all the CTCs of type I and type II were shown in Fig. 3J. Type I CTCs were found at a larger proportion in the OSA group ($46.08\%$) than in the NCs group ($29.51\%$) after CPAP treatment (Fig. 3K). Furthermore, type I CTCs were significantly higher in OSA group (each patient had an average of 4.273 CTCs) than in NCs group (each patient had an average of 2.769 CTCs) after CPAP treatment (Fig. 3L). Additionally, Fig. 3M showed the average curves generated by all the CTCs of type I and type II in the two groups. According to the statistics, the peak concentration values of all CTCs in the basal ganglia showed adequate accuracy for distinguishing type II CTCs from type I CTCs (sensitivity of $95.67\%$, specificity of $98.68\%$, AUROC of 0.9969, $P \leq 0.0321$) (Fig. 3N, Supplementary Table 4). In addition, the wash-in rate values of all CTCs in the basal ganglia had desirable accuracy for distinguishing type II CTCs from type I CTCs (sensitivity of $76.17\%$, specificity of $90.79\%$, AUROC of 0.8889, $P \leq 0.0086$) (Fig. 3N, Supplementary Table 4). The wash-out rate values of all CTCs in the basal ganglia also had appropriate accuracy for distinguishing type II CTCs from type I CTCs (sensitivity of $81.23\%$, specificity of $70.39\%$, AUROC of 0.8408, $P \leq 0.0020$) (Fig. 3N, Supplementary Table 4). Moreover, no significant differences between the two groups were detected in any of the type I CTC parameters (peak concentration, wash-in rate or wash-out rate) (Fig. 3O). However, the OSA patients had considerably higher peak concentration values of type II CTCs than those of NCs group after CPAP treatment, while there were no significant differences in the wash-in rate or wash-out rate values between the OSA and NCs groups (Fig. 3P). ## Discussion The glymphatic drainage system is a key waste clearance network made up of normally functioning PVSs, which provides a fluid channel for metabolites and neurotoxic waste products to be removed from the brain [1, 25, 26]. Previous studies put forward that glymphatic system dysfunction in OSA evidenced by DTI-ALPS [27, 28]. Unfortunately, there are no methods available to directly quantitatively assess the fluid dynamics of the glymphatic drainage system, including the inflow and outflow of brain interstitial fluid. In the present study, using DCE-MRI, the CTCs of PVSs were retrieved to quantify the perfusion and drainage of gadobutrol in PVSs, which exactly represented the glymphatic drainage function. Gadobutrol from arterioles would drain via PASs to the brain parenchyma, whereas gadobutrol from venules would drain through PVESs to the subarachnoid space. The wash-in rate and peak concentration values were utilized as CTC characteristic parameters, which demonstrated the physiological differences between PASs and PVESs. Moreover, the wash-out rate values of the PASs represented the drainage from the PASs to brain tissue. Besides, the type II CTCs ware detected in both NCs and OSA groups, which suggested that the PVESs could be found in normal aged individuals [29]. Our findings indicated that the glymphatic drainage system was markedly impaired in OSA patients. Furthermore, we tried to identify the improvement of glymphatic drainage function in OSA patients after 1 month CPAP treatment. We considered that the higher peak concentration value of OSA patients before CPAP treatment may due to the hypertension of the patients, and the slower wash-out rate value of OSA patients before CPAP treatment proved the more serious impairment of glymphatic drainage. After CPAP treatment, there were no differences between the OSA group and the NCs group in the peak concentration value and the wash-out rate value. Unfortunately, one limitation of these methods is that how to consider the CPAP treatment as adequate to improve the glymphatic drainage system. However, these limitations notwithstanding, this study does suggest that the CPAP treatment could improve the fluid dynamics of the glymphatic drainage system and thus improve the clinical symptoms and the cognitive function in OSA patients. Additionally, we found that morphological changes of PVSs occurred in OSA group as compared to the NCs group. What’s more, our data uncovered that the enlarged PVSs could also be found more significantly in OSA patients with severe hypoxemia than mild hypoxemia. In conjunction with the previous studies that the glial metabolism and the vascular unit of the glymphatic drainage system are altered in hypoxic conditions [30], we speculate that the hypoxic condition during sleep may also cause enlarged PVSs and glymphatic drainage system dysfunction in patients of OSA. We further explored the potential relationship between the glymphatic drainage dysfunction and the cognitive decline in OSA patients. The results suggested the potential causal association of the outflow dysfunction of glymphatic drainage system and the cognitive decline. Previous research has shown that sleep fragmentation of OSA impede sleep-associated glymphatic circulation of CSF-ISF exchange, which may lead to hydrocephalus [31]. Ventricle enlargement was detected in the OSA group in this study, which was more noticeable in severe OSA or OSA with severe hypoxemia. These findings further proved that glymphatic drainage system dysfunction occurred in patients of OSA, especially in patients with severe OSA or hypoxemia. It was previously reported that dilated PVSs is linked to an increased risk of dementia [5, 32]. Our results showed that the MMSE score and the MoCA score were negatively associated with the enlarged PVSs, the expanded ventricles, and the AHI and the ODI scores, which further suggested that patients with OSA have a higher risk of cognitive impairment due to the glymphatic drainage system dysfunction. ## Conclusion We found that glymphatic drainage system dysfunction occurred in OSA patients, including enlarged PVSs, ventricle expansion, and abnormal fluid dynamics of PVSs, which might be important imaging markers. In addition, the correlation between glymphatic drainage system dysfunction and hypoxia severity or cognitive decline, suggesting that impaired glymphatic drainage might contribute to cognitive decline via OSA-induced chronic intermittent hypoxia in OSA patients. Moreover, the fluid dynamics of the glymphatic drainage system was improved by CPAP treatment, indicating the novel therapeutic mechanism of CPAP treatment for cognitive function in OSA patients. ## Supplementary Information Below is the link to the electronic supplementary material. Supplementary file1 (DOCX 788 KB) ## References 1. Rasmussen MK, Mestre H, Nedergaard M. The glymphatic pathway in neurological disorders. Lancet Neurol (2018.0) 17 1016-1024. DOI: 10.1016/S1474-4422(18)30318-1 2. Iliff JJ, Wang M, Liao Y. A paravascular pathway facilitates CSF flow through the brain parenchyma and the clearance of interstitial solutes, including amyloid beta. Sci Transl Med (2012.0) 4 147. DOI: 10.1126/scitranslmed.3003748 3. Bohr T, Hjorth PG, Holst SC. The glymphatic system: Current understanding and modeling. Science (2022.0) 25 104987. DOI: 10.1016/j.isci.2022.104987 4. Rasmussen MK, Mestre H, Nedergaard M. Fluid transport in the brain. Physiol Rev (2022.0) 102 1025-1151. DOI: 10.1152/physrev.00031.2020 5. Paradise M, Crawford JD, Lam BCP. Association of dilated perivascular spaces with cognitive decline and incident dementia. Neurology (2021.0) 96 e1501-e1511. DOI: 10.1212/WNL.0000000000011537 6. Li Y, Rusinek H, Butler T. Decreased CSF clearance and increased brain amyloid in Alzheimer's disease. Fluids Barriers CNS (2022.0) 19 21. DOI: 10.1186/s12987-022-00318-y 7. Han F, Brown G, Zhu Y. Decoupling of global brain activity and cerebrospinal fluid flow in Parkinson's disease cognitive decline. Mov Disord (2021.0) 36 2066-2076. DOI: 10.1002/mds.28643 8. Young T, Palta M, Dempsey J. The occurrence of sleep-disordered breathing among middle-aged adults. N Engl J Med (1993.0) 328 1230-1235. DOI: 10.1056/nejm199304293281704 9. Shieu M, Zaheed A, Shannon C. Positive airway pressure and cognitive disorders in adults with obstructive sleep apnea: a systematic review of the literature. Neurology (2022.0). DOI: 10.1212/wnl.0000000000200383 10. Carnicelli L, Maestri M, Di Coscio E. A longitudinal study of polysomnographic variables in patients with mild cognitive impairment converting to Alzheimer's disease. J Sleep Res (2019.0) 28 e12821. DOI: 10.1111/jsr.12821 11. Yaffe K, Laffan A, Harrison S. Sleep-disordered breathing, hypoxia, and risk of mild cognitive impairment and dementia in older women. JAMA (2011.0) 306 613-619. DOI: 10.1001/jama.2011.1115 12. Blackwell T, Yaffe K, Laffan A. Associations between sleep-disordered breathing, nocturnal hypoxemia, and subsequent cognitive decline in older community-dwelling men: the osteoporotic fractures in men sleep study. J Am Geriatr Soc (2015.0) 63 453-461. DOI: 10.1111/jgs.13321 13. Bubu OM, Pirraglia E, Andrade AG. Obstructive sleep apnea and longitudinal Alzheimer's disease biomarker changes. Sleep (2019.0) 42 2. DOI: 10.1093/sleep/zsz048 14. Leng Y, Mcevoy CT, Allen IE. Association of sleep-disordered breathing with cognitive function and risk of cognitive impairment: a systematic review and meta-analysis. JAMA Neurol (2017.0) 74 1237-1245. DOI: 10.1001/jamaneurol.2017.2180 15. Liguori C, Cremascoli R, Maestri M. Obstructive sleep apnea syndrome and Alzheimer's disease pathology: may continuous positive airway pressure treatment delay cognitive deterioration?. Sleep Breath (2021.0) 25 2135-2139. DOI: 10.1007/s11325-021-02320-4 16. Richards KC, Gooneratne N, Dicicco B. CPAP adherence may slow 1-year cognitive decline in older adults with mild cognitive impairment and apnea. J Am Geriatr Soc (2019.0) 67 558-564. DOI: 10.1111/jgs.15758 17. Rosenzweig I, Glasser M, Crum WR. Changes in neurocognitive architecture in patients with obstructive sleep apnea treated with continuous positive airway pressure. EBioMedicine (2016.0) 7 221-229. DOI: 10.1016/j.ebiom.2016.03.020 18. Osorio RS, Gumb T, Pirraglia E. Sleep-disordered breathing advances cognitive decline in the elderly. Neurology (2015.0) 84 1964-1971. DOI: 10.1212/WNL.0000000000001566 19. Anonymous EEG. arousals: scoring rules and examples: a preliminary report from the sleep disorders atlas task force of the American sleep disorders association. Sleep (1992.0) 15 173-184. PMID: 11032543 20. 20.Anonymous. Sleep-related breathing disorders in adults: recommendations for syndrome definition and measurement techniques in clinical research. The Report of an American Academy of Sleep Medicine Task Force. Sleep. 1999;22:667–89. 21. Xie L, Wu Q, Hu W. Performance of brief ICF-sleep disorders and obesity core set in obstructive sleep apnea patients. Respir Res (2020.0) 21 156. DOI: 10.1186/s12931-020-01404-1 22. Ding X, Wang X, Xia D. Impaired meningeal lymphatic drainage in patients with idiopathic Parkinson's disease. Nat Med (2021.0) 27 411-418. DOI: 10.1038/s41591-020-01198-1 23. Wardlaw J, Smith E, Biessels G. Neuroimaging standards for research into small vessel disease and its contribution to ageing and neurodegeneration. Lancet Neurol (2013.0) 12 822-838. DOI: 10.1016/s1474-4422(13)70124-8 24. Charrad M, Ghazzali N, Boiteau V. NbClust: an R package for determining the relevant number of clusters in a data set. J Stat Softw (2014.0) 61 1-36. DOI: 10.18637/jss.v061.i06 25. Cserr H, Cooper D, Suri P. Efflux of radiolabeled polyethylene glycols and albumin from rat brain. Am J Physiol (1981.0) 240 F319-F328. DOI: 10.1152/ajprenal.1981.240.4.F319 26. Wardlaw J, Benveniste H, Nedergaard M. Perivascular spaces in the brain: anatomy, physiology and pathology. Nat Rev Neurol (2020.0) 16 137-153. DOI: 10.1038/s41582-020-0312-z 27. Lee H, Lee D, Shin K. Glymphatic system dysfunction in obstructive sleep apnea evidenced by DTI-ALPS. Sleep Med (2022.0) 89 176-181. DOI: 10.1016/j.sleep.2021.12.013 28. Roy B, Nunez A, Aysola RS. Impaired glymphatic system actions in obstructive sleep apnea adults. Front Neurosci (2022.0) 16 884234. DOI: 10.3389/fnins.2022.884234 29. Wu C, Lirng J, Ling Y. Noninvasive characterization of human glymphatics and meningeal lymphatics in an in vivo model of blood-brain barrier leakage. Ann Neurol (2021.0) 89 111-124. DOI: 10.1002/ana.25928 30. Lana D, Ugolini F, Giovannini M. An overview on the differential interplay among neurons-astrocytes-microglia in CA1 and CA3 hippocampus in hypoxia/ischemia. Front Cell Neurosci (2020.0) 14 585833. DOI: 10.3389/fncel.2020.585833 31. Román G, Jackson R, Fung S. Sleep-disordered breathing and idiopathic normal-pressure hydrocephalus: recent pathophysiological advances. Curr Neurol Neurosci Rep (2019.0) 19 39. DOI: 10.1007/s11910-019-0952-9 32. Brown R, Benveniste H, Black SE. Understanding the role of the perivascular space in cerebral small vessel disease. Cardiovasc Res (2018.0) 114 1462-1473. DOI: 10.1093/cvr/cvy113
--- title: Adipokines in multiple sclerosis patients are related to clinical and radiological measures authors: - Floor C. Loonstra - Kim F. Falize - Lodewijk R. J. de Ruiter - Menno M. Schoonheim - Eva M. M. Strijbis - Joep Killestein - Helga E. de Vries - Bernard M. J. Uitdehaag - Merel Rijnsburger journal: Journal of Neurology year: 2022 pmcid: PMC10025234 doi: 10.1007/s00415-022-11519-8 license: CC BY 4.0 --- # Adipokines in multiple sclerosis patients are related to clinical and radiological measures ## Abstract ### Background An imbalance of adipokines, hormones secreted by white adipose tissue, is suggested to play a role in the immunopathology of multiple sclerosis (MS). In people with MS (PwMS) of the same age, we aimed to determine whether the adipokines adiponectin, leptin, and resistin are associated with MS disease severity. Furthermore, we aimed to investigate whether these adipokines mediate the association between body mass index (BMI) and MS disease severity. ### Methods Adiponectin, resistin, and leptin were determined in serum using ELISA. 288 PwMS and 125 healthy controls (HC) were included from the Project Y cohort, a population-based cross-sectional study of people with MS born in the Netherlands in 1966, and age and sex-matched HC. Adipokine levels and BMI were related to demographic, clinical and disability measures, and MRI-based brain volumes. ### Results Adiponectin levels were 1.2 fold higher in PwMS vs. HC, especially in secondary progressive MS. Furthermore, we found a sex-specific increase in adiponectin levels in primary progressive (PP) male patients compared to male controls. Leptin and resistin levels did not differ between PwMS and HC, however, leptin levels were associated with higher disability (EDSS) and resistin strongly related to brain volumes in progressive patients, especially in several grey matter regions in PPMS. Importantly, correction for BMI did not significantly change the results. ### Conclusion In PwMS of the same age, we found associations between adipokines (adiponectin, leptin, and resistin) and a range of clinical and radiological metrics. These associations were independent of BMI, indicating distinct mechanisms. ## Introduction Obesity during childhood or adolescence is associated with a ~ 2.5-fold increased risk of developing multiple sclerosis (MS), suggesting that accompanying metabolic and immunological alterations promote disease pathogenesis [1, 2]. Additionally, increased body mass index (BMI) and obesity have been associated with higher disability, increased pro-inflammatory cytokines in relapsing remitting (RR) MS, and reduced gray matter volumes on MRI [3, 4]. However, the underlying mechanisms of these associations are still largely unknown. Enhanced body weight (BMI > 25) is associated with low-grade inflammation, which is driven by the (altered) release of adipokines in the circulation. Recent data suggest that an imbalance of pro- and anti-inflammatory adipokines, a class of cytokines released by the white adipose tissue (WAT), contributes to immune-pathological processes related to MS and may therefore represents one of the possible links between increased BMI and MS disease severity [5]. Adipokines are key players in the maintenance of energy balance as well as immune homeostasis and especially adiponectin, leptin and resistin have been described to play a pivotal role in both processes [5]. Adiponectin, the most abundant adipokine found in human plasma, exerts anti-inflammatory effects by suppressing the production of pro-inflammatory cytokines. While it suppresses activation of T and B lymphocytes by stimulating secretion of anti-inflammatory IL-10[6, 7], adiponectin may also exert pro-inflammatory effects [8, 9]. Leptin possesses various pro-inflammatory roles in endothelial cells and macrophages by producing pro-inflammatory molecules such as TNF-α, IL-6 and CXCL10 [10]. Moreover, leptin increases type 1 T helper cell activity while it decreases the proliferation of regulatory T cells [10]. While adiponectin and leptin are predominantly produced by the WAT, resistin is believed to be mainly produced in peripheral blood mononuclear cells (PBMCs) and increases when stimulated with pro-inflammatory cytokines. In turn, resistin treatment stimulates the release of other pro-inflammatory cytokines from human PBMC’s [11, 12]. Adipokines may predict progression in other chronic inflammatory diseases[13], but only few reports address the involvement of adipokines in MS. Adipokines interact with the blood–brain barrier (BBB) and may either cross the BBB or affect BBB integrity. Consequently, abnormal adipokine secretion could exert potent CNS effects through oxidative stress and inflammation [14]. Independent studies show that leptin, resistin, and adiponectin levels are increased in people with MS (PwMS). However, results are contrasting and a direct association with MS phenotype is unknown.[15–18] Additionally, studies were confounded by unknown treatment status, sex and most importantly, differences in age [5, 19, 20]. This study, therefore, aimed to investigate the association between adipokines and disease severity in the different MS phenotypes. To remove potential bias effects of age, this was assessed in a cohort of PwMS and healthy controls (HC) of the same age (Project Y) [21]. We explored [1] differences in adiponectin, leptin and resistin levels in MS and how these relate to clinical, disability, and MRI measures, adjusted and unadjusted for BMI and [2] how BMI relates to clinical, disability and MRI measures. ## Cohort Patients were selected from the cohort project Y, a population-based cross sectional birth year cohort aimed to include all PwMS (as defined by the 2017 McDonald Criteria) [22] born in the Netherlands in 1966 and HC’s born in 1965–1967. Details on this cohort have been described previously [21]. All participants with available plasma samples were included for the aim of this study. All participants gave written informed consent and this study was approved by the Medical Ethical Committee of the Amsterdam UMC, location VUmc. ## Adipokine measurements Serum was collected via standard vena puncture. Adipokine quantification was performed using ELISA. Samples were diluted 1:5000 to 1:10,000 for adiponectin, 1:40 for leptin (Human standard ABTS development kit, PeproTech, London, UK) and 1:20 for resistin (Human standard ABTS development kit, PeproTech) and randomized between plates. The samples of each participant were analyzed in duplo within one run and each plate contained a sample from a control donor, to control for inter-assay variability. Inter-assay and intra-assay variability was $21\%$ and $4.3\%$ for resistin, $29\%$ and $21\%$ for leptin and $12\%$ and $11\%$ for adiponectin, respectively. The personnel performing the analyses was blinded for the clinical data. Leptin levels below limit of detection (78.10 pg/mL) were imputated by assigning random numbers between 0 and 78.10. ## Neuroimaging 228 patients and 113 HC underwent 3 T MR imaging of brain and spinal cord, which is described in detail elsewhere [21]. In short, the MRI protocol included cerebral high resolution anisotropic sagittal three dimensional (3D)-T1 sagittal slices and 3D- Fluid Attenuation Inversion Recovery (3D-FLAIR) sequences. Lesions were automatically segmented on FLAIR for lesion volumes (LV) and filled on T1 images. Normalized total brain volume (NBV), normalized cortical gray matter volume (NCGMV) and normalized white matter volume (NWMV) were calculated using SIENAX, providing normalization for head size. Total normalized deep grey matter (NDGMV) and thalamic volume (NThalV) were calculated using FIRST in FSL6. The Harvard_Oxford atlas was used to measure cerebellar grey matter volume (NCbV) using a previously described procedure [23]. Cerebral 3DT1 images were used to measure mean upper cervical cord area (MUCCA) using SCT, as previously described [24]. ## Clinical assessment A comprehensive interview was conducted and included date of onset, MS phenotype, exacerbations and disease progression, use of disease modifying therapies (DMT), general medical history and use of other medication. For the purpose of this study, we used information on diabetes mellitus (yes/no), hyperlipidemia and statin use. Expanded Disability Status Scale (EDSS) scores were used to assess overall MS-related disability. Upper and lower extremity function was measured using the nine-hole peg test (9HPT) and timed 25 foot walking test (T25FWT), respectively. The T25FWT was completed twice and the 9HPT was performed twice each hand; the average of both trials was used. BMI was obtained by dividing weight in kilograms by length in meters squared. ## Statistical analyses Statistical analysis was performed using SPSS (Version 26.0, IBM, USA). Histograms and the Shapiro–Wilk test were used to test for normality of distribution. Adiponectin, leptin, resistin and LV were log-transformed to achieve a normal distribution. Independent T-tests, ANOVA and non-parametric tests were used to compare characteristics between PwMS and HC and between MS subtypes. A p value < 0.05 was considered statistically significant; Bonferroni corrections were applied to adjust for multiple comparisons. We hypothesized that if adipokines would serve as mediators in the relation between BMI and MS disease severity [1] the analysis would not yield significant relations between adipokines and MS while adjusting for BMI [2] adipokines would be significantly related with BMI [3] both adipokines and BMI would yield similar significant associations with MS disease severity and [4] the relation between adipokines and MS disease severity would significantly change while excluding BMI as covariate. Analyses thus consisted of 4 steps. ## Group differences in adipokine levels General linear models (GLM) were performed to assess differences in adipokine levels between patients and HC, between MS subtypes, EDSS groups, patients using DMT and between RR onset patients with a relapse within 3 months prior to sampling and patients without a relapse. When comparing patients vs. HC, BMI, sex, type 1 or type 2 diabetes mellitus (yes/no), statin use (yes/no) and hyperlipidemia (yes/no) were used as covariates based on their known confounding effects. If comparing patients groups based on clinical parameters, the following covariates were used: BMI, sex, disease duration and DMT use (duration and current DMT yes/no), diabetes mellitus, statin use and hyperlipidemia. No significant changes in adipokine levels were found between smokers and non-smokers and analyses were therefore not corrected for smoking. ## Univariate regressions Relations of adipokines with disability (EDSS, 9HPT, T25FWT) and volumetric MRI measures were assessed using univariate linear regression analysis, correcting for BMI, sex, disease duration, DMT use (duration and current DMT yes/no), onset type (RRMS vs. progressive), diabetes mellitus and statin use. Regression analyses were stratified by sex and MS subtype. In each strata, effect modification by sex and onset type was assessed. Cases were classified as outliers if Cook’s distance was ≥ 1.0 and/or residuals were three or more standard deviations from the mean and/or based on visual inspection. Patients unable to perform the T25FWT or 9HPT were excluded from regression analyses. Relations of BMI with disability and volumetric MRI measures were analyzed using linear regressions, stratified by sex and MS subtype and adjusted for disease duration, DMT use (duration and current DMT yes/no), diabetes mellitus and statin use wherever appropriate. When stratified by sex, analysis were also corrected for onset type (RRMS vs. progressive) and vice versa. For similar significant associations between adipokines and MRI volumes on the one hand and associations between BMI and MRI volumes on the other hand, we explored whether BMI and adipokines were independently associated with MRI volumes using linear regression analyses including both the respective adipokine and BMI in the same model. Sex, disease duration, DMT use (duration and current DMT yes/no), onset type (RRMS vs. progressive), diabetes mellitus and statin use were also entered as covariates. Lastly, relations of BMI and adipokines with MRI volumes in HC were assessed using linear regressions. To assess whether BMI adjustment significantly changed the relation between adipokines and MS disease severity, all analyses of the first step were repeated with similar covariates while excluding BMI. ## Correlation analysis For each sex-specific subtype, Spearman’s correlation was used for correlation analyses between adipokines and BMI. ## Group differences in MRI volumes A GLM was used to assess differences in MRI volumes between BMI categories “lean” (BMI < 25), “overweight” (BMI 25–30) and “obese” (BMI ≥ 30). Patients were stratified by sex and adjusted for disease duration, DMT use (duration and current DMT yes/no), diabetes mellitus and statin use. ## General characteristics 288 PwMS (RRMS: 170; SPMS: 80; PPMS: 37) and 125 HC of the Project Y cohort were included. Table 1 depicts the demographic, clinical and MRI characteristics for all patients and HC.Table 1General characteristics of people with multiple sclerosis and healthy controlsHealthy controls ($$n = 125$$)All PwMS ($$n = 288$$)RRMS ($$n = 170$$)SPMS ($$n = 80$$)PPMS ($$n = 37$$)Age, years (SD)52.9 ± 1.252.9 ± 0.952.9 ± 0.953.1 ± 0.953.1 ± 0.9Female (%)92 ($74\%$)207 ($72\%$)139 ($82\%$)†48 ($60\%$)19 ($51\%$)‡BMI (SD)25.6 ± 3.726.1 ± 4.926.6 ± 5.225.5 ± 4.425.4 ± 3.7Adiponectin (ng/mL), median (IQR)10,591.7 (8416.6)12,455.9 (7922.4)12,293.4 (7365.7)13,289.5 (8329.4)12,681.1 (8071.0)Leptin (pg/mL), median (IQR)31,856.7 (50,711.1)34,987.1 (55,344.0)42,361.5 (61,425.3)52,378.2 (52,585.1)27,216.6 (45,793.4)Resistin (pg/mL), median (IQR)4789.2 (2666.0)4618.9 (2436.4)4607.5 (2670.4)†4687. ( 2859.4)4562.3 (1791.4)‡EDSS, median (IQR)–3.75 (2.0)3.0 (2.0)†6.0 (2.5)4.0 (2.5)‡Disease duration since symptom onset (IQR)15.3 (15.9)14.2 (15.2)†20.7 (11.5)§8.1 (9.0)‡Current DMT, n (%)–134 ($47\%$)94 ($55\%$)33 ($41\%$)8 ($22\%$)‡ First line DMT90 ($31\%$)68 ($40\%$)†19 ($24\%$)§2 ($5\%$)‡ Second line DMT46 ($16\%$)26 ($15\%$)14 ($18\%$)6 ($16\%$)DMT total duration (IQR)–6.1 (9.0)6.4 (9.1)7.8 (8.8) §1.5 (2.7)‡Number of relapses, n (%)– < 5 relapses206 ($72\%$)122 ($72\%$)46 ($58\%$)§37 ($100\%$)‡ > 5 relapses82 ($29\%$)48 ($28\%$)34 ($43\%$)§–‡Statine use, n (%)1 ($1\%$)12 ($4\%$)4 ($2\%$)6 ($8\%$)2 ($5\%$)Diabetes mellitus, n (%)1 ($1\%$)8 ($3\%$)6 ($5\%$)1 ($1\%$)1 ($3\%$)MRI volumes($$n = 113$$)($$n = 230$$)($$n = 144$$)($$n = 54$$)($$n = 32$$)Normalized brain volume, L (mean, SD)1.54 ± 0.0771.48 ± 0.0781.49 ± 0.0771.48 ± 0.0671.48 ± 0.097Normalized cortical gray matter volume, L (mean, SD)0.79 ± 0.0510.76 ± 0.0520.76 ± 0.0520.74 ± 0.0500.75 ± 0.055Normalized deep gray matter volume, mL (mean, SD)63.53 ± 4.8959.01 ± 5.4359.58 ± 4.86†57.45 ± 6.6559.03 ± 5.24Thalamic volume, mL (mean, SD)21.36 ± 1.6719.56 ± 2.0219.76 ± 1.8719.01 ± 2.3719.53 ± 1.93Normalized white matter volume, L (mean, SD)0.71 ± 0.0430.69 ± 0.0430.69 ± 0.0410.70 ± 0.400.69 ± 0.055Normalized cerebellar gray matter volume, mL (mean, SD)108.40 ± 14.13101.88 ± 13.64102.82 ± 12.4399.80 ± 16.37101.21 ± 13.87Mean upper cervical cord area, mm2(mean, SD)72.9 ± 7.9867.69 ± 8.6469.57 ± 7.8463.53 ± 9.3266.24 ± 8.42Lesion volume, mL (median, IQR)2.79 (2.43)10.79 (14.25)9.54 (10.43)14.01 (15.46)13.70 (18.36)BMI body mass index, DMT disease modifying therapy, EDSS expanded disability status scale, SD standard deviation, PPMS primary progressive multiple sclerosis, RRMS relapsing remitting multiple sclerosis, SPMS secondary progressive multiple sclerosisIndicates a significant difference ($p \leq 0.05$) between all PwMS and healthy controls (uncorrected)†Indicates a significant difference ($p \leq 0.05$) between RRMS and SPMS (uncorrected)‡Indicates a significant difference ($p \leq 0.05$) between RRMS and PPMS (uncorrected)§Indicates a significant difference ($p \leq 0.05$) between SPMS and PPMS (uncorrected) ## Adipokine levels in PwMS and healthy controls Adiponectin concentrations were higher in PwMS compared to HC ($$p \leq 0.004$$), whereas resistin ($$p \leq 0.088$$) and leptin ($$p \leq 0.945$$) levels did not differ between PwMS and HC’s (Fig. 1). Male patients had higher adiponectin concentrations compared to male HC ($$p \leq 0.023$$), however no differences were found between female PwMS and female HC ($$p \leq 0.055$$) (Fig. 1). Both adiponectin and leptin levels were higher in female PwMS compared to male PwMS and in female HC compared to male HC (all $p \leq 0.001$) (Fig. 1). After BMI stratification, only adiponectin levels were higher in patients with BMI ≥ 25 ($$n = 153$$) compared to HC with BMI ≥ 25 ($$n = 75$$) ($$p \leq 0.002$$), whereas all adipokine levels in patients with BMI < 25 did not differ from HC’s with BMI < 25.Fig. 1Levels of adiponectin (a), resistin (b) and leptin (c) in healthy controls versus people with multiple sclerosis (PwMS) and levels of adiponectin (d), resistin (e) and leptin (f) in PwMS stratified by sex. Each dot in the scatter box-plot represents a sample. p values were calculated with a general linear model, adjusted for sex, BMI, diabetes mellitus (yes/no), statin use (yes/no) and hyperlipidemia (yes/no). F female, HC healthy controls, M male, MS multiple sclerosis Stratification by MS subtype showed significantly higher adiponectin levels in SPMS compared to HC ($$p \leq 0.012$$). When stratified by sex, only in male PPMS higher adiponectin concentrations were observed compared to male HC ($$p \leq 0.025$$) and a trend towards increased adiponectin levels in female SPMS compared to HC ($$p \leq 0.060$$) (Fig. 2).Fig. 2Adiponectin levels in healthy controls and people with multiple sclerosis, stratified by sex and MS subtype. Each dot in the scatter box-plot represents a sample. p values were calculated with a general linear model (adjusted for BMI, diabetes status, statin use and hyperlipidemia) followed by post-hoc analyses, Bonferroni corrected. HC healthy control, PPMS primary progressive multiple sclerosis, RRMS relapsing remitting multiple sclerosis, SPMS secondary progressive multiple sclerosis ## Adipokine levels across PwMS with disease modifying therapy Resistin levels were lower in PwMS using teriflunomide ($$n = 15$$) compared to PwMS using glatiramer acetate ($$n = 22$$; $$p \leq 0.020$$), dimethyl fumarate ($$n = 36$$; $$p \leq 0.032$$) and ocrelizumab ($$n = 14$$; $$p \leq 0.042$$). No differences were detected in adiponectin and leptin concentrations between all DMT’s. Moreover, adipokine levels did not differ between patients without DMT, patients with first-line DMT (interferon-beta, dimethyl fumarate, glatiramer acetate and teriflunomide) and patients with second-line DMT (ocrelizumab, natalizumab, fingolimod). ## Relation of adipokine levels and clinical measures Higher adiponectin levels were related to a longer disease duration in female progressive MS (β = 0.349, $$p \leq 0.034$$), but not in female SPMS or PPMS. Leptin levels inversely correlated with disease duration (β = -0.212, $$p \leq 0.020$$) in female SPMS, while resistin did not correlate with disease duration. In addition, none of the adipokines in relapse onset patients with a relapse within 3 months prior to sampling ($$n = 8$$) did significantly differ from relapse onset patients in remission ($$n = 240$$). Regression analyses of clinical disability scores as dependent variables are depicted in Table 2. Higher leptin levels were associated with higher EDSS in all strata, except for SPMS. In female RRMS, increased leptin levels were associated with higher EDSS (β = 0.260, $$p \leq 0.011$$), but not in male RRMS.Table 2Association between BMI and disability measures (EDSS, 9HPT and T25FWT)—adjusted univariate linear regression analysisEDSSStd. β9HPTStd. βT25FWTStd. βBMI Male patients− 0.1010.0970.077 Female patients− 0.024− 0.0160.107 RRMS− 0.0090.1080.112 SPMS− 0.0060.0150.237 PPMS0.226− 0.012− 0.066Adiponectin Male patients0.070− 0.0030.203 Female patients0.033− 0.0170.039 RRMS− 0.108− 0.064− 0.072 SPMS0.1080.0320.187 PPMS0.0010.0370.038Resistin Male patients0.1270.3370.106 Female patients− 0.128− 0.106− 0.125 RRMS− 0.059− 0.072− 0.073 SPMS0.0710.0310.023 PPMS− 0.015− 0.205− 0.195Leptin Male patients0.2590.1560.030 Female patients0.1710.0420.005 RRMS0.269*0.0480.049 SPMS0.1000.099− 0.243 PPMS0.4360.2060.280Bold values denote statistical significance at the $p \leq 0.05$ levelThe following disease specific and disease modifying factors were included in the univariate linear regression analysis as covariates: disease duration, DMT duration, DMT use (yes/no), BMI, sex, onset type, statin use (yes/no) and diabetes mellitus (yes/no).BMI body mass index, EDSS expanded disability status scale, 9HPT Nine hole peg test, 25FWT 25 foot timed walking test, Std. β standardized beta, PPMS primary progressive multiple sclerosis, RRMS relapsing remitting multiple sclerosisp value < 0.05; p value < 0.01 ## Relation of adipokine levels and radiological measures Adiponectin was significantly related to NBV (β = − 0.316, $$p \leq 0.026$$) and NWMV (β = − 0.533, $$p \leq 0.002$$) in SPMS (Table 3). Lower leptin levels were associated with lower NCGMV in all female patients. Additionally, in RRMS males, leptin was significantly related to LV (β = 0.677, $$p \leq 0.039$$). Resistin levels were negatively associated with NBV, NCGMV, NDGMV and NThalV in PPMS (Fig. 3). In RRMS males, resistin was also inversely related to NDGMV (β = − 0.637, $$p \leq 0.028$$). However in SPMS, lower resistin concentrations were associated with lower thalamic volume (β = 0.268, $$p \leq 0.044$$).Table 3Association between adipokines and MRI volumes: adjusted univariate linear regression analysisNBVStd. βNWMStd. βNCGMVStd. βNDGMVStd. βNThalVStd. βNCbVStd. βMUCCAStd. βLVStd. βAdiponectin Male patients− 0.123− 0.074− 0.128− 0.087− 0.021− 0.0130.1210.258 Female patients0.0290.0330.0100.0660.029− 0.003− 0.039− 0.066 RRMS0.4750.119− 0.0060.0970.0490.008− 0.0080.012 SPMS− 0.361− 0.533*− 0.103− 0.239− 0.229− 0.0870.2270.019 PPMS− 0.093− 0.078− 0.0940.0470.107− 0.031− 0.2360.150Resistin Male patients− 0.052− 0.079− 0.001− 0.174− 0.0260.0950.129− 0.042 Female patients− 0.025− 0.1530.086− 0.009− 0.0130.002− 0.0800.029 RRMS0.020− 0.1600.145− 0.0070.0080.066− 0.072− 0.006 SPMS0.1690.2640.0810.1760.2680.097− 0.109− 0.107 PPMS− 0.408− 0.303− 0.366− 0.630*− 0.557− 0.3300.1180.295Leptin Male patients− 0.177− 0.268− 0.047− 0.080− 0.041− 0.193− 0.122− 0.085 Female patients0.145− 0.0550.272*0.067− 0.0010.118− 0.0180.018 RRMS0.078− 0.1020.195− 0.044− 0.0900.0890.0050.008 SPMS0.3020.0850.2850.3250.2640.1460.187− 0.107 PPMS− 0.029− 0.1690.135− 0.102− 0.139− 0.039− 0.3810.206Bold values denote statistical significance at the $p \leq 0.05$ levelThe following disease specific and disease modifying factors were included in the univariate linear regression analysis as covariates: disease duration, DMT duration, DMT use (yes/no), BMI, sex, onset type, statin use (yes/no) and diabetes mellitus (yes/no)Std. β standardized Beta, PPMS primary progressive multiple sclerosis, RRMS relapsing remitting multiple sclerosis, SPMS secondary progressive MS, LV Lesion volume, MUCCA Mean upper cervical cord area, NBV normalized total brain volume, NCbV cerebellar gray matter volume, NCGMV normalized cortical gray matter volume, NDGMV normalized deep gray matter volume, NThalV normalized thalamic volume, NWMV normalized white matter volumep value < 0.05; p value < 0.01; p value < 0.001Fig. 3Scatterplot of resistin levels and total brain volume (a) deep gray matter volume (b) and thalamic volume (c) in primary progressive multiple sclerosis. Dashed lines indicate the $95\%$ confidence intervals for the regression line. PPMS primary progressive multiple sclerosis ## Relation of BMI with adipokines First, univariate analyses were performed to assess the relation between adipokine levels and BMI (Table 4). Adiponectin negatively correlated with BMI, whereas resistin and leptin correlated positively. Table 4Correlations between adipokines and BMI in different subgroupsHCMS allRRMSSPMSPPMSMaleFemaleMaleFemaleMaleFemaleMaleFemaleMaleFemaleAdiponectin− 0.29− 0.26− 0.10− 0.28*0.10− 0.29− 0.05− 0.31− 0.34− 0.11Leptin0.74*0.580.450.660.310.670.590.570.460.66Resistin0.120.21− 0.010.26*0.080.31− 0.180.260.14− 0.44Bold values denote statistical significance at the $p \leq 0.05$ levelBMI body mass index, HC healthy controls, PPMS primary progressive multiple sclerosis, RRMS relapsing remitting multiple sclerosis, SPMS secondary progressive multiple sclerosisValues are Spearman rho, p value < 0.05; p value < 0.01; p value < 0.001 ## Relation of BMI with clinical and radiological measures In female PPMS, increased BMI was significantly associated with higher EDSS (β = 0.464, $$p \leq 0.032$$). No other relations between BMI and clinical disability measures were observed (Table 2). Next, MRI volumes were compared between BMI categories. Obese female patients (BMI ≥ 30, $$n = 28$$) had significantly lower NCGMV compared to female patients in the lowest BMI group (BMI < 25, $$n = 76$$) ($$p \leq 0.023$$). In addition, overweight male patients (BMI 25–30, $$n = 28$$) had significantly lower NCGMV compared to lean male patients (BMI < 25, $$n = 27$$) ($$p \leq 0.030$$). Regression analyses of volumetric MRI measures are shown in Table 5. BMI was positively associated with NWMV in female and in PPMS, while BMI was inversely related to NCGMV in SPMS. Additional analysis revealed that in female SPMS, higher BMI was associated with higher LV (β = 0.395, $$p \leq 0.040$$), whereas increased BMI was associated with higher NWMV (β = 0.619, $$p \leq 0.030$$) in female PPMS.Table 5Association between BMI and disability measures and MRI volumes—adjusted univariate linear regression analysisEDSSStd. β9HPTStd. βT25FWTStd. βNBVStd. βNWMVStd. βNCGMVStd. βNDGMVStd. βNThalVStd. βNCbVStd. βMUCCAStd. βLVStd. βBMI Male patients− 0.1010.0970.077− 0.0270.184− 0.179− 0.0470.087− 0.1740.106− 0.116 Female patients− 0.024− 0.0160.1070.0080.168− 0.1400.0610.137− 0.115− 0.014− 0.024 RRMS− 0.0090.1080.112− 0.0120.118− 0.1250.0780.118− 0.100− 0.042− 0.046 SPMS− 0.0060.0150.237− 0.1110.205− 0.298− 0.0500.027− 0.2390.0380.279 PPMS0.226− 0.012− 0.0660.3090.4070.1340.1720.526*0.0340.167− 0.358Bold values denote statistical significance at the $p \leq 0.05$ levelThe following disease specific and disease modifying factors were included in the univariate linear regression analysis as covariates: disease duration, DMT duration, DMT use (yes/no), BMI, sex, onset type, statine use (yes/no) and diabetes mellitus (yes/no)EDSS Expanded disability status scale, 9HPT Nine hole peg test, 25FWT 25 Foot timed walking test, Std. β standardized Beta, PPMS primary progressive multiple sclerosis, RRMS relapsing remitting multiple sclerosis, SPMS secondary progressive multiple sclerosis LV Lesion volume, MUCCA Mean upper cervical cord area, NBV normalized total brain volume, NCbV Cerebellar gray matter volume, NCGMV Normalized cortical gray matter volume, NDGMV Normalized deep gray matter volume, NThalV Normalized thalamic volume, NWMV Normalized white matter volumep value < 0.05; p value < 0.01 The only corresponding significant association between BMI and MRI volumes and adipokines and MRI volumes was the association with NThalV in PPMS. Both BMI (β = 0.630, $$p \leq 0.001$$) and resistin (β = − 0.515, $$p \leq 0.002$$) were significantly associated with NThalV when analyzed in the same association model, indicating that BMI and resistin are independently associated with NThalV in PPMS. ## Relation of BMI and adipokines with MRI volumes in healthy controls In HCs, higher BMI was associated with higher NWMV (β = 0.230, $$p \leq 0.016$$), whereas higher BMI was related to lower CGMV (β = − 0.197, $$p \leq 0.026$$) and lower NCbV (β = − 0.289, $$p \leq 0.001$$). In male HC, higher BMI was associated with higher LV (β = 0.382, $$p \leq 0.037$$). Finally, adipokine levels were not associated with MRI volumes in HC, although in female HC, leptin levels showed a trend towards significant negative association with MUCCA (β = − 0.221, $$p \leq 0.054$$). ## Relation of adipokines with clinical and radiological measures: BMI uncorrected All analyses (step 1) were repeated without BMI as covariate. Overall, without BMI correction, relations did not significantly change. The association between leptin and EDSS in male and female patients even lost its significance without adjusting for BMI and the association between leptin and EDSS in RRMS became weaker. In addition, the relation between adiponectin and NBV in SPMS and the relation between resistin and NthalV in PPMS lost its significance without BMI correction. The only association that became significant without BMI correction was the relation between resistin and NWMV in SPMS (Uncorrected: β = 0.288, $$p \leq 0.049$$; corrected: β = 0.264, $$p \leq 0.074$$), nevertheless, no significant association was reported between BMI and NWMV in SPMS. ## Discussion In a nation-wide MS cohort including patients and HC of the same age, we demonstrate independent associations of adipokines (adiponectin, resistin and leptin) and BMI with disability measures and MRI volumes. Although only adiponectin differed between PwMS and controls, adipokine levels showed several relations with clinical and radiological measures in specific subgroups of patients, with an opposite effect of leptin and resistin compared to adiponectin. These relations were observed while correcting for BMI, suggesting an additional role for adipokines in MS. Importantly, adipokine levels were only related to MRI volumes in PwMS and not in HC, further underlining the possible specific role of adipokines in MS. ## Adiponectin levels We found increased adiponectin levels in SPMS and male PPMS compared to controls. While several studies described reduced[15, 25, 26] or unaltered levels[27] in MS, the majority of studies reported increased adiponectin levels in patients in remission [28–32]. Earlier studies found positive associations of adiponectin with progression and disease severity in MS, as well as with inflammation and progression in rheumatoid arthritis, chronic kidney disease and inflammatory bowel disease [31, 33, 34]. The positive correlation with disease duration in female progressive MS and negative correlations with brain volumes in our cohort adds to these earlier findings. The presence of high adiponectin levels in MS might be indicative of an attempt, albeit ineffective, of the body to respond to (chronic) inflammation. Of note, our data mainly shows an association of adiponectin with progressive MS in which the inflammatory component is less pronounced. The observed sex-specific associations in progressive patients might be due to different cellular responses to adipokines between the sexes, since accumulating evidence points towards differential function and morphology of cell types explained by sex [35]. Our data may also indicate a dual role of adiponectin in pathological conditions. Since adiponectin is thought to primarily exert an anti-inflammatory effect, the increase of adiponectin in PwMS seems paradoxical. It has, however, been shown that adiponectin exhibits both anti-inflammatory and pro-inflammatory effects dependent on cell type and adiponectin receptor (Adipo-R) expression [36, 37]. Research has demonstrated that expression of Adipo-R1 and 2 is induced in pro-inflammatory mouse macrophages and adiponectin increased TNFα, IL-6 and IL-12, while in anti-inflammatory macrophages, Adipo-R expression is preserved and adiponectin induced the anti-inflammatory IL-10 [9]. Another study showed that adiponectin deficiency in mice led to exacerbated inflammatory responses in microglia in vivo while adiponectin treatment counteracted inflammatory cytokines in microglia, but worsened the response in astrocytes in vitro [38]. Thus, the actions of adiponectin are highly dependent on cell type and phenotype-specific receptor expression. ## Leptin levels Contrary to previous research, we found no apparent differences in leptin levels between PwMS and controls [39–41]. As most studies did not correct for confounders, the question remains whether previous reported higher leptin levels are actual differences or result from these confounders. Nevertheless, the relatively high age of our cohort and associated decrease in inflammation could have contributed to these discrepancies. Except for SPMS, we did find positive correlations between leptin and EDSS, as well as with LV in male RRMS. We also found a positive correlation with NCGMV in female patients, in line with earlier observations [42, 43]. While these results warrant further exploration, it is possible that leptin exerts differential functions during the more inflammatory disease phase compared to the progressive phase as well as in white versus grey matter. Several animal studies have shown that leptin-deficient mice carrying the obese mutation (ob/ob) are not susceptible for EAE, whereas subsequent intraperitoneal leptin replacement induced clinical symptoms. In contrast, intracerebral leptin injections stimulated proliferation of neuronal precursors [44] and reduced infarct volume in ischemic mice [45]. Such neuroprotective effects are further reinforced by studies which found that leptin is associated with larger brain volumes in healthy individuals [46] and regional GM volumes in elderly subjects [47]. It is well-established that adipokine levels significantly differ between sexes [20]. For instance, leptin and adiponectin levels are increased in females compared to males, which is again confirmed in our cohort. This sexual dimorphism is not entirely explained by either sex hormones or body fat distribution and may involve the additional release of leptin from non-adipose sources such as the brain [48]. ## Resistin levels To our knowledge, previous reports on correlations between resistin and MRI volumes are lacking. We found negative associations between resistin and grey matter volumes in PPMS. The role of resistin in neurodegeneration has not been elucidated yet and several modes of action could contribute to this pathophysiological process. In macrophages, resistin induces inflammatory cytokines and increases the expression of cell adhesion molecules [49]. Moreover, resistin is shown to induce endothelial dysfunction in blood vessels and promotes endothelial-monocyte adhesion and infiltration [49, 50]. Importantly, resistin leads to mitochondrial dysfunction, which contributes to progressive neurodegeneration [49, 51]. Thus, resistin could contribute to neurodegeneration via BBB dysfunction and subsequent immune cell infiltration and mitochondrial dysfunction [14]. In contrast, we observed positive associations between resistin and NBV and NThalV in SPMS. Larger SPMS and PPMS groups are required to explore whether resistin has differential effects in progressive phenotypes. ## Disease modifying therapy An important potential confounder in numerous studies is a lack of controlling for treatment status, which can significantly affect the levels of adipokines, since many treatments are based on immunosuppression [52]. Nevertheless, in our cohort, adipokine levels did not significantly differ between patients without DMT, patients with first-line DMT and patients with second-line DMT. Interestingly, we observed a significant reduction in resistin levels in patients treated with teriflunomide. Teriflunomide is hypothesized to ameliorate MS by reducing proliferation of activated lymphocytes [53], but also exerts direct inhibitory effects on pro-inflammatory cytokine release in monocytes [54]. As main sources of resistin are monocytes and macrophages, the decrease of resistin specifically in this treatment group might be explained by the anti-inflammatory effect of teriflunomide on monocytes. ## BMI Our findings that [1] adipokines are associated with clinical-, disability- and MRI measures while corrected for BMI [2] BMI associates with different outcome measures compared to adipokines and [3] associations between adipokines and MS metrics do not significantly change without BMI as covariate, suggesting that other (pathophysiological) mechanisms in MS, independent of BMI, are responsible for adipokine alterations. Our initial hypothesis implied that increased BMI in MS may lead to altered adipokine release, which results in the activation of inflammatory pathways [5]. Higher levels of pro-inflammatory cytokines in MS further enhance pro-inflammatory adipokine secretion, creating a positive feedback loop [5]. This would explain the absence of a direct link between BMI, adipokines and MS disease severity in our cohort. However, it remains unclear which stimulus induces this proposed positive feedback loop. In both HC and in PwMS, BMI was positively associated with NWMV and negatively associated with NCGMV. The positive relation between BMI and NWMV seems paradoxical and results should interpreted with caution. However, other studies have described similar positive relations, which hypothesized that pathological lipid metabolism in the brain of obese individuals may result in increased NWMV [55, 56]. Associations between higher BMI and reductions in normalized GM volume in MS as well as reduction in NBV in the healthy population have been previously described [4, 57]. The specific mechanisms through which obesity affects brain atrophy remain however poorly understood. The lack of association between adipokine levels and MRI volumes in HC suggest that adipokine alterations do not provide a direct link between increased BMI and brain atrophy in healthy individuals. ## Strengths and limitations The main strength of this study is that all patients and HC are of the same age. Age has a well-known effect on brain volume and on the immune system [58, 59]. Importantly, age significantly affects synthesis and function of adipokines; nearly all adipokines are increased in the older population compared to younger individuals with similar fat mass [19]. However, our study has certain drawbacks. Results are based on cross-sectional data and we could therefore not discriminate cause and effect. Moreover, while our cohort is one of the largest cohorts assessing adipokine levels to date, stratification may have led to loss of statistical power. Lastly, results cannot be generalized to younger MS populations that have generally a more active inflammatory profile. ## Conclusion In a cohort of PwMS and HC of the same age, we demonstrated associations of adipokines with clinical measures and brain volumes, indicating that adipokines are involved in MS. Associations between adipokines with a range of clinical and radiological metrics were independent from BMI, suggesting a different mechanism in the relation with MS disease severity. Our results aid to the understanding of the neuroprotective and neurotoxic effects of adipokines on the MS brain and could stimulate the development of targeted therapies based on hormonal interventions. ## References 1. Gianfrancesco MA, Acuna B, Shen L, Briggs FB, Quach H, Bellesis KH. Obesity during childhood and adolescence increases susceptibility to multiple sclerosis after accounting for established genetic and environmental risk factors*. Obes Res Clin Pract* (2014) 8 e435-e447. DOI: 10.1016/j.orcp.2014.01.002 2. Munger KL, Bentzen J, Laursen B, Stenager E, Koch-Henriksen N, Sorensen TI. Childhood body mass index and multiple sclerosis risk: a long-term cohort study. Mult Scler (2013) 19 1323-1329. DOI: 10.1177/1352458513483889 3. Stampanoni Bassi M, Iezzi E, Buttari F, Gilio L, Simonelli I, Carbone F. Obesity worsens central inflammation and disability in multiple sclerosis. Mult Scler (2020) 26 1237-1246. DOI: 10.1177/1352458519853473 4. Mowry EM, Azevedo CJ, McCulloch CE, Okuda DT, Lincoln RR, Waubant E. Body mass index, but not vitamin D status, is associated with brain volume change in MS. Neurology (2018) 91 e2256-e2264. DOI: 10.1212/WNL.0000000000006644 5. Rijnsburger M, Djuric N, Mulder IA, de Vries HE. Adipokines as immune cell modulators in multiple sclerosis. Int J Mol Sci (2021) 22 10845. DOI: 10.3390/ijms221910845 6. Wang Y, Wang X, Lau WB, Yuan Y, Booth D, Li JJ. Adiponectin inhibits tumor necrosis factor-alpha-induced vascular inflammatory response via caveolin-mediated ceramidase recruitment and activation. Circ Res (2014) 114 792-805. DOI: 10.1161/CIRCRESAHA.114.302439 7. Tilg H, Moschen AR. Adipocytokines: mediators linking adipose tissue, inflammation and immunity. Nat Rev Immunol (2006) 6 772-783. DOI: 10.1038/nri1937 8. Luo Y, Liu M. Adiponectin: a versatile player of innate immunity. J Mol Cell Biol (2016) 8 120-128. DOI: 10.1093/jmcb/mjw012 9. van Stijn CM, Kim J, Lusis AJ, Barish GD, Tangirala RK. Macrophage polarization phenotype regulates adiponectin receptor expression and adiponectin anti-inflammatory response. FASEB J (2015) 29 636-649. DOI: 10.1096/fj.14-253831 10. Francisco V, Pino J, Campos-Cabaleiro V, Ruiz-Fernandez C, Mera A, Gonzalez-Gay MA. Obesity, fat mass and immune system: role for leptin. Front Physiol (2018) 9 640. DOI: 10.3389/fphys.2018.00640 11. Bokarewa M, Nagaev I, Dahlberg L, Smith U, Tarkowski A. Resistin, an adipokine with potent proinflammatory properties. J Immunol (2005) 174 5789-5795. DOI: 10.4049/jimmunol.174.9.5789 12. Tarkowski A, Bjersing J, Shestakov A, Bokarewa MI. Resistin competes with lipopolysaccharide for binding to toll-like receptor 4. J Cell Mol Med (2010) 14 1419-1431. DOI: 10.1111/j.1582-4934.2009.00899.x 13. Thapa B, Lee K. Metabolic influence on macrophage polarization and pathogenesis. BMB Rep (2019) 52 360-372. DOI: 10.5483/BMBRep.2019.52.6.140 14. Parimisetty A, Dorsemans AC, Awada R, Ravanan P, Diotel N, Lefebvre DC. Secret talk between adipose tissue and central nervous system via secreted factors-an emerging frontier in the neurodegenerative research. J Neuroinflamm (2016) 13 67. DOI: 10.1186/s12974-016-0530-x 15. Düzel B, Tamam Y, Çoban A, Tüzün E. Adipokines in multiple sclerosis patients with and without optic neuritis as the first clinical presentation. Immunol Invest (2019) 48 190-197. DOI: 10.1080/08820139.2018.1528270 16. Emamgholipour S, Eshaghi SM, Hossein-nezhad A, Mirzaei K, Maghbooli Z, Sahraian MA. Adipocytokine profile, cytokine levels and foxp3 expression in multiple sclerosis: a possible link to susceptibility and clinical course of disease. PLoS ONE (2013) 8 e76555. DOI: 10.1371/journal.pone.0076555 17. Musabak U, Demirkaya S, Genç G, Ilikci RS, Odabasi Z. Serum adiponectin, TNF-α, IL-12p70, and IL-13 levels in multiple sclerosis and the effects of different therapy regimens. NeuroImmunoModulation (2011) 18 57-66. DOI: 10.1159/000317393 18. Natarajan R, Hagman S, Hämälainen M, Leppänen T, Dastidar P, Moilanen E. Adipsin is associated with multiple sclerosis: a follow-up study of adipokines. Mult Scler Int (2015) 2015 371734. PMID: 26634156 19. Mancuso P, Bouchard B. The impact of aging on adipose function and adipokine synthesis. Front Endocrinol (Lausanne) (2019) 10 137. DOI: 10.3389/fendo.2019.00137 20. Valencak TG, Osterrieder A, Schulz TJ. Sex matters:the effects of biological sex on adipose tissue biology and energy metabolism. Redox Biol (2017) 12 806-813. DOI: 10.1016/j.redox.2017.04.012 21. Loonstra FC, De Ruiter LRJ, Doesburg D. Project Y: The search for clues explaining phenotype variability in MS. Mult Scler Relat Disord (2022) 57 103337. DOI: 10.1016/j.msard.2021.103337 22. Thompson AJ, Banwell BL, Barkhof F, Carroll WM, Coetzee T, Comi G. Diagnosis of multiple sclerosis: 2017 revisions of the McDonald criteria. Lancet Neurol (2018) 17 162-173. DOI: 10.1016/S1474-4422(17)30470-2 23. Schoonheim MM, Douw L, Broeders TA, Eijlers AJ, Meijer KA, Geurts JJ. The cerebellum and its network: Disrupted static and dynamic functional connectivity patterns and cognitive impairment in multiple sclerosis. Mult Scler (2021) 27 2031-2039. DOI: 10.1177/1352458521999274 24. Sastre-Garriga J, Pareto D, Battaglini M, Rocca MA, Ciccarelli O, Enzinger C. MAGNIMS consensus recommendations on the use of brain and spinal cord atrophy measures in clinical practice. Nat Rev Neurol (2020) 16 171-182. DOI: 10.1038/s41582-020-0314-x 25. Kraszula L, Jasinska A, Eusebio M, Kuna P, Glabinski A, Pietruczuk M. Evaluation of the relationship between leptin, resistin, adiponectin and natural regulatory T cells in relapsing-remitting multiple sclerosis. Neurol Neurochir Pol (2012) 46 22-28. DOI: 10.5114/ninp.2012.27211 26. Piccio L, Cantoni C, Henderson JG, Hawiger D, Ramsbottom M, Mikesell R. Lack of adiponectin leads to increased lymphocyte activation and increased disease severity in a mouse model of multiple sclerosis. Eur J Immunol (2013) 43 2089-2100. DOI: 10.1002/eji.201242836 27. Musabak U, Demirkaya S, Genc G, Ilikci RS, Odabasi Z. Serum adiponectin, TNF-alpha, IL-12p70, and IL-13 levels in multiple sclerosis and the effects of different therapy regimens. NeuroImmunoModulation (2011) 18 57-66. DOI: 10.1159/000317393 28. Coban A, Duzel B, Tuzun E, Tamam Y. Investigation of the prognostic value of adipokines in multiple sclerosis. Mult Scler Relat Disord (2017) 15 11-14. DOI: 10.1016/j.msard.2017.04.006 29. Rasooli Tehrani A, Gholipour S, Sharifi R, Yadegari S, Abbasi-Kolli M, Masoudian N. Plasma levels of CTRP-3, CTRP-9 and apelin in women with multiple sclerosis. J Neuroimmunol (2019) 333 576968. DOI: 10.1016/j.jneuroim.2019.576968 30. Kvistad SS, Myhr KM, Holmoy T, Benth JS, Wergeland S, Beiske AG. Serum levels of leptin and adiponectin are not associated with disease activity or treatment response in multiple sclerosis. J Neuroimmunol (2018) 323 73-77. DOI: 10.1016/j.jneuroim.2018.07.011 31. Signoriello E, Lus G, Polito R, Casertano S, Scudiero O, Coletta M. Adiponectin profile at baseline is correlated to progression and severity of multiple sclerosis. Eur J Neurol (2019) 26 348-355. DOI: 10.1111/ene.13822 32. Hietaharju A, Kuusisto H, Nieminen R, Vuolteenaho K, Elovaara I, Moilanen E. Elevated cerebrospinal fluid adiponectin and adipsin levels in patients with multiple sclerosis: a Finnish co-twin study. Eur J Neurol (2010) 17 332-334. DOI: 10.1111/j.1468-1331.2009.02701.x 33. Frommer KW, Zimmermann B, Meier FM, Schroder D, Heil M, Schaffler A. Adiponectin-mediated changes in effector cells involved in the pathophysiology of rheumatoid arthritis. Arthritis Rheum (2010) 62 2886-2899. DOI: 10.1002/art.27616 34. Gremese E, Tolusso B, Gigante MR, Ferraccioli G. Obesity as a risk and severity factor in rheumatic diseases (autoimmune chronic inflammatory diseases). Front Immunol (2014) 5 576. DOI: 10.3389/fimmu.2014.00576 35. Chen KE, Lainez NM, Coss D. Sex differences in macrophage responses to obesity-mediated changes determine migratory and inflammatory traits. J Immunol (2021) 206 141-153. DOI: 10.4049/jimmunol.2000490 36. Esmaili S, Xu A, George J. The multifaceted and controversial immunometabolic actions of adiponectin. Trends Endocrinol Metab (2014) 25 444-451. DOI: 10.1016/j.tem.2014.06.001 37. Choi HM, Doss HM, Kim KS. Multifaceted physiological roles of adiponectin in inflammation and diseases. Int J Mol Sci (2020) 21 1219. DOI: 10.3390/ijms21041219 38. Nicolas S, Cazareth J, Zarif H, Guyon A, Heurteaux C, Chabry J. Globular adiponectin limits microglia pro-inflammatory phenotype through an AdipoR1/NF-kappaB signaling pathway. Front Cell Neurosci (2017) 11 352. DOI: 10.3389/fncel.2017.00352 39. Chatzantoni K, Papathanassopoulos P, Gourzoulidou E, Mouzaki A. Leptin and its soluble receptor in plasma of patients suffering from remitting-relapsing multiple sclerosis (MS) In vitro effects of leptin on type-1 and type-2 cytokine secretion by peripheral blood mononuclear cells, T-cells and monocytes of MS patients. J Autoimmun (2004) 23 169-177. DOI: 10.1016/j.jaut.2004.05.007 40. Dashti M, Alroughani R, Jacob S, Al-Temaimi R. Leptin rs7799039 polymorphism is associated with multiple sclerosis risk in Kuwait. Mult Scler Relat Disord (2019) 36 101409. DOI: 10.1016/j.msard.2019.101409 41. Evangelopoulos ME, Koutsis G, Markianos M. Serum leptin levels in treatment-naive patients with clinically isolated syndrome or relapsing-remitting multiple sclerosis. Autoimmune Dis (2014) 2014 486282. PMID: 25505980 42. Turan S, Sarioglu FC, Erbas IM, Cavusoglu B, Karagoz E, Sisman AR. Altered regional grey matter volume and appetite-related hormone levels in adolescent obesity with or without binge-eating disorder. Eat Weight Disord (2021) 26 2555-2562. DOI: 10.1007/s40519-021-01117-4 43. Pannacciulli N, Le DS, Chen K, Reiman EM, Krakoff J. Relationships between plasma leptin concentrations and human brain structure: a voxel-based morphometric study. Neurosci Lett (2007) 412 248-253. DOI: 10.1016/j.neulet.2006.11.019 44. Perez-Gonzalez R, Antequera D, Vargas T, Spuch C, Bolos M, Carro E. Leptin induces proliferation of neuronal progenitors and neuroprotection in a mouse model of Alzheimer's disease. J Alzheimers Dis (2011) 24 17-25. DOI: 10.3233/JAD-2011-102070 45. Zhang F, Wang S, Signore AP, Chen J. Neuroprotective effects of leptin against ischemic injury induced by oxygen-glucose deprivation and transient cerebral ischemia. Stroke (2007) 38 2329-2336. DOI: 10.1161/STROKEAHA.107.482786 46. Lieb W, Beiser AS, Vasan RS, Tan ZS, Au R, Harris TB. Association of plasma leptin levels with incident Alzheimer disease and MRI measures of brain aging. JAMA (2009) 302 2565-2572. DOI: 10.1001/jama.2009.1836 47. Narita K, Kosaka H, Okazawa H, Murata T, Wada Y. Relationship between plasma leptin level and brain structure in elderly: a voxel-based morphometric study. Biol Psychiatry (2009) 65 992-994. DOI: 10.1016/j.biopsych.2008.10.006 48. Cote I, Green SM, Toklu HZ, Morgan D, Carter CS, Tumer N. Differential physiological responses to central leptin overexpression in male and female rats. J Neuroendocrinol. (2017) 29 e12552. DOI: 10.1111/jne.12552 49. Acquarone E, Monacelli F, Borghi R, Nencioni A, Odetti P. Resistin: a reappraisal. Mech Ageing Dev (2019) 178 46-63. DOI: 10.1016/j.mad.2019.01.004 50. Hsu WY, Chao YW, Tsai YL, Lien CC, Chang CF, Deng MC. Resistin induces monocyte-endothelial cell adhesion by increasing ICAM-1 and VCAM-1 expression in endothelial cells via p38MAPK-dependent pathway. J Cell Physiol (2011) 226 2181-2188. DOI: 10.1002/jcp.22555 51. Chen Z, Tao S, Li X, Yao Q. Resistin destroys mitochondrial biogenesis by inhibiting the PGC-1alpha/ NRF1/TFAM signaling pathway. Biochem Biophys Res Commun (2018) 504 13-18. DOI: 10.1016/j.bbrc.2018.08.027 52. De Angelis F, John NA, Brownlee WJ. Disease-modifying therapies for multiple sclerosis. BMJ (2018) 363 k4674. DOI: 10.1136/bmj.k4674 53. Bar-Or A, Pachner A, Menguy-Vacheron F, Kaplan J, Wiendl H. Teriflunomide and its mechanism of action in multiple sclerosis. Drugs (2014) 74 659-674. DOI: 10.1007/s40265-014-0212-x 54. Li L, Liu J, Delohery T, Zhang D, Arendt C, Jones C. The effects of teriflunomide on lymphocyte subpopulations in human peripheral blood mononuclear cells in vitro. J Neuroimmunol (2013) 265 82-90. DOI: 10.1016/j.jneuroim.2013.10.003 55. Haltia LT, Viljanen A, Parkkola R, Kemppainen N, Rinne JO, Nuutila P. Brain white matter expansion in human obesity and the recovering effect of dieting. J Clin Endocrinol Metab (2007) 92 3278-3284. DOI: 10.1210/jc.2006-2495 56. Yokum S, Ng J, Stice E. Relation of regional gray and white matter volumes to current BMI and future increases in BMI: a prospective MRI study. Int J Obes (Lond) (2012) 36 656-664. DOI: 10.1038/ijo.2011.175 57. Bobb JF, Schwartz BS, Davatzikos C, Caffo B. Cross-sectional and longitudinal association of body mass index and brain volume. Hum Brain Mapp (2014) 35 75-88. DOI: 10.1002/hbm.22159 58. Mogilenko DA, Shchukina I, Artyomov MN. Immune ageing at single-cell resolution. Nat Rev Immunol (2021) 22 484-498. DOI: 10.1038/s41577-021-00646-4 59. De Stefano N, Stromillo ML, Giorgio A, Bartolozzi ML, Battaglini M, Baldini M. Establishing pathological cut-offs of brain atrophy rates in multiple sclerosis. J Neurol Neurosurg Psychiatry (2016) 87 93-99. PMID: 25904813
--- title: Pharmacological and mechanistic study of PS1, a Pdia4 inhibitor, in β-cell pathogenesis and diabetes in db/db mice authors: - Hui-Ju Tseng - Wen-Chu Chen - Tien-Fen Kuo - Greta Yang - Ching-Shan Feng - Hui-Ming Chen - Tzung-Yan Chen - Tsung-Han Lee - Wen-Chin Yang - Keng-Chang Tsai - Wei-Jan Huang journal: 'Cellular and Molecular Life Sciences: CMLS' year: 2023 pmcid: PMC10025235 doi: 10.1007/s00018-022-04677-5 license: CC BY 4.0 --- # Pharmacological and mechanistic study of PS1, a Pdia4 inhibitor, in β-cell pathogenesis and diabetes in db/db mice ## Abstract Pdia4 has been characterized as a key protein that positively regulates β-cell failure and diabetes via ROS regulation. Here, we investigated the function and mechanism of PS1, a Pdia4 inhibitor, in β-cells and diabetes. We found that PS1 had an IC50 of 4 μM for Pdia4. Furthermore, PS1 alone and in combination with metformin significantly reversed diabetes in db/db mice, 6 to 7 mice per group, as evidenced by blood glucose, glycosylated hemoglobin A1c (HbA1c), glucose tolerance test, diabetic incidence, survival and longevity ($P \leq 0.05$ or less). Accordingly, PS1 reduced cell death and dysfunction in the pancreatic β-islets of db/db mice as exemplified by serum insulin, serum c-peptide, reactive oxygen species (ROS), islet atrophy, and homeostatic model assessment (HOMA) indices ($P \leq 0.05$ or less). Moreover, PS1 decreased cell death in the β-islets of db/db mice. Mechanistic studies showed that PS1 significantly increased cell survival and insulin secretion in Min6 cells in response to high glucose ($P \leq 0.05$ or less). This increase could be attributed to a reduction in ROS production and the activity of electron transport chain complex 1 (ETC C1) and Nox in Min6 cells by PS1. Further, we found that PS1 inhibited the enzymatic activity of Pdia4 and mitigated the interaction between Pdia4 and Ndufs3 or p22 in Min6 cells ($P \leq 0.01$ or less). Taken together, this work demonstrates that PS1 negatively regulated β-cell pathogenesis and diabetes via reduction of ROS production involving the Pdia4/Ndufs3 and Pdia4/p22 cascades. ### Supplementary Information The online version contains supplementary material available at 10.1007/s00018-022-04677-5. ## Introduction The International Federation of Diabetes estimated that in 2021 there were 537 million people living with diabetes worldwide and 5 million lives were lost due to the disease [1]. One of the characterizing traits of diabetes is the failure of functional β-cells to modulate insulin secretion to offset increasing insulin resistance, driving disease development [2]. Unfortunately, no anti-diabetes drugs are available in the clinic for preserving β-cells. Beta-cell failure is known to be important in diabetes development [3], and preservation of functional β-cells can reverse the outcome of clinical diabetes [4, 5]. Thus, the search for new genes implicated in β-cell failure can assist in developing therapeutic approaches to cure diabetes. Despite some progress on β-cell failure, the molecular mechanism that orchestrates β-cell number and function is poorly studied. Diabetogenic stimuli always cause endoplasmic reticulum (ER) stress and oxidative stress [6, 7]. Consequently, exuberant reactive oxygen species (ROS) result in β-cell demise and dysfunction [8, 9] and peripheral insulin resistance [10] in animals and humans. Besides, β-cells are known to be susceptible to ROS because they possess fewer anti-oxidant proteins than the other cell types [2]. The mitochondrial electron transport chain (ETC), NADPH oxidase (Nox), and ER oxidoreductin 1 are common machineries for ROS generation [11–13]. Some potential molecular targets of β-cell failure have been identified including BACH2 [14], MST1 [15], GSK3 [16], PDX1 [16], P2RY1 [17], PHLPP [18], Pdia4 [2], etc. Among them, Pdia4 was the only target whose deficiency could reverse diabetes [2]. Pdia4 is a molecular chaperone with 3 GCHC motifs in the Pdi family. Pdia4 has been reported to be mainly expressed in β-cells, and its expression was up-modulated in β-cells and sera of rodents by nutrient overload [2]. Furthermore, data obtained from Pdia4 knockout and transgenic mice showed that Pdia4 promoted β-cell failure, including cell death and dysfunction, and diabetes via up-regulation of ROS production as indicated by fasting blood glucose (FBG), postprandial blood glucose (PBG), glycosylated hemoglobin A1c (HbA1c), glucose tolerance test (GTT), islet architecture, diabetic incidence, and homeostatic model assessment for β-cell function and insulin resistance (HOMA) indices. Mechanistically speaking, Pdia4 was found to augment ROS production in β-cells via its interplay and activation of Ndufs3 and p22 in the ETC complex 1 (ETC C1) and Nox pathways, respectively [2]. This seminal publication identified Pdia4 as a novel therapeutic target of β-cell pathogenesis and diabetes as a result of ROS dysregulation [2]. PS1 has been approved as an investigational new drug for diabetes by the Food and Drug Administration of the United States (https://clinicaltrials.gov/ct2/show/NCT05176210?term=PS1&draw=2&rank=1). We hypothesized that PS1, a drug candidate that inhibits Pdia4, could reverse β-cell pathogenesis and diabetes in db/db mice. In the current investigation, we studied the pharmacological effect and mechanism of PS1 at the molecular, cellular and animal levels. ## Cells, plasmids, and reagents A murine β-cell line, Min6 cells, and pancreatic islets isolated from B6 mice were grown in DMEM medium (Sigma D5648) supplemented with $20\%$ FBS, 3.4 mg/mL NaHCO3, 75 μg/mL penicillin, 50 μg/mL streptomycin, and glucose at the indicated dosages. The cells were grown at 37 °C in a $5\%$ CO2 incubator. Expression plasmids, Flag-Pdia4, Myc/Flag-tagged Ndufs3, and Myc/Flag-tagged p22, were constructed and transfected as published previously [2]. PS1 with a molecular formula of C12H11NO5 (Pharmasaga, Taipei, Taiwan), sitagliptin (STG, Merck, Kenilworth, NJ), metformin (Merck) and optimal cutting temperature (OCT, Thermo Fisher, Waltham, MA) were obtained. Diaminobenzidine tetrahydrochloride (DAB), NADPH, histopaque-1077, BSA, glucose, PBS, trypsin, and palmitate were bought from Sigma (St Louis, MO). Propidium iodide (PI), dihydroethidium (DHE), DAPI, chloromethyl-2′,7′-dichlorodihydrofluorescein diacetate (CM-H2DCFDA), CellROX, MitoSOX, MitoGreen, and Hochest 33,342 were bought from Molecular Probes (Eugene, OR). Elite glucometers and chips were purchased from Bayer (Germany). Enzyme-linked immunosorbent assay (ELISA) kits for insulin and c-peptide were purchased from Mercodia (Uppsala, Sweden) and Crystal Chem (Elk Grove Village, IL). Lucigenin, HEPES, and PBS were purchased from Roche (Switzerland). Antibodies against Ki67 (MIB-1, Santa Cruz Biotech, Santa Cruz, CA) and insulin (H-86, Santa Cruz Biotech) and terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL) kits (Chemicon, Temecula, CA) were purchased. ## Drug administration and measurement of metabolic parameters PBS vehicle (CTR), PS1 (0.8 mg/kg, 2.5 mg/kg, and 7.5 mg/kg) alone and in combination (2.5 mg/kg PS1 with 60 mg/kg metformin), and two therapeutic positive controls, 15 mg/kg STG alone and in combination (15 mg/kg STG plus 60 mg/kg metformin) were given orally to female db/db mice, a mouse model of type 2 diabetes, once a day from 8 to 24 weeks of age. The number of mice in each group was 9 females per group except for the group of PS1 at 2.5 and 7.5 mg/kg, where 10 females were used. By 18 weeks of age, 3 mice in each group were removed for histochemical analysis. The drugs were withdrawn from each group of mice from 25 weeks of age to death. Food intake, water consumption, body weight, FBG, PBG, HbA1c, diabetic incidence, HOMA-β and HOMA-IR, GTT, serum insulin, serum c-peptide, and serum ROS of the mice were monitored at the time points shown. To measure FBG and PBG, the mice were fasted overnight and bled for blood samples. The mice had free access to food for 2 h. After 30 min, they were bled for blood samples. The levels of blood glucose from fasted and starved mice were measured using an Elite glucometer (Bayer, Germany). Mice with FBG or PBG over 126 or 200 mg/dL, respectively, for two consecutive measurements were considered diabetic. Diabetic incidence was calculated based on FBG and PBG. For GTT, db/db mice at 8, 16, 24 weeks were denied access to chow for 16 h. The mice were administered an intraperitoneal injection of 1 g/kg glucose. Blood samples were collected from tail veins at 0, 30, 60, 120 and 180 min after glucose injection. The levels of blood glucose were measured using an Elite glucometer. HOMA-β [20 × fasting insulin (mU/mL)/(fasting glucose (mmol/L)—3.5] and HOMA-IR [fasting glucose (mmol/L) × fasting insulin (μU/mL)/22.5] indices were calculated. The level of serum c-peptide, serum insulin, and HbA1c in mouse blood samples was determined as published previously [2]. The mouse sera were incubated with lucigenin [19], and ROS was assessed using a SpectraMax i3x reader (Molecular Devices, San Jose, CA). All animals were maintained at 21–23 °C with 12 h light-12 h dark cycles in the institutional animal facility and handled according to the Academia Sinica Institutional Animal Care and Utilization Committee [11-03-158]. ## Flow cytometric analysis Min6 cells were pre-incubated with complete DMEM medium containing PBS (CTR and HG), STG (10 μg/mL) and PS1 (0.2, 1, and 5 μg/mL) for 2 h. The cells were then incubated with DMEM medium, and the medium was supplemented with 30 mM glucose (HG, STG, and PS1) for an additional 2 h. After washing, the cells were stained with PI (1 μg/mL) for 5 min and analyzed using a LSRII analyzer. FlowJo software was used to analyze the cytometric data. ## Pdia4 assays As published [2], recombinant Pdia4 was incubated with PBS vehicle and PS1 in the presence of insulin substrates at 25 °C for 30 min. Following stop solution and detection reagents, the turbidity of each sample was measured using a Biotek Cytation 5 reader at 595 nm and, subsequently, Pdia4 activity was determined. Alternatively, Pdia4 was precipitated from the pancreata of db/db mice given PBS and drugs at the indicated dosages using anti-Pdia4 antibody. ## Immunoblotting assays Min6 cells were transiently transfected with the construct expressing Flag-Pdia4 and that encoding Myc/Flag-tagged Ndufs3 (B) or Myc/Flag-tagged p22 using a TransIT-LT1 Transfection kit (Mirus Bio). Following 24 h culture, the cells were treated with PBS and PS1 (5 μM) for 0.5 h. Following cell lysis, total lysates and anti-Pdia4 precipitates were subjected to immunoblot analysis with anti-Flag antibody. ## Measurement of ROS and the activity of ETC C1 and Nox [2] Min6 cells were treated with medium, 16.7 mM glucose, and a combination of 16.7 mM glucose with STG (10 μg/mL) and PS1 at 0.2, 1, and 5 μg/mL for 30 min. The cells were incubated with Hoechst 33,342, a nuclear dye, plus CM-H2DCFDA, a dye for cytosolic ROS, and MitoGreen, a mitochondrial tracker, plus MitoSOX, a dye for mitochondrial ROS, and analyzed using confocal microscopy. To measure the ETC C1 and Nox activity, membrane and mitochondrial fractions of the cells were individually extracted (protein extraction kits, Abcam, UK) and tested for the activity of Nox and ETC C1, respectively, as previously described [2]. ## Insulin quantification Min6 cells were pre-treated in complete DMEM with 30 mM glucose at 37 °C for 2 h. The cells were washed extensively and stimulated with medium containing high glucose (16.7 mM), STG (10 µg/mL) and PS1 (0.2 µg/mL, 1 µg/mL, and 5 µg/mL) at 37 °C for 30 min. Their supernatants were collected. Alternatively, Min6 cells were stimulated with DMEM medium or the medium KCl (30 mM), STG (10 µg/mL) and PS1 (0.2 µg/mL, 1 µg/mL, and 5 µg/mL) at 37 °C for 30 min. The supernatants were collected. The supernatants from Min6 cells or mouse serum were measured for insulin ELISA assays according to the manufacturer’s protocol (Mercodia, Uppsala, Sweden). ## Immunohistochemical (IHC) analysis After sacrifice, the pancreata from 3 mice per group were frozen in OCT medium and cryosectioned. The sections were stained with DHE for ROS content. To stain for insulin, the pancreatic sections were fixed and incubated with anti-insulin antibody and developed with DAB. To assess β-cell proliferation, the pancreatic sections were stained and developed with anti-Ki67 kits (BD Biosciences, San Jose, CA). For cell death detection, the pancreatic sections underwent TUNEL assays (Chemicon, Temecula, CA). The above slides were imaged and analyzed using the AxioVision program (Carl Zeiss). To quantify cell proliferation and death in β-cells, the Ki67-positive or TUNEL-positive cells were counted under a microscope. ## Statistics Data from three or more independent experiments are expressed as mean ± standard deviation (SD). Nonparametric tests and log rank were used to determine if there are any statistical differences between groups. P () < 0.05; P () < 0.01 and P (*) < 0.001 were considered statistically significant. The number of mice (n) is shown in parentheses. ## PS1 alone and in combination with metformin can reduce diabetes and increase survival and life span in db/db mice PS1 was rationally developed as a Pdia4 small-molecule inhibitor using a combination of a molecular docking strategy and total chemical synthesis (Figure S1A). Furthermore, PS1 had a half maximal inhibitory concentration (IC50) value of 4 μM for Pdia4. These data prompted us to assess the anti-diabetic effect of PS1 in db/db mice in new-onset diabetes (Figure S1B). First, we assessed the glucose-reduction caused by PS1 in diabetic db/db mice. As anticipated, db/db mice had manifested diabetes by 8 weeks and this disease lasted for their life time (CTR, Fig. 1A). In contrast, STG is a commercial DPP4 inhibitor for diabetes. Sixteen-week treatment with STG, a positive control of monotherapy, and 16-week treatment with STG plus metformin, a positive control of combinational therapy, modestly lowered FBG and PBG in diabetic db/db mice (STG@15 mg/kg versus STG@15 mg/kg + Met@60 mg/kg, Fig. 1A). Of note, PS1 dose-dependently reduced diabetes in diabetic db/db mice as evidenced by FBG (PS1, left, Fig. 1A) and PBG (PS1, right, Fig. 1A). A combination of PS1 and metformin had slightly better glucose-lowering effects than PS1 alone (FBG and PBG, PS1@2.5 + Met@60 mg/kg, Fig. 1A).Fig. 1Beneficial effects of PS1 on diabetes development in db/db mice. A Female db/db mice that had new-onset diabetes were grouped and treated with PBS (CTR), STG (15 mg/kg), a combination of STG (15 mg/kg) and metformin (Met, 60 mg/kg), PS1 (0.8 mg/kg, 2.5 mg/kg, and 7.5 mg/kg), and a combination of PS1 (2.5 mg/kg) and metformin (60 mg/kg) from 8 to 56 weeks of age. Their fasting blood glucose (FBG) and postprandial blood glucose (PBG) levels were found using a glucometer at the indicated ages. B HbA1c of the mice A was monitored. C GTT of the mice A was monitored by the aged of 8, 16 and 24 weeks. Mouse number (n) is shown in parentheses. P () < 0.05, P () < 0.01 and P () < 0.001 were considered statistically significant HbA1c is known as a reliable marker of chronic glycemia. Therefore, we next checked the effects of PS1 on HbA1c in diabetic db/db mice. Each group of db/db mice, aged 8 weeks, had an HbA1c value of $4\%$ (8 weeks, Fig. 1B). This value increased to $10\%$ in control db/db mice aged 16 weeks and beyond (CTR, 16–56 weeks, Fig. 1B). In contrast, db/db mice treated with STG per se and in combination with metformin reduced the HbA1c value to 8.3–$7.7\%$ in age-matched db/db mice (STG@15 mg/kg versus STG@15 mg/kg + Met@60 mg/kg, Fig. 1B). Furthermore, PS1 dose-dependently reduced the HbA1c value to 7.9–$5.9\%$ in age-matched db/db mice (PS1, Fig. 1B). The db/db mice given a combination of PS1 and metformin had an HbA1c value of $5.3\%$, which was slightly lower than the HbA1c value in the age-matched mice treated with PS1 alone (PS1@2.5 + Met@60 mg/kg versus PS1, Fig. 1B). Next, GTT was measured in each group of db/db mice aged 8, 16 and 24 weeks. No difference was observed in the GTT of each group of mice at the age of 8 weeks (left, Fig. 1C). However, STG per se and in combination with metformin significantly improved GTT compared to PBS vehicle in db/db mice aged 16 weeks (STG@15 mg/kg versus STG@15 mg/kg + Met@60 mg/kg, middle, Fig. 1C). PS1 alone and in combination with metformin further improved GTT in 16-week-old db/db mice in a dose-dependent fashion (PS1 at 0.8, 2.5 and 7.5 mg/kg versus PS1@2.5 mg/kg + Met@60 mg/kg, middle, Fig. 1C). A combination of PS1 and metformin improved GTT more than PS1 alone in 16-week-old db/db mice (PS1 at 0.8, 2.5 and 7.5 mg/kg versus PS1@2.5 mg/kg + Met@60 mg/kg, middle, Fig. 1C). By 24 weeks of age, PS1 alone and in combination with metformin for 16 weeks improved GTT in db/db mice more than 8-week treatment with the same drugs (right, Fig. 1C). As far as diabetic incidence is concerned, $100\%$ of new-onset diabetic db/db mice developed severe diabetes over time (CTR, Fig. 2A). Although STG per se and in combination with metformin lowered the blood glucose level of the db/db mice (Fig. 1A), such treatments failed to reduce diabetic incidence (STG@15 mg/kg versus STG@15 mg/kg + Met@60 mg/kg, Fig. 2A). However, PS1 at 2.5, and 7.5 mg/kg reduced diabetic incidence in db/db mice by $0\%$, 13–$29\%$, and 43–$86\%$, respectively (Fig. 2A). In sharp contrast, a combination of PS1 and metformin fully reversed diabetes in db/db mice (PS1@2.5 mg/kg + Met@60 mg/kg, Fig. 2A). We also examined survival rate and life span in different mouse groups. Sixteen-week treatment with PS1 significantly improved survival (PS1, Fig. 2B) and life span (PS1, Fig. 2C) in db/db mice according to dose. Furthermore, db/db mice treated with PS1 and metformin had better survival and life span (PS1@2.5 mg/kg + Met@60 mg/kg, Fig. 2B, C). Overall, the data indicated that the Pdia4 inhibitor, PS1, alone and in combination, could treat and reverse diabetes. Fig. 2Promotion of diabetes reversal, survival rate and longevity in db/db mice by PS1. Diabetic incidence (A), survival rate (B), and life span (C) of the mice from Fig. 1A from birth to death were measured. Diabetic incidence of the mice was calculated based on their PBG over 126 mg/dL. Mouse number (n) is shown in parentheses. P () < 0.05, P () < 0.01 and P () < 0.001 were considered statistically significant ## PS1 increases serum insulin and serum c-peptide and improves HOMA indices in db/db mice In parallel, we tested how PS1 affected the function of pancreatic islets in db/db mice. No difference in levels of postprandial serum insulin was observed in any of the groups of mice aged 7 weeks (left, Fig. 3A). Levels of postmeal serum insulin in control db/db mice gradually decreased over time (CTR, 7–58 weeks, Fig. 3A). However, mice given STG alone and in combination with metformin had slightly more postmeal serum insulin than control mice (STG@15 mg/kg versus STG@15 mg/kg + Met@60 mg/kg, 18–58 weeks, Fig. 3A). In contrast, PS1 dose-dependently elevated the levels of postmeal serum insulin in db/db mice (PS1 at 0.8, 2.5 and 7.5 mg/kg, 18–58 weeks, Fig. 3A). Since c-peptide is a predictor of β-cell function, next, the effect of PS1 on the levels of c-peptide in db/db mice was examined. There was no difference in the levels of postprandial serum c-peptide in any of the groups of mice aged 7 weeks (7 weeks, Fig. 3B). Levels of serum c-peptide in control db/db mice gradually decreased over time (CTR, 7–58 weeks, Fig. 3B). However, mice given STG per se and in combination with metformin slightly increased the levels of serum c-peptide in db/db mice over time (STG@15 mg/kg versus STG@15 mg/kg + Met@60 mg/kg, 18–58 weeks, Fig. 3B). In contrast, PS1 per se and in combination with metformin dose-dependently augmented the levels of serum c-peptide in db/db mice (PS1 at 0.8, 2.5 and 7.5 mg/kg versus PS1@2.5 mg/kg + Met@60 mg/kg, 18–58 weeks, Fig. 3B). Since HOMA indices are useful markers of β-cell function and insulin resistance, next, we checked the effect of PS1 on the HOMA-β and HOMA-IR indices in diabetic db/db mice. There was no difference in the HOMA-β index in any group of db/db mice at 8 weeks (Fig. 3C). However, the HOMA-β index in db/db mice dramatically declined over time (CTR, Fig. 3C). STG per se and in combination with metformin modestly reduced this decline in db/db mice (STG@15 mg/kg versus STG@15 mg/kg + Met@60 mg/kg, 16–56 weeks, Fig. 3C). In remarkable contrast, PS1 per se and in combination with metformin significantly increased the HOMA-β index in db/db mice according to dose (PS1 at 0.8, 2.5 and 7.5 mg/kg versus PS1@2.5 mg/kg + Met@60 mg/kg, 16—56 weeks, Fig. 3C). In agreement, no difference in the HOMA-IR index was perceived in any group of 8-week-old db/db mice (Fig. 3D). However, the HOMA-IR index in db/db mice gradually increased over time (CTR, Fig. 3D). STG per se and in combination with metformin moderately reduced this increase in db/db mice (STG@15 mg/kg versus STG@15 mg/kg + Met@60 mg/kg, 16–56 weeks, Fig. 3D). PS1 per se and in combination with metformin significantly reduced the HOMA-IR index in db/db mice in a dose-dependent fashion (PS1@0.8, 2.5 and 7.5 mg/kg versus PS1@2.5 + Met@60 mg/kg, 16—56 weeks, Fig. 3D). As a result, the HOMA-IR index in each group of mice was inversely proportional to the HOMA-β index (Fig. 3C, D). Collectively, the data showed that PS1 ameliorated the function of the pancreatic islets. Fig. 3Improvement of serum insulin, c-peptide and HOMA indices in db/db mice by PS1. A The insulin concentration of the mice (Fig. 1A) was measured at the age of 7, 18, 30 and 58 weeks. B C-peptide of the mice (Fig. 1A) was determined at the indicated ages. HOMA-β (C) and (D) HOMA-IR of the mice (Fig. 1A) were measured at the age of 8, 16, 24 and 56 weeks. Mouse number (n) is shown in parentheses. P () < 0.05, P () < 0.01 and P () < 0.001 were considered statistically significant ## PS1 reduces islet atrophy, islet ROS, and serum ROS in db/db mice We also looked at the effect of PS1 on the architecture of pancreatic islets in db/db mice. Control db/db mice, aged 18 weeks, had smaller islet size than those treated with STG alone or in combination with metformin (STG@15 mg/kg versus STG@15 mg/kg + Met@60 mg/kg, Ins, Fig. 4A, B). Of note, db/db mice treated with PS1 per se and in combination with metformin further increased the islet size (PS1 at 0.8, 2.5 and 7.5 mg/kg versus PS1@2.5 + Met@60 mg/kg, Ins, Fig. 4A, B). This increase seemed to be dose-dependent. The data suggested that PS1 reduced the islet atrophy in a dose-dependent fashion. Consistently, ROS content in the islets of db/db mice treated with PS1 and a combination of PS1 plus metformin was inversely proportional to their islet size as evidenced by the signals of DHE staining (DHE, Fig. 4A, C). The data on the ROS content of the mouse islets were in good agreement with the data on serum ROS of db/db mice (Fig. 4D).Fig. 4Reduction of islet atrophy, islet ROS, and serum ROS in db/db mice by PS1. A The pancreata* of the mice (Fig. 1A) at the age of 18 weeks were fixed and stained with anti-insulin antibody (top) and DHE (bottom). Representative images were photographed. Scale bar: 50 µm. Islet area (μm2) (B) and relative fluorescence intensity (RFI) (C) of the pancreatic islets of the mice A were quantified and re-plotted into histograms. Scale bar = 100 μm. The dashed circles show islet regions. D Serum ROS of the mice (Fig. 1A) was measured at the age of 18 weeks. The number of mice (n) is indicated in parentheses. P () < 0.05, P () < 0.01 and P (*) < 0.001 were considered statistically significant The overall data demonstrated that PS1 reduced islet atrophy and elevated the ROS level of the islets and sera in db/db mice. ## PS1 reduces cell death rather than cell proliferation in the islets of db/db mice To probe the mechanism by which PS1 protected against β-cell failure, we first studied the action of PS1 on β-cell proliferation and demise using Ki67 staining and TUNEL assays, respectively. There was no statistical difference in the proliferation of cells of the islets among control db/db and the mice treated with PS1 and other drugs based on the IHC staining with the antibody against Ki67, a cell proliferation marker (Fig. 5A). However, control db/db mice had a comparable number of TUNEL-positive islet cells, primarily dead islet cells, as the age-matched mice treated with STG alone and in combination with metformin (STG@15 mg/kg versus STG@15 mg/kg + Met@60 mg/kg, Fig. 5B). In remarkable contrast, treatment with PS1 per se or in combination with metformin significantly diminished the number of TUNEL-positive islet cells in db/db mice (PS1 at 0.8, 2.5 and 7.5 mg/kg versus PS1@2.5 mg/kg + Met@60 mg/kg, Fig. 5B). Furthermore, the diminishment of TUNEL-positive islet cells in the mouse pancreata by PS1 seemed to be dose-dependent. We also explored the effect of PS1 on the expression of Aldh1a3, an indicator of β-cell dedifferentiation. PS1 dose-dependently decreased the expression level of Aldh1a3 in mouse islets (Fig. 5C). To sum up, both types of assays suggested that PS1 was able to reduce cell death but not cell proliferation in pancreatic islets of db/db mice. Fig. 5Decrease in cell death and Aldh1a3 expression in the pancreatic islets of db/db mice by PS1. A, B The pancreata* of the mice (Fig. 1A) were fixed and stained with anti-Ki67 and TUNEL reagent. Ki67-positive cells per islet area (0.05 mm2) A were visualized in the islets of the mice (left). After quantification, the signals of each group were quantified and re-plotted into histograms (right). TUNEL-positive cells per islet area (0.05 mm2) B were visualized in the islets of the mice (left). After quantification, the signals of each group were quantified and re-plotted into histograms (right). C The pancreata of the mice (Fig. 1A) were fixed and stained with anti-Aldh1a3 (red). Representative images of Aldh1a3 in the islets were acquired using a confocal microscope. After quantification, the mean fluorescence intensity (MFI) of each group was quantified and re-plotted into histograms (right). The dashed circles show islet regions and the black arrowheads point out Ki67+ or TUNEL+ cells. Mouse number (n) is shown in parentheses. P () < 0.05, P () < 0.01 and P (*) < 0.001 are considered statistically significant ## PS1 decreases cell death and increases insulin secretion in β-cells To probe the action of PS1 in β-cells, we investigated the impact of PS1 on cell demise and function in Min6 cells, a murine β-cell line, in response to high glucose. Min6 cells were pre-incubated with cell medium, and the medium was supplemented with STG (10 μg/mL) and PS1 (0.2, 1, and 5 μg/mL) in the presence of 5 mM or 30 mM glucose. Flow cytometric data showed that control Min6 cells grown in the medium containing 5 mM glucose, had a basal level of cell death, $7.4\%$ of PI-positive cells (CTR, Fig. 6A). Glucose at 30 mM increased cell death to $18.6\%$ (HG, Fig. 6). STG failed to lower cell death ($17\%$) (STG, Fig. 6A). In contrast, PS1 at 0.2 μg/mL, 1 μg/mL, and 5 μg/mL reduced cell death to $12.8\%$, $8.8\%$, and $4\%$, respectively (PS1, Fig. 6A). Remarkably, PS1 at 5 μg/mL even reduced this death in Min6 cells by $3.4\%$ (from 7.4 to $4\%$). The data demonstrated that PS1 dose-dependently rescued Min6 cells from cell death as a result of high glucose. However, STG failed to rescue Min6 cells from cell death under the same conditions. Fig. 6Down-regulation of cell death and up-regulation of insulin secretion in Min6 cells by PS1. A Min6 cells were treated with 5 mM (CTR) or 30 mM glucose with PBS (HG), STG (10 µg/mL) and PS1 (0.2 µg/mL, 1 µg/mL, and 5 µg/mL). The cells were stained with propidium iodide (PI) and analyzed with flow cytometry. The percentage of PI-positive cells from 3 experiments was analyzed and is expressed as mean ± SD. B Min6 cells were stimulated with DMEM medium or the medium containing KCl (30 mM), STG (10 µg/mL) and PS1 (0.2 µg/mL, 1 µg/mL, and 5 µg/mL). The insulin level of their supernatants was quantified and replotted into histograms. The data from 3 experiments are expressed as mean ± SD. C Min6 cells were pre-treated with DMEM medium containing 30 mM. After washing, the cells were treated with high glucose (16.7 mM), STG (10 µg/mL) and PS1 (0.2 µg/mL, 1 µg/mL, and 5 µg/mL). The insulin level of their supernatants was quantified and replotted into histograms. The data from 3 experiments are expressed as mean ± SD. P () < 0.05, P () < 0.01 and P () < 0.001 were considered statistically significant We also examined the effect of PS1 on insulin release in Min6 cells. First, we tested the insulinotropic effect of PS1 in Min6 cells. As expected, there was a basal level of insulin in the supernatant of Min6 cells (CTR, Fig. 6B). Potassium chloride, a positive control, elevated the level of insulin in the supernatant of Min6 cells (KCL, Fig. 6B). STG did not alter the level of insulin in the supernatant of Min6 cells (STG, Fig. 6B). However, PS1 dose-dependently up-regulated the level of insulin in the supernatant of Min6 cells (PS1, Fig. 6B). In parallel, we also treated Min6 cells with 30 mM glucose with or without PS1. As expected, high glucose compromised the glucose stimulated insulin secretion (GSIS) in Min6 cells (CTR, Fig. 6C), However, STG failed to improve the GSIS in Min6 cells (STG, Fig. 6C). In remarkable contrast, PS1 dose-dependently improved the GSIS in Min6 cells (PS1, Fig. 6C). The data suggest that PS1 protected against β-cell dysfunction and death. ## PS1 down-regulates ROS production via the Pdia4/Ndufs3 and Pdia4/p22 axes in β cells To investigate whether PS1 exerted its action through ROS pathways, we first analyzed the amount of ROS in the mitochondria and cytosol of Min6 cells. Min6 cells had a basal level of mitochondrial ROS (CTR, Fig. 7A). High glucose augmented mitochondrial ROS in Min6 cells (HG, Fig. 7A). STG failed to diminish mitochondrial ROS in Min6 cells (STG, Fig. 7A). In sharp contrast, PS1 dose-dependently reduced mitochondrial ROS in Min6 cells (PS1, Fig. 7A). Similarly, we found that Min6 cells had a basal level of cytosolic ROS (CTR, Fig. 7B). High glucose increased cytosolic ROS in Min6 cells (HG, Fig. 7B). STG failed to diminish cytosolic ROS in Min6 cells (STG, Fig. 7B). In sharp contrast, PS1 dose-dependently reduced cytosolic ROS in Min6 cells (PS1, Fig. 7B). The data showed that PS1 reduced ROS in β-cells. Fig. 7Inhibition of the Pdia4/ETC C1 and Pdia4/Nox pathways in Min6 cells by PS1. ( A-B) Min6 cells were treated with 5 mM glucose (CTR) or 30 mM glucose with PBS (HG), STG (15 μg/mL), and PS1 (0.2 µg/mL, 1 µg/mL, and 5 µg/mL). The cells were then incubated with MitoGreen plus MitoSOX (A) and Hoechst 33,342 (Ho) plus CellROX (B) in the presence of glucose at 5 mM (CTR) and 30 mM (HG). The cells were visualized (left) and quantified (right). Scale bar: 5 µm. C The ETC C1 (left) activity of mitochondria isolated from control and PS1-treated Min6 cells A was measured using MitoCheck assays. D The Nox activity of the membrane fraction of control and PS1-treated Min6 cells A was measured using luminometric assays. Min6 cells that expressed Flag-Pdia4 plus Myc/Flag-tagged Ndufs3 (E) or p22 (F) were treated with PS1 (5 μg/mL) for 30 min. The cells were lysed and precipitated using anti-Pdia4 antibodies and protein G beads. Their total lysates (TL) and immunoprecipitates (IP) were analyzed with immunoblots using anti-Flag antibody. P () < 0.05, P () < 0.01 and P () < 0.001 were considered statistically significant. G Schema outlining the regulation of the Pdia4/Ndufs3 and Pdia4/p22 pathways. The intermolecular association of Pdia4 with Ndufs3 and p22phox increases the activity of ETC C1 and Nox. As a result, the excess ROS induces β-cell pathology and diabetes (top). On the other hand, Pdia4 inhibition by PS1 can disrupt the interplay of Pdia4 and Ndufs3 or p22, resulting in declined ROS production, β-cell failure and diabetes (bottom) To pinpoint the pathways that PS1 targeted, we checked the effect of PS1 on Pdia4 activity and the intermolecular interaction between Pdia4 and Ndufs3 or p22. The data showed that PS1 could inhibit the enzymatic activity of Pdia4 in Min6 cells (Figure S1) and, in turn, decreased the activity of ETC C1, Fig. 7C) and Nox (Fig. 7D). Further, immunoblotting data showed that PS1 intervened with the intermolecular interaction between Pdia4 and Ndufs3 (Fig. 7E) and that between Pdia4 and p22 (Fig. 7F). Taken together, the data suggested that PS1 down-regulated production of mitochondrial and cytosolic ROS via the Pdia4/Ndufs3 and Pdia4/p22 pathways in β-cells. A schematic model describing the pharmacological function and mechanism of PS1 is shown in Fig. 7G. Under hyperglycemia, nutrient overload up-modulated Pdia4 expression. This up-modulation increased the activity of Ndufs3 and p22 through intermolecular interplay and thus escalated the generation of ROS in β-cells. Eventually, the excessive ROS caused β-cell dysfunction and death and, subsequently, diabetes (Fig. 7G). On the other hand, PS1 reversed diabetes through reduced ROS production and β-cell pathology in diabetic animals (Fig. 7G). The data also unveiled the pharmacological action and mechanism of PS1 in β-cell pathology and diabetes. ## Discussion The db/db mouse, whose leptin receptor is mutated, is a commonly used model of type 2 diabetes. Our previous seminal paper showed the regulation of β-cell pathology and diabetes by Pdia4 in db/db mice [2]. This regulation was through the association of Pdia4 with Ndufs3 and p22, resulting in the activation of ETC C1 and Nox pathways and, hence, ROS production in β-cells [2]. One hit, 2-β-D-glucopyranosyloxy1-hydroxytrideca 5,7,9,11-tetrayne, was shown to inhibit Pdia4 activity with an IC50 of 358 μM. Consequently, it suppressed diabetes progression in db/db mice [2]. In this study, monotherapy and combinational therapy of PS1, a Pdia4-based drug candidate with an IC50 of 4 μM, could lessen β-cell failure and, in turn, diabetes development in db/db mice. Indeed, PS1 was more potent than the aforesaid hit. Mechanistically, PS1 down-regulated the ROS-generating pathways by disrupting the association of Pdia4 with Ndufs3 and p22, two important players of ETC C1 and Nox 1–4, respectively (Fig. 7). More importantly, we unveiled, for the first time, the pharmacological function of PS1 in endocrine β-cells, addressed the importance, molecular basis and therapeutic potential of PS1 in pathological process of β-cells and diabetes, and provided a novel molecular mechanism through which PS1 intervened with the ROS-generating machinery, ECT C1 and Nox, leading to shutting down ROS production. Oxidative stress is known to dictate β-cell physiology and pathology and diabetes [20]. Consistently, metabolic stress has been reported to elevate oxidative stress through the production of ROS, which dampens insulin release and insulin action during diabetes [21]. Akin to the previous findings [2], PS1 inhibited Pdia4 activity (Figure S1) and, thus, abolished the interplay between Pdia4 and Ndufs3 or p22 in β-cells (Fig. 7E, F). Disruption of Pdia4 might destabilize Ndufs3 or p22 and dampened ROS-generating ETC C1 and Nox pathways as published [2]. Accordingly, PS1 also reduced the activity of ETC C1 and Nox (Fig. 7C, D). As expected, PS1 reduced ROS content in the pancreatic islets of db/db mice (Fig. 4A, C) and in the cytosol and mitochondria of β-cells (Fig. 7A, B). In fact, the above data are in good agreement with several reports that have stated the escalated expression and/or activity of p22 and Ndufs3 in humans and animals of diabetes [22, 23]. Since Pdia4 is not an essential gene [2, 24], its specific inhibitor, PS1, seemed to have no noticeable adverse effects. On the contrary, PS1 was able to reverse diabetes, increase survival and prolong the longevity in db/db mice (Fig. 2). Overall, we believe the data presented herein move us one step further towards clinical use of the Pdia4-based therapy for β-cell pathogenesis and diabetes. Our previous publication demonstrated that Pdia4 knockout reduced cell death but not cell proliferation in the β-cells of db/db mice [2]. As a result, Pdia4 ablation accumulated functional β-cells and, in turn, reversed diabetes in diabetic animals. In good agreement with the publication on Pdia4 ablation, 16-week administration of PS1 alone and together with metformin normalized diabetes in db/db mice as shown by blood glucose, HbA1c, GTT, diabetic incidence, and water consumption (Figs. 1, 2, S2B). Of note, PS1 per se and in combination with metformin increased longevity of db/db mice by 29 and 34 weeks, respectively (Fig. 2C). The data suggested a good efficacy of PS1 in diabetes. Strikingly, the reversal rate of diabetes for PS1 per se and in combination with metformin was $86\%$ and $100\%$ in db/db mice even after 68-week drug withdrawal (Fig. 2A). Consequently, we found that PS1 reduced cell death but not cell proliferation in the β-cells of db/db mice (Fig. 5A, B). The data clearly demonstrated that PS1 diminished β-cell demise in diabetic animals. Flow cytometric data showed that in response to high glucose, PS1 dose-dependently protected against cell death in β-cells (Fig. 6A). Meanwhile, PS1 reduced the decline of c-peptide and insulin in the sera of db/db mice (Fig. 3A, B). Moreover, PS1 increased the HOMA-β index and decreased the HOMA-IR index in db/db mice (Fig. 3C, D). ELISA data also showed that PS1 enhanced insulin release in β-cells (Fig. 6B). More importantly, PS1 could reduce the deterioration of insulin release in β-cells pre-treated with high glucose (Fig. 6C). Overall, the data suggested that PS1 ameliorated β-cell function in diabetic animals. Likewise, PS1 decreased the expression level of Aldh1a3 (Fig. 5C), which was reported to participate in β-cell failure [25, 26]. The above data demonstrate the importance and efficacy of PS1, which targets Pdia4, to preserve functional β-cells in diabetes therapy. In this study, db/db mice developed diabetes, and also obesity. Accumulating evidence supports a causal relationship between obesity and ER stress in adipose tissue. Pdia4 has been reported to be a novel ER chaperone implicated in β-cell pathogenesis in diabetes [2]. However, the role of Pdia4 in obesity progression remains poorly understood. One report stated that Pdia4 regulated adipocytes by down-regulating adiponectin [27]. Consistently, serum Pdia4 was shown to be related to obesity, insulin sensitivity, and diabetes [28]. The adiponectin/leptin ratio was proposed as a functional marker of fat inflammation in human patients of diabetes [29]. In this sense, it is necessary to verify whether Pdia4 affects this ratio in db/db mice. In addition, moderate islet inflammation was thought to contribute to β-cell failure in type 2 diabetes. Alternatively activated macrophages facilitated a loss of β-cell identity in db/db mice [30]. Given the relationship between Pdia4 and obesity, it should be noted that other factors like aquaporins, a central player in fat metabolism, could be involved in the Pdia4 pathways [31]. Given the importance of Pdia4 for obesity, it will be absolutely necessary to further explore the function and mechanism of PS1 in obesity. ## Supplementary Information Below is the link to the electronic supplementary material. Supplementary file1 (PDF 309 kb) ## References 1. 1.IDF Diabetes Atlas tEC ed (2021) IDF Diabetes Atlas. International Diabetes Federation 2. Kuo TF, Hsu SW, Huang SH, Chang CL, Feng CS, Huang MG. Pdia4 regulates beta-cell pathogenesis in diabetes: molecular mechanism and targeted therapy. EMBO Mol Med* (2021.0) 13 e11668. DOI: 10.15252/emmm.201911668 3. Wysham C, Shubrook J. Beta-cell failure in type 2 diabetes: mechanisms, markers, and clinical implications. Postgrad Med (2020.0) 132 676-686. DOI: 10.1080/00325481.2020.1771047 4. Defronzo RA. Banting Lecture. From the triumvirate to the ominous octet: a new paradigm for the treatment of type 2 diabetes mellitus. Diabetes (2009.0) 58 773-795. DOI: 10.2337/db09-9028 5. Leahy JL, Hirsch IB, Peterson KA, Schneider D. Targeting beta-cell function early in the course of therapy for type 2 diabetes mellitus. J Clin Endocrinol Metab (2010.0) 95 4206-4216. DOI: 10.1210/jc.2010-0668 6. Herbert TP, Laybutt DR. A reevaluation of the role of the unfolded protein response in Islet dysfunction: maladaptation or a failure to adapt?. Diabetes (2016.0) 65 1472-1480. DOI: 10.2337/db15-1633 7. Lei XG, Vatamaniuk MZ. Two tales of antioxidant enzymes on beta cells and diabetes. Antioxid Redox Signal (2011.0) 14 489-503. DOI: 10.1089/ars.2010.3416 8. Newsholme P, Rebelato E, Abdulkader F, Krause M, Carpinelli A, Curi R. Reactive oxygen and nitrogen species generation, antioxidant defenses, and beta-cell function: a critical role for amino acids. J Endocrinol (2012.0) 214 11-20. DOI: 10.1530/JOE-12-0072 9. Weaver JR, Grzesik W, Taylor-Fishwick DA. Inhibition of NADPH oxidase-1 preserves beta cell function. Diabetologia (2015.0) 58 113-121. DOI: 10.1007/s00125-014-3398-2 10. Evans JL, Maddux BA, Goldfine ID. The molecular basis for oxidative stress-induced insulin resistance. Antioxid Redox Signal (2005.0) 7 1040-1052. DOI: 10.1089/ars.2005.7.1040 11. Bindokas VP, Kuznetsov A, Sreenan S, Polonsky KS, Roe MW, Philipson LH. Visualizing superoxide production in normal and diabetic rat islets of Langerhans. J Biol Chem (2003.0) 278 9796-9801. DOI: 10.1074/jbc.M206913200 12. Harrity T, Farrelly D, Tieman A, Chu C, Kunselman L, Gu L. Muraglitazar, a novel dual (alpha/gamma) peroxisome proliferator-activated receptor activator, improves diabetes and other metabolic abnormalities and preserves beta-cell function in db/db mice. Diabetes (2006.0) 55 240-248. DOI: 10.2337/diabetes.55.01.06.db05-0648 13. Leung KK, Leung PS. Effects of hyperglycemia on angiotensin II receptor type 1 expression and insulin secretion in an INS-1E pancreatic beta-cell line. JOP (2008.0) 9 290-299. PMID: 18469441 14. Son J, Ding H, Farb TB, Efanov AM, Sun J, Gore JL. BACH2 inhibition reverses beta cell failure in type 2 diabetes models. J Clin Investig (2021.0) 131 e153876. DOI: 10.1172/JCI153876 15. Ardestani A, Paroni F, Azizi Z, Kaur S, Khobragade V, Yuan T. MST1 is a key regulator of beta cell apoptosis and dysfunction in diabetes. Nat Med (2014.0) 20 385-397. DOI: 10.1038/nm.3482 16. Sacco F, Seelig A, Humphrey SJ, Krahmer N, Volta F, Reggio A. Phosphoproteomics reveals the GSK3-PDX1 axis as a key pathogenic signaling node in diabetic islets. Cell Metab (2019.0) 29 1422-1432. DOI: 10.1016/j.cmet.2019.02.012 17. Marcheva B, Weidemann BJ, Taguchi A, Perelis M, Ramsey KM, Newman MV. P2Y1 purinergic receptor identified as a diabetes target in a small-molecule screen to reverse circadian beta-cell failure. Elife (2022.0) 11 e75132. DOI: 10.7554/eLife.75132 18. Lupse B, Ardestani A, Annamalai K, Maedler K. PHLPP1/2 phosphatases are keymediators of beta cell failure in type 2 diabetes. Diabetologia (2019.0) 62 S35 19. Tsai PY, Ka SM, Chang JM, Chen HC, Shui HA, Li CY. Epigallocatechin-3-gallate prevents lupus nephritis development in mice via enhancing the Nrf2 antioxidant pathway and inhibiting NLRP3 inflammasome activation. Free Radic Biol Med (2011.0) 51 744-754. DOI: 10.1016/j.freeradbiomed.2011.05.016 20. Dos Santos JM, Tewari S, Mendes RH. The role of oxidative stress in the development of diabetes mellitus and its complications. J Diabetes Res (2019.0) 2019 4189813. PMID: 31192263 21. Tangvarasittichai S. Oxidative stress, insulin resistance, dyslipidemia and type 2 diabetes mellitus. World J Diabetes (2015.0) 6 456-480. DOI: 10.4239/wjd.v6.i3.456 22. Huang X, Sun M, Li D, Liu J, Guo H, Dong Y. Augmented NADPH oxidase activity and p22phox expression in monocytes underlie oxidative stress of patients with type 2 diabetes mellitus. Diabetes Res Clin Pract (2011.0) 91 371-380. DOI: 10.1016/j.diabres.2010.12.026 23. Wu J, Luo X, Thangthaeng N, Sumien N, Chen Z, Rutledge MA. Pancreatic mitochondrial complex I exhibits aberrant hyperactivity in diabetes. Biochem Biophys Rep (2017.0) 11 119-129. PMID: 28868496 24. Almeida S, Zhou L, Gao FB. Progranulin, a glycoprotein deficient in frontotemporal dementia, is a novel substrate of several protein disulfide isomerase family proteins. PLoS One (2011.0) 6 e26454. DOI: 10.1371/journal.pone.0026454 25. Efrat S. Beta-cell dedifferentiation in type 2 diabetes: concise review. Stem Cells (2019.0) 37 1267-1272. DOI: 10.1002/stem.3059 26. Kim-Muller JY, Fan J, Kim YJ, Lee SA, Ishida E, Blaner WS. Aldehyde dehydrogenase 1a3 defines a subset of failing pancreatic beta cells in diabetic mice. Nat Commun (2016.0) 7 12631. DOI: 10.1038/ncomms12631 27. Su SC, Chien CY, Chen YC, Chiang CF, Lin FH, Kuo FC. PDIA4, a novel ER stress chaperone, modulates adiponectin expression and inflammation in adipose tissue. BioFactors (2022.0) 48 1060-1075. DOI: 10.1002/biof.1872 28. Su SC, Hung YJ, Lin FH, Hsieh CH, Lu CH, Chien CY. Circulating protein disulfide isomerase family member 4 is associated with type 2 diabetes mellitus, insulin sensitivity, and obesity. Acta Diabetol (2022.0) 59 1001-1009. DOI: 10.1007/s00592-022-01892-1 29. Frühbeck G, Catalán V, Rodríguez A, Ramírez B, Becerril S, Salvador J. Adiponectin-leptin ratio is a functional biomarker of adipose tissue inflammation. Nutrients (2019.0) 11 454. DOI: 10.3390/nu11020454 30. Chan JY, Lee K, Maxwell EL, Liang C, Laybutt DR. Macrophage alterations in islets of obese mice linked to beta cell disruption in diabetes. Diabetologia (2019.0) 62 993-999. DOI: 10.1007/s00125-019-4844-y 31. Fruhbeck G. Obesity: aquaporin enters the picture. Nature (2005.0) 438 436-437. DOI: 10.1038/438436b
--- title: Comprehensive Proteomics Analysis Identifies CD38-Mediated NAD+ Decline Orchestrating Renal Fibrosis in Pediatric Patients With Obstructive Nephropathy authors: - Yuandong Tao - Jifeng Wang - Xuexue Lyu - Na Li - Dong Lai - Yuanyuan Liu - Xingyue Zhang - Pin Li - Shouqing Cao - Xiaoguang Zhou - Yang Zhao - Lifei Ma - Tian Tao - Zhichun Feng - Xiubin Li - Fuquan Yang - Huixia Zhou journal: 'Molecular & Cellular Proteomics : MCP' year: 2023 pmcid: PMC10025283 doi: 10.1016/j.mcpro.2023.100510 license: CC BY 4.0 --- # Comprehensive Proteomics Analysis Identifies CD38-Mediated NAD+ Decline Orchestrating Renal Fibrosis in Pediatric Patients With Obstructive Nephropathy ## Body Obstructive nephropathy is a leading cause of kidney injury in infants and children, which is frequently induced by hydronephrosis in pediatric patients with ureteropelvic junction obstruction (UPJO) [1, 2, 3]. Prolonged obstructive hydronephrosis leads to chronic kidney disease (CKD) and renal fibrosis, characterized by an excessive extracellular matrix accumulation and the progressive loss of kidney function [4]. To date, effective antifibrotic therapies are still lacking. Therefore, it is critical to expanding our understanding of obstructive nephropathy to facilitate the development of new treatments. The current understanding of the pathogenesis of obstruction-induced kidney injury is mostly derived from animal experiments, for instance, the unilateral ureteral obstruction (UUO) model [5]. It is believed that obstructive hydronephrosis is associated with a higher intrapelvic and intratubular hydrostatic pressure, which stimulates apoptosis and necrosis of tubular cells and the resultant release of damage-associated molecular patterns [6]. These lead to the infiltration of immune cells, including macrophages and neutrophils, and the synthesis of proinflammatory cytokines and profibrotic factors, such as interleukin-1 beta (IL-1β), tumor necrosis factor-alpha (TNF-α), transforming growth factor-beta (TGF-β), and platelet-derived growth factor (PDGF). The profibrotic microenvironment promotes the activation of myofibroblasts, the major source of extracellular matrix, thus driving the development of renal fibrosis [7]. Although these findings have provided important insights into obstructive nephropathy, the etiology of hydronephrosis-induced kidney injury in pediatric patients is largely unknown, and only large-scale, unbiased discovery experiments can enable the assessment of multiple biological processes, networks, and regulators simultaneously. Nicotinamide adenine dinucleotide (NAD+) is an essential cofactor for various metabolic processes, such as glycolysis, the Krebs cycle, and fatty acid oxidation (FAO) [8]. The cellular NAD+ can be synthesized from the essential amino acid tryptophan via the de novo pathway as well as from dietary nicotinic acid via the Preiss–Handler pathway. Alternatively, NAD+ can be produced from different forms of vitamin B3 via the salvage pathway [9, 10]. The NAD+ consumption is catalyzed by three classes of enzymes: sirtuins (SIRTs), poly (ADP-ribose) polymerases (PARPs), and cyclic ADPribose synthetases (CD38) [11]. Interestingly, recent studies indicate that NAD+ is implicated in the development of kidney diseases [8]. A rapid decrease in NAD+ levels is observed in acute kidney injury, which is partially driven by impaired do novo NAD+ biosynthesis [12, 13]. Recovery of NAD+ levels via supplementation with NAD+ precursors, such as nicotinamide mononucleotide or nicotinamide (NAM), protects against kidney damage in acute kidney injury models [12, 14, 15]. Moreover, several studies in animals and clinical observations suggest a linkage between NAD+ metabolism and the development of CKDs [8]. However, few data are available on the details of NAD+ metabolism and its roles and underlying mechanisms in CKDs, including obstruction-induced renal fibrosis [10]. The multifunctional NADase CD38 is a kind of ADP-ribosyl cyclase that degrades NAD+ and modulates cellular NAD+ homeostasis [16]. It is most highly expressed in immune cells, such as macrophages, T cells, B cells, and monocytes [17]. CD38 has a central role in age-related NAD+ decline in mammals, depending on its ectoenzyme or endoenzyme activity [18]. Of note, recent studies reported that CD38 deficiency induces autoimmune characteristics and kidney damage in 16-month-old mice [19] and CD38 inhibition by apigenin reduces renal injury in diabetic rats through the restoration of the NAD+/NADH ratio [20]. These findings suggest tantalizing links of CD38 and NAD+ to kidney diseases. Nonetheless, the roles and underlying mechanisms of CD38 in obstructive nephropathy are completely unknown. In the present study, we analyzed the global proteome of obstructed kidneys from patients and UUO mice. We uncovered the dysregulation of NAD+ metabolism in obstructive nephropathy and found that NADase CD38 promoted obstruction-induced renal fibrosis and kidney inflammation. ## Abstract Obstructive nephropathy is one of the leading causes of kidney injury and renal fibrosis in pediatric patients. Although considerable advances have been made in understanding the pathophysiology of obstructive nephropathy, most of them were based on animal experiments and a comprehensive understanding of obstructive nephropathy in pediatric patients at the molecular level remains limited. Here, we performed a comparative proteomics analysis of obstructed kidneys from pediatric patients with ureteropelvic junction obstruction and healthy kidney tissues. Intriguingly, the proteomics revealed extensive metabolic reprogramming in kidneys from individuals with ureteropelvic junction obstruction. Moreover, we uncovered the dysregulation of NAD+ metabolism and NAD+-related metabolic pathways, including mitochondrial dysfunction, the Krebs cycle, and tryptophan metabolism, which led to decreased NAD+ levels in obstructed kidneys. Importantly, the major NADase CD38 was strongly induced in human and experimental obstructive nephropathy. Genetic deletion or pharmacological inhibition of CD38 as well as NAD+ supplementation significantly recovered NAD+ levels in obstructed kidneys and reduced obstruction-induced renal fibrosis, partially through the mechanisms of blunting the recruitment of immune cells and NF-κB signaling. Thus, our work not only provides an enriched resource for future investigations of obstructive nephropathy but also establishes CD38-mediated NAD+ decline as a potential therapeutic target for obstruction-induced renal fibrosis. ## Graphical Abstract ## Highlights •Proteomics uncovered a profibrotic and inflammatory phenotype in obstructed kidneys.•Obstruction induced a global metabolic dysregulation and an aberrant NAD+ metabolism.•CD38 was strongly induced in human and experimental obstructive nephropathy.•CD38 deletion or inhibition mitigated obstruction-induced renal fibrosis. ## In Brief Obstructive nephropathy is a leading cause of kidney injury in infants and children. In this work, we performed comparative proteomics of control and obstructed kidneys from human and experimental obstructive nephropathy and uncovered the aberrant NAD+ metabolism, which was partially induced by CD38 upregulation. Deletion or inhibition of CD38 reduced obstruction-associated renal fibrosis and inflammation. These findings emphasized the therapeutic potential of CD38 and NAD+ metabolism in obstructive nephropathy. ## Patient Samples The human kidney samples used in this study were obtained from The Seventh Medical Center of Chinese PLA General Hospital, with the approval of the Research Ethics Committee of the hospital. Written informed consent was provided by legal parents. Eleven obstructed kidneys were from pediatric patients with late-presented hydronephrosis induced by UPJO during nephrectomy. Eight control kidneys were sampled from the tumor-free kidney cortex 5 cm away from the tumor region of patients with nephroblastoma during radical nephrectomy. Before nephrectomy in both UPJO and tumor cases, the kidney vessels were clamped within 15 min. The operations were performed by a single skilled surgeon. Surgically resected kidney tissues were frozen in liquid nitrogen for storage before use. The clinical information of patients was shown in Table 1.Table 1Clinical information of patientsCharacteristicsObstructed kidneys ($$n = 11$$)Control kidneys ($$n = 8$$)p valueAge (months) (mean) [s.d.]25.85 (22.53)48.23 (44.26)0.220Gender0.040 Male36 Female82Degree of hydronephrosis0 008 100 200 300 4110DiagnosisHydronephrosisNephroblastomaOperation side0.599 Left42 Right76BUN (mM) (mean) [s.d.]4.58 (1.86)4.00 (1.39)0.473Cre (mM) (mean) [s.d.]34.50 (6.64)38.78 (7.62)0.210Abbreviations: BUN, serum urea nitrogen; Cre, creatinine. ## Mice and UUO Model C57BL/6 wildtype mice were purchased from Charles River in Beijing (Vital River). Cd38−/− mice on a C57BL/6 background were kindly provided by Dr Xingyue Zhang (The Seventh Medical Center of Chinese PLA General Hospital). All animal experimental procedures were approved by the Institutional Animal Care and Utilization Committees of the Chinese PLA General Hospital. All mice were maintained in a specific pathogen-free condition. For the UUO model, male mice aged between 7 and 8 weeks were used. Generally, a median abdominal incision was performed after anesthetization and the left ureter was double ligated. The sham group underwent the same procedure except for the ureteral ligation. Mice were sacrificed 7 days (unless specified otherwise) after surgery and kidneys were then harvested. For the inhibition of CD38, compound 78c (S8960, Selleck) was administrated via oral gavage for six consecutive days (200 μg/mouse/day). NAD+ (N7004, Sigma) was supplemented by peritoneal injection (100 μg/mouse/day). ## Sample Preparation The kidney tissues were disrupted by using a Bertin homogenizer on ice in lysis buffer (8 M urea/0.1 M Tris-HCl, pH 8.0) containing 1× Protease Inhibitor Cocktail (Roche). After centrifugation, the extracted proteins were reduced with 10 mM DTT for 2 h at room temperature followed by alkylation with 20 mM iodoacetamide for 30 min in the dark. Samples were then digested with trypsin (1:50) at 37 °C overnight. The digestion was desalted on an OASIS HLB column, and peptides eluted with $60\%$ acetonitrile were lyophilized via vacuum centrifugation and dissolved in $0.1\%$ formic acid before mass spectrometry (MS) data acquisition. ## Data-Independent Acquisition Mass Spectrometry Data Acquisition All nano-liquid chromatography tandem mass spectrometry experiments were performed on Orbitrap Eclipse (Thermo Scientific) equipped with an Easy n-LC 1200 HPLC system (Thermo Scientific). The peptides were loaded onto a 100 μm id × 2 cm fused silica trap column packed in-house with reversed-phase silica (Reprosil-Pur C18 AQ, 5 μm, Dr. Maisch GmbH) and then separated on a 75 μm id × 25 cm C18 column packed with reversed-phase silica (Reprosil-Pur C18 AQ, 1.9 μm, Dr. Maisch GmbH). The peptides bounded on the column were eluted with a 103-min linear gradient. Solvent A consisted of $0.1\%$ formic acid in water solution, and solvent B consisted of $80\%$ acetonitrile and $0.1\%$ formic acid. The segmented gradient was 4 to $11\%$ B, 4 min; 11 to $21\%$ B, 28 min; 21 to $30\%$ B, 29 min; 30 to $42\%$ B, 27 min; 42 to $99\%$ B, 5 min; $99\%$ B, 10 min at a flow rate of 300 nl/min. The MS analysis was performed with Orbitrap Eclipse mass spectrometer (Thermo Scientific). With the data-independent acquisition mode, the MS data were acquired at a high resolution 120,000 (m/z 200) across the mass range of 400 to 1210 m/z. The target value was 4.00E+05 with a maximum injection time of 50 ms. One full scan was followed by 40 windows with an isolation width of 16 m/z for fragmentation in the Ion Routing Multipole with HCD normalized collision energy of $30\%$. Tandem mass spectrometry spectra were acquired at resolution of 30,000 at m/z 200 across the mass range of 200 to 2000 m/z. The target value was 4.00E+05 with a maximum injection time of 50 ms. For the nanoelectrospray ion source setting, the spray voltage was 2.0 kV; no sheath gas flow; the heated capillary temperature was 320 °C. ## Data-Independent Acquisition Data Analysis The DIA raw data from Orbitrap Eclipse were analyzed using Spectronaut version 14 (Biognosys) with the “DirectDIA” mode for protein identification and quantification. The UniProt human or mouse proteome database (download date 2021-07-05) was used for searching the data from kidney samples, and the total number of database entries searched for human and mouse was 20,371 and 55,341, respectively. The most important searching parameters were set as the default settings: trypsin was selected as enzyme and two missed cleavages were allowed for searching; the mass tolerance of MS1 and MS2 was set as correction factor 1; cysteine carbamidomethylation was specified as fixed modification; the methionine oxidation and acetylation of protein N-term were chosen as variable modifications. False discovery rate <$1\%$ was set for peptide spectrum matches, peptides, and proteins identification. The data were filtered by Qvalue, and the “Global Normalization” was set as “Median” with enabled cross run normalization. Statistical analyses of proteins were performed in Spectronaut. Two-sample t test was used for the calculation of p value, and the resulting p value was adjusted using the Benjamini–Hochberg method. Proteins with p value <0.05, q value <0.05, and fold change >2 or <0.5 were considered as up- or downregulated differentially expressed proteins. ## Western Blots Kidney protein extracts were prepared according to standard protocols [21]. Cell lysates were separated by $10\%$ SDS-PAGE and transferred to polyvinylidene difluoride membranes (Millipore). The following antibodies were used (all from Cell Signaling Technology unless specified otherwise): monoclonal mouse anti-mouse/human CD38 (sc-374650, Santa Cruz), monoclonal mouse anti-mouse GAPDH (5174T), monoclonal rabbit anti-mouse/human SMAD2 (5339T) and phospho-SMAD2 (18338T), monoclonal rabbit anti-mouse/human NF-κB p65 (8242T) and phospho-NF-κB p65 (3033T). ## Immunohistochemistry Human and mouse kidneys were fixed in $4\%$ paraformaldehyde and embedded with paraffin. The kidney sections were stained with hematoxylin and eosin, Masson's trichrome, and rabbit anti-mouse αSMA antibody (19245, Cell Signaling Technology). Images of kidney slides were obtained on a Nano Zoomer Slide Scanner (Hamamatsu Photonics). The percentages of collagen-positive areas and αSMA-positive areas were quantified by ImageJ software. ## Real-Time Quantitative PCR Mouse kidneys were homogenized and total RNA was extracted using Trizol according to the manufacturer’s protocol (Thermo Fisher). Complementary DNA was generated using a Reverse Transcription kit (Takara). Real-time quantitative PCR was performed using the iCycler iQ5 Real-Time PCR detection system (Bio-Rad). The expression of the target gene was normalized to the expression of the housekeeping gene, Gapdh. Relative gene expression was calculated using the standard 2−ΔΔCt method. The primers used were shown in Table 2.Table 2Primers for real-time quantitative PCRGenesForwardReversemGapdhATCTTCTTGTGCAGTGCCAGCGTTGATGGCAACAATCTCCACmTgfb1CGCAACAACGCCATCTATGAACTGCTTCCCGAATGTCTGAmIl1bTGTAATGAAAGACGGCACACCTCTTCTTTGGGTATTGCTTGGmKim1CTATGTTGGCATCTGCATCGAAGGCAACCACGCTTAGAGAmNgalGATGAACTGAAGGAGCGATTCTCGGTGGGAACAGAGAAAAC ## Kidney Leukocyte Isolation and Flow Cytometry Analysis Mice under indicated conditions were anesthetized, and kidneys were harvested and cut into pieces. Then kidney tissues were digested in a buffer (HBSS supplemented with $0.05\%$ collagenase I and 2 mM CaCl2) under 37 °C for 25 min. After digestion, the kidney tissues were filtered through a 70-μm nylon mesh. The cell suspension was centrifuged at 500g for 5 min, and the resulting cell suspension was treated with Fcγ receptor blocker (101320, BioLegend) for 10 min followed by incubating with the following fluorescent antibodies (all from BioLegend): CD45 BV421 [103134], CD11b FITC [101206], Ly6G APC/Cyanine7 [127624], Ly6C PE [128008], F$\frac{4}{80}$ APC [123116], 7AAD [420404], and CD38 PE/Cyanine7 [102717]. Flow cytometry was performed on a FACSCanto II (BD Biosciences), and data were analyzed by FlowJo software 10.4. ## Renal NAD+ Detection NAD+ levels in mouse kidney tissues were measured using a NAD/NADH quantification kit (MAK037, Sigma) according to the manufacturer’s protocol. ## Experimental Design and Statistical Rationale The study aimed to obtain a comprehensive understanding of human obstructive nephropathy at the proteome level. The experimental design is shown in Graphical Abstract. Eleven human obstructed kidneys and eight control kidneys were used to perform the proteomic study. Mouse kidney proteomics was performed with six sham and six UUO kidneys. The number of replicates and biological and statistical methods used for analysis were described in the figure legends. All tissues were lysed and digested in parallel. Data were presented as mean ± SEM. Statistical analyses were performed with GraphPad Prism version 6.0c or R version 4.1. Two-tailed Student’s t test was used for comparisons between two groups. Spearman's rank correlation was used to evaluate relationships between two variables. A p value less than 0.05 was considered significant. The receiver operating characteristic (ROC) curve analysis was performed in the pROC package. ## Proteomic Profiling of Human Obstructed Kidneys The study collected 11 obstructed kidneys (Ob) from pediatric patients with hydronephrosis and eight control kidneys (Ctr) of peritumor tissues from pediatric patients with nephroblastoma (Table 1). Histological examination revealed tubule atrophy, urinary cast, and leukocyte infiltration in obstructed kidneys (Fig. 1A). Moreover, the obstructed kidneys showed excessive collagen deposition (Fig. 1B) and an increased number of αSMA-positive cells compared with controls (Fig. 1C). These results indicate that sustained ureteral obstruction led to kidney injury and renal fibrosis. Fig. 1Proteomic profiling of human obstructed kidneys. A–C, representative images of hematoxylin & eosin (H&E) staining (A), Masson's trichrome staining (B), and αSMA immunohistochemistry staining (C) of kidney sections from control and patients with obstructive nephropathy. The scale bars represent: (A and B) upper panels 400 μm, lower panels 20 μm, (C) 100 μm. D, protein numbers identified by data-independent acquisition mass spectrometry in kidney samples obtained from control and patients with obstructive nephropathy. E, heatmap indicating all proteins identified by data-independent acquisition mass spectrometry in kidneys from control and patients with obstructive nephropathy. F, principle component analysis of control kidney proteomes ($$n = 8$$) and obstructed kidney proteomes ($$n = 11$$). To obtain an extensive molecular understanding of human obstructive nephropathy, a DIA-MS approach was used to perform the proteomic study. Proteomics measurement of all human kidney samples resulted in a total of 6258 proteins with an average of 5780 proteins per sample (Fig. 1, D and E). Principle component analysis demonstrated a clear boundary between the two proteomes (Fig. 1F), indicating an abnormal proteomic landscape in the kidneys from patients with obstructive nephropathy. ## Proteomic Landscape of Human Obstructed Kidneys Next, differentially expressed proteins between control and obstructed kidneys were identified. Compared with the control, 614 proteins were upregulated and 855 proteins were downregulated (p value <0.05, q value <0.05, and fold change >2 or <0.5) in human obstructive nephropathy (Fig. 2A). Functional enrichment analyses were performed by using Metascape [22] to identify the altered biological processes and signaling pathways in human obstructed kidneys. The upregulated proteins were significantly enriched in terms of collagen formation and immune response (Fig. 2B), which was consistent with the histological results (Fig. 1, A–C). Several collagens were dramatically increased in kidneys from patients with obstructive nephropathy, such as COL1A$\frac{1}{2}$, COL6A1, COL5A2, and COL3A1 (Fig. 2A and supplemental Fig. S1A). POSTN [23], a previously demonstrated signature protein for renal fibrosis, exhibited a 5-fold increase in obstructed kidneys (Fig. 2A). In line with previous studies demonstrating that inflammation is a critical contributor to renal fibrosis [24, 25], our study showed significant enrichment of neutrophil response, phagocytosis, cytokine signaling, and interferon signaling in the upregulated proteins (Fig. 2B and supplemental Fig. S1, B and C). Furthermore, gene set enrichment analysis revealed substantial upregulation of collagen fibril organization and collagen formation, as well as a significant increase in TGF-β signaling and Wnt signaling in obstructed kidneys (Fig. 2C), which are strongly associated with the development of renal fibrosis [26]. Supporting this, kidneys from individuals with obstructive nephropathy showed increased phosphorylation of SMAD2 (supplemental Fig. S1D). Taken together, these results indicate that prolonged ureteral obstruction induces a fibrotic and inflammatory phenotype in human kidneys. Fig. 2Proteomic landscape of human obstructed kidneys. A, volcano plot generated by differential analysis of the proteomic profiles of obstructed kidneys versus control kidneys. Significantly upregulated proteins were shown as blue dots and downregulated proteins were shown as pink dots (p value <0.05, q value <0.05, and fold change >2 or <0.5). B, top 20 most significant terms enriched by upregulated proteins in obstructed kidneys. C, gene set enrichment analysis showing pathways significantly upregulated in kidneys from patients with obstructive nephropathy. D, top 20 most significant terms enriched by downregulated proteins in obstructed kidneys. E, gene set enrichment analysis showing pathways dramatically downregulated in kidneys from patients with obstructive nephropathy. F, receiver operating characteristic curve analysis was performed and proteins with area under ROC curve = $100\%$ were used for protein–protein interaction analysis and Gene Ontology enrichment analysis. However, the downregulated proteins in obstructed kidneys were significantly enriched in the biological processes of proximal tubule transport and proton transmembrane transport, suggesting an impairment of renal function induced by ureteral obstruction (Fig. 2D). Furthermore, proteomics revealed extensive metabolic reprogramming events in obstructed kidneys. Biological processes including the Krebs cycle, mitochondrion organization, carbohydrate metabolic process, and tryptophan metabolism were remarkably enriched in the downregulated proteins (Fig. 2D). Meanwhile, gene set enrichment analysis revealed dramatic downregulation of mitochondrion organization and the Krebs cycle in kidneys from patients with obstructive nephropathy (Fig. 2E and supplemental Fig. S2). We also observed a dramatic downregulation of FAO in obstructed kidneys (Fig. 2E), validating the previous study showing that defective FAO in renal tubular epithelial cells has a key role in kidney fibrosis development [27]. Then, we performed ROC curves analysis on the differentially expressed proteins to identify the most discriminant proteins (area under the ROC curve = $100\%$) between control and obstructed kidneys. Interestingly, proteins with area under the ROC curve = $100\%$ were found to be mainly enriched in oxidative phosphorylation and mitochondrial function (Fig. 2F), suggesting that mitochondrial dysfunction might be a hallmark of obstruction-induced kidney injury. These results were consistent with the previous findings that mitochondrial abnormalities are critical features in the pathogenesis of several kidney diseases [28, 29]. Collectively, these data indicate that ureteral obstruction leads to extensive metabolic alterations and renal dysfunction in human kidneys. To reproduce the phenotype in animal experiments, we took advantage of the UUO mouse model, a well-established model to study obstructive nephropathy and renal fibrosis [30]. The proteomics revealed tremendous changes between sham and UUO kidneys (supplemental Fig. S3, A and B). Similarly to human proteomics, substantial enrichment of extracellular matrix organization and inflammation were identified in the upregulated proteins (p value <0.05, q value <0.05, and UUO versus sham fold change >2 or <0.5) (supplemental Fig. S3C). Moreover, we detected a global dysregulation of cellular metabolism in mouse fibrotic kidneys as well, such as mitochondrial organization, the Krebs cycle, and fatty acid metabolism (supplemental Fig. S3D). Therefore, the UUO mouse model replicated the phenotype of human obstructive nephropathy proteomically. ## Aberrant NAD+ Metabolism in Human and Mouse Obstructed Kidneys Since the above-identified metabolic dysregulations, including the Krebs cycle, mitochondrial dysfunction, oxidative phosphorylation, and tryptophan metabolism (Fig. 2E), are always associated with NAD+ levels [10], we asked whether NAD+ metabolism altered during obstructive nephropathy. Strikingly, by quarrying STRING database we found that enriched Gene Ontology molecular functions in the downregulated proteins were frequently related to NAD+ metabolism, in both human and experimental obstructive nephropathy (Fig. 3, A and B).Fig. 3Aberrant NAD+metabolism in human and mouse obstructed kidneys. A and B, top 20 most significant terms enriched by downregulated proteins in obstructed kidneys from patients with obstructive nephropathy (A) and UUO mice (B) by quarrying STRING database. C, schematic of cellular NAD+ metabolism showing key enzymes catalyzing NAD+ production and consumption. D, protein levels of CD38 in kidneys from control and patients with obstructive nephropathy. E, protein levels of CD38 in mouse control and obstructed kidneys. F and G, Spearman's rank correlation between the expression of CD38 and fibrosis markers in human (F) and mouse (G) kidney proteomics. H, tissue levels of NAD+ in kidneys from mice subjected to sham or UUO for 7 days. Representative data of three independent experiments. The results represent mean ± SEM. ∗$p \leq 0.01$, ∗∗$p \leq 0.01$, ∗∗∗$p \leq 0.001.$ p values were calculated by two-tailed Student’s t test (D, E and H) and Spearman's rank correlation analysis (F and G). UUO, unilateral ureteral obstruction. To determine if NAD+ homeostasis altered during the development of obstructive nephropathy, we then focused on the analysis of NAD+ metabolic enzymes (Fig. 3C). Several key enzymes catalyzing NAD+ de novo biosynthesis were dramatically decreased in human and mouse obstructed kidneys, such as HAAO and QPRT (supplemental Fig. S4). NAPRT and NMNAT3, mediating NAD+ synthesis via the salvage pathway, were significantly downregulated in patients as well as in UUO mice (supplemental Fig. S4). As for the NAD+ consumer, SIRT3 and SIRT5 were downregulated in both human and mouse obstructed kidneys, while SIRT2 slightly decreased only in patients (supplemental Fig. S4). PARPs showed different expression patterns between patients and mice. The human PARP4 was downregulated, whereas mouse PARP4 was upregulated in obstructed kidneys (supplemental Fig. S4). The proteomics identified unchanged expression of human PARP9, PARP10, and PARP14; however, mouse PARP9 and PARP10 were significantly increased in UUO kidneys (supplemental Fig. S4). Notably, CD38, the major NADase in mammals, was strongly induced in human and experimental obstructive nephropathy (Fig. 3, D and E). Moreover, CD38 was significantly correlated with PDGFRβ in human and mouse obstructed kidneys ($r = 0.66$, $p \leq 0.01$; $r = 0.72$, $p \leq 0.05$, respectively) (Fig. 3, F and G). PDGFRβ is a potent marker of myofibroblasts, which are the primary source of extracellular matrix in renal fibrosis [31]. CD38 was also positively correlated with CTNNB1 and SMAD2 in UUO mice, the key downstream component of Wnt and TGF-β signaling, respectively [32] (Fig. 3G). Therefore, the proteomic study reveals an aberrant NAD+ metabolism and a fibrosis-associated upregulation of CD38 in obstructive nephropathy. To determine whether dysregulated NAD+ metabolism resulted in altered NAD+ content in kidneys, we then assessed renal NAD+ levels in UUO mice. As expected, a significant decrease in NAD+ levels was observed in UUO kidneys (Fig. 3H), demonstrating that dysregulated NAD+ metabolism diminished renal NAD+ levels. ## The Elevated CD38 Contributed to NAD+ Decline in Obstructed Kidneys Given that CD38 was the major NADase that was significantly upregulated in both patients and UUO mice (Fig. 3, D and E) and CD38 was positively correlated with fibrosis markers (Fig. 3, F and G), we sought to determine the expression patterns of CD38 and whether CD38 contributed to NAD+ decline in obstructive nephropathy. In line with proteomics, Western blots confirmed the elevated levels of CD38 in human and mouse obstructed kidneys (Fig. 4, A and B). Healthy kidneys barely expressed CD38, whereas immunohistochemistry analysis of kidneys from patients with obstructive nephropathy showed accumulated CD38-expressing cells, which were mainly located in the renal interstitium but not in tubular cells or glomeruli (Fig. 4C). Moreover, CD38-positive cells were dramatically increased in UUO kidneys as well (Fig. 4D). To further characterize the expression patterns of CD38, we utilized a single cell RNA sequencing (scRNA-seq) dataset of mouse UUO kidneys (GSE140023) [33]. We learned that CD38 was mainly expressed in endothelial cells, macrophages, and dendritic cells during renal fibrosis (supplemental Fig. S5, A–C). Flow cytometry analysis verified the accumulation of CD38-positive immune cells in UUO kidneys (Fig. 4E), among which $53.1\%$ were macrophages (Fig. 4F). In addition, we also observed CD38 expression in lymphocytes in mouse obstructed kidneys (Fig. 4F). Taken together, these data indicate that CD38 is elevated and is partially expressed in immune cells during obstructive nephropathy. Fig. 4The elevated CD38 contributed to NAD+decline in obstructed kidneys. A and B, CD38 protein levels in human (A) and mouse (B) control and obstructed kidneys were determined by Western blots. Represented results of three separate experiments. C, representative images of kidney sections from control and patients with obstructive nephropathy were stained for CD38. The scale bar represents 20 μm. D–F, flow cytometry analysis of kidneys from mice subjected to sham or UUO operation for 7 days ($$n = 4$$). The experiment was repeated three times. Quantification of CD38+ cells (D), CD38+ immune cells (E), and composition of CD38+ immune cells (F) were shown. G, flow cytometry analysis indicating CD38+ cells in obstructed kidneys from wildtype (WT) or Cd38−/− mice subjected to UUO for 7 days. The experiment was repeated twice. H, NAD+ levels in kidneys from wildtype or Cd38−/− mice subjected to UUO for 7 days ($$n = 6$$). The experiment was repeated three times. All data represent the mean ± SEM. ∗∗$p \leq 0.01$, ∗∗∗$p \leq 0.001.$ p values were calculated by two-tailed Student’s t test (D, E and H). UUO, unilateral ureteral obstruction. CD38 deficiency increases renal NAD+ content in mice under normal conditions [34]. We next determined the impact of CD38 on renal NAD+ levels in obstructive nephropathy by using Cd38−/− mice [35]. The absence of CD38 at the protein level in obstructed kidneys was confirmed by flow cytometry (Fig. 4G). Importantly, the obstructed kidneys from Cd38−/− mice showed a significant rise in NAD+ levels compared with wildtype mice (Fig. 4H), demonstrating that the fibrosis-associated CD38 upregulation contributed to the decline of renal NAD+ levels during obstructive nephropathy. ## CD38 Deletion or Inhibition Ameliorated UUO-Induced Renal Fibrosis Since elevated CD38 was positively correlated with fibrosis markers (Fig. 3, F and G) and contributed to declined renal NAD+ levels (Fig. 4H), we then attempted to investigate the pathogenic role of CD38 in obstructive nephropathy. Histological examination showed that genetic deletion of CD38 ameliorated UUO-induced renal fibrosis, as evidenced by the decrease in collagen accumulation and αSMA-positive cells (Fig. 5, A and B). The kidney injury markers, Kim1 and Ngal, were dramatically downregulated in Cd38−/− mice, suggesting a protective role of CD38 deletion against obstruction-induced tubular damage (Fig. 5C). Moreover, compared with wildtype mice, Tgfb1 had a ∼$30\%$ decrease in UUO kidneys from Cd38−/− mice (Fig. 5D). Collectively, these results suggest that CD38 promotes UUO-induced renal fibrosis. Fig. 5CD38 deletion or inhibition ameliorated UUO-induced renal fibrosis. A, representative images of H&E staining (upper panels), MT staining (middle panels), and αSMA immunohistochemistry staining (lower panels) of kidneys from wildtype and Cd38−/− mice subjected to sham or UUO for 7 days ($$n = 6$$). The experiment was repeated three times. The scale bar represents 50 μm. B, quantification of collagen-positive areas and αSMA-positive areas in (A) ($$n = 6$$). C and D, mRNA levels of Kim1, Ngal (C), and Tgfb1 (D) of obstructed kidneys from wildtype and Cd38−/− mice subjected to UUO for 7 days ($$n = 5$$/4). E, NAD+ levels in obstructed kidneys from vehicle- or 78c-treated wildtype mice subjected to UUO for 7 days ($$n = 6$$). F, representative images of H&E staining (upper panels), MT staining (middle panels), and αSMA immunohistochemistry staining (lower panels) of kidneys from vehicle- or 78c-treated wildtype mice subjected to UUO for 7 days ($$n = 6$$). The experiment was repeated three times. The scale bar represents 50 μm. G, quantification of collagen-positive areas and αSMA-positive areas in (F) ($$n = 6$$). H, immunoblot analysis of phosphorylated (p-) and total protein of SMAD2 in kidney samples from vehicle- and 78c-treated mice subjected to UUO for 7 days. The experiment was repeated three times. All data represent the mean ± SEM. ∗$p \leq 0.05$, ∗∗$p \leq 0.01.$ p values were calculated by two-tailed Student’s t test (B–E and G). MT, Masson's trichrome; UUO, unilateral ureteral obstruction. Next, we explored the therapeutic potential of targeting CD38 in obstructive nephropathy. 78c is a specific and potent CD38 inhibitor, which is able to induce tissue NAD+ boosting through the inhibition of the catalytic activity of CD38 [36]. We found that 78c significantly restored NAD+ levels in UUO kidneys (Fig. 5E). Importantly, 78c treatment diminished collagen deposition and αSMA expression in obstructed kidneys (Fig. 5, F and G). Moreover, 78c treatment blunted TGF-β signaling by suppressing SMAD2 phosphorylation in UUO kidneys (Fig. 5H). These data suggest that CD38 inhibition mitigates obstruction-induced renal fibrosis. ## NAD+ Supplementation Blunted UUO-Induced Renal Fibrosis Having uncovered the NAD+-degrading and profibrotic role of CD38 in obstructive nephropathy, we then asked whether NAD+ supplementation can confer protective effects against obstruction-induced renal fibrosis. We injected NAD+ peritoneally for six consecutive days into wildtype mice subjected to UUO operation, and kidneys were harvested 7 days later (Fig. 6A). NAD+ administration significantly increased tissue levels of NAD+ in obstructed kidneys (Fig. 6B). Histologically, collagen accumulation and αSMA expression were attenuated in UUO kidneys from NAD+-treated mice, whereas their levels were much higher in vehicle-treated mice (Fig. 6, C and D). Moreover, NAD+ supplementation downregulated the expression of kidney injury marker Kim1 in UUO kidneys (Fig. 6E). These data indicate that NAD+ protects kidneys from UUO-induced renal fibrosis. Fig. 6NAD+supplementation blunted UUO-induced renal fibrosis. A, schematic of the experimental design. Eight-week-old male wildtype C57BL/6 mice were subjected to UUO operation and were injected with vehicle or NAD+ peritoneally once a day before kidney harvest at day 7 post UUO. B, NAD+ levels in kidneys from vehicle- and NAD+-treated mice subjected to UUO for 7 days ($$n = 6$$). C, representative images of obstructed kidneys from vehicle- and NAD+-treated UUO mice were stained for H&E, Masson's trichrome, and αSMA ($$n = 6$$). The experiment was repeated three times. The scale bar represents 50 μm. D, quantitation of the collagen-positive areas and αSMA-positive areas of kidney sections in (C). E, mRNA levels of Kim1 in obstructed kidneys from vehicle- or NAD+-treated UUO mice ($$n = 6$$). All data represent the mean ± SEM. ∗$p \leq 0.05$, ∗∗$p \leq 0.01.$ p values were calculated by two-tailed Student’s t test (B, D and E). UUO, unilateral ureteral obstruction. ## Deletion or Inhibition of CD38 and NAD+ Supplementation Reduced Inflammation in Obstructed Kidneys Next, we asked the mechanisms by which CD38 and NAD+ decline promoted renal fibrosis induced by obstruction. Since CD38 and NAD+ are known for their roles in regulating immune response [17] and considerable evidence demonstrates that unresolved inflammation plays a substantial role in kidney fibrosis development [7, 25], we then determined the impact of CD38 and NAD+ on kidney inflammation. We found that CD38 deletion and NAD+ supplementation significantly decreased the recruitment of immune cells into UUO kidneys (Fig. 7, A and B). The obstructed kidneys from Cd38−/− mice recruited fewer macrophages, monocytes, and neutrophils compared with wildtype mice (Fig. 7C). Interestingly, the obstructed kidneys from Cd38−/− mice expressed a lower level of Mcp1 (Fig. 7D), encoding monocyte chemotactic protein 1, which is essential to recruit monocytes/macrophages into inflamed tissue [37]. Moreover, scRNA-seq analysis revealed significant enrichment of leukocyte migration and cell adhesion in CD38-positive macrophages (supplemental Fig. S5D), suggesting that CD38 was required for the recruitment of immune cells. Besides, the proinflammatory cytokine IL-1β was significantly downregulated in obstructed kidneys after CD38 deletion (Fig. 7D) and NAD+ supplementation (Fig. 7E). CD38 and NAD+ decline are also associated with activated NF-κB signaling, which is involved in the development of renal fibrosis [25, 38]. Notably, CD38 inhibition blunted NF-κB signaling by downregulating the expression of NF-κB p65 and its phosphorylation (Fig. 7F). In line with these results, the mouse kidney proteomics showed that the expression of CD38 was positively correlated with the abundance of CD45 ($r = 0.86$, $p \leq 0.001$), IKKB ($r = 0.93$, $p \leq 0.001$), and NF-κB p65 ($r = 0.74$, $p \leq 0.01$) (Fig. 7G). Taken together, these results indicate that targeting CD38 and NAD+ metabolism reduces kidney inflammation, partially by suppressing the infiltration of immune cells and NF-κB signaling, thus mitigating obstruction-induced renal fibrosis. Fig. 7Deletion or inhibition of CD38 and NAD+supplementation reduced inflammation in obstructed kidneys. A, flow cytometry analysis and quantification showing the infiltration of CD45+ cells in obstructed kidneys from wildtype and Cd38−/− mice subjected to UUO for 7 days ($$n = 6$$). B, quantification of the infiltration of CD45+ cells in obstructed kidneys from UUO mice received vehicle or NAD+ treatment ($$n = 6$$). C, numbers of macrophages (Mφs), monocytes (Mono), and neutrophils (PMNs) in obstructed kidneys from wildtype and Cd38−/− mice subjected to UUO for 7 days ($$n = 6$$). D, mRNA levels of Mcp1 and Il1b in obstructed kidneys from wildtype and Cd38−/− mice subjected to UUO for 7 days ($$n = 5$$/4). E, mRNA levels of Il1b in obstructed kidneys from UUO mice that received vehicle or NAD+ treatment ($$n = 4$$). F, immunoblot analysis of phosphorylated (p-) and total protein of NF-κB p65 in kidney samples from vehicle- and 78c-treated mice subjected to UUO for 7 days. Representative data of three separate experiments. G, Spearman's rank correlation between the expression of IKKB, NF-κB p65, CD45, and CD38 in UUO kidneys. The data were extracted from mouse kidney proteomes. All data represent the mean ± SEM. ∗$p \leq 0.05$, ∗∗$p \leq 0.01$, ∗∗∗$p \leq 0.001.$ p values were calculated by two-tailed Student’s t test (A–E) and Spearman's rank correlation analysis (G). UUO, unilateral ureteral obstruction. ## Discussion Considerable advances in understanding obstruction-induced renal fibrosis have been made, yet most of them were based on animal experiments and a comprehensive understanding of human obstructive nephropathy at the proteome level is lacking. In the current study, we presented a proteomic landscape of human obstructed kidneys allowing the analysis and extraction of altered signaling pathways at the proteome level. Importantly, the proteomics uncovered the previously underexplored dysregulation of NAD+ metabolism in obstructed kidneys. Moreover, the major NAD consumer CD38 was strongly induced in human and experimental obstructive nephropathy, which led to declined renal NAD+ levels. CD38 deletion or inhibition and NAD+ supplementation restored NAD+ levels and ameliorated UUO-induced renal fibrosis, partially through the mechanism of reducing kidney inflammation. Our study provides solid molecular evidence at the proteome level for the first time derived directly from pediatric patients with obstructive nephropathy. Several studies implied that NAD+ levels are associated with diabetic nephropathy [10, 39, 40], whereas little is known about their roles in obstructive nephropathy. Our study directly addressed the aberrant NAD+ metabolism in human obstructed kidneys at the proteome level. Moreover, NAD+ restoration by NAD+ supplementation protects renal fibrosis in UUO mice, which was consistent with a previous study showing that supplementation of the NAD+ precursor NAM inhibited UUO-induced renal fibrosis, although NAD+ levels were not assessed after NAM treatment in this study [41]. Therefore, NAD+ boosting might represent an optional therapy for obstructive nephropathy. Several clinical trials are currently ongoing to test the therapeutic potential of NAD+ boosters in human kidney diseases, some of which showed clinical benefits, while others did not [10], indicating that challenges remain for their clinical translation. Supplementation of NAD+ precursors and inhibition of NAD+ consumers are common strategies to recover NAD+ levels. In the present study, the observation of defective NAD+ synthesis in obstructed kidneys raised the possibility that precursor supplementation could be less effective. Importantly, the major NADase CD38 was strongly induced in human and experimental obstructive nephropathy, and its deletion or inhibition recovered renal NAD+ levels and ameliorated UUO-induced renal fibrosis. This was consistent with a recent study demonstrating that targeting CD38-dependent NAD+ degradation mitigates skin, lung, and peritoneal fibrosis [42]. These findings suggest that restoration of NAD+ levels by inhibiting NAD+ consumers (for example, CD38) might be a therapeutic strategy for obstructive nephropathy. CD38 is expressed by various cell types and can be induced during inflammatory conditions, especially in hematopoietic cells [17, 43]. We found that CD38 was partially expressed in immune cells during UUO-induced renal fibrosis, half of which were CD11b+ F$\frac{4}{80}$+ macrophages, suggesting that a subset of CD38-positive macrophages emerged during the fibrotic process. A study reported lately that CD38 on macrophages is capable of reducing tissue levels of NAD+ during aging, depending on its ectoenzyme activity [18]. Moreover, CD38 can modulate inflammation by regulating cell recruitment, phagocytosis, and cytokine release [17]. Our work showed that CD38 was required for the infiltration of innate immune cells and IL-1β expression. Of note, emerging evidence supports the idea that CD38-mediated NAD+ depletion contributes to the inflammatory response [17]. In our hand, inhibition of CD38 catalytic activity by 78c blunted NF-κB signaling, suggesting that the immunomodulatory role of CD38 was associated with its NAD+-degrading activity. Besides, we cannot exclude the contribution of CD38 expressed in other cells (for example, endothelial cells) to NAD+ consumption and kidney pathologies. The kidney is a high energy-demanding organ that maintains the homeostasis of electrolyte, water, and acid–base balance, requiring a large number of mitochondria to meet energy needs. A correlation between mitochondrial dysfunction and kidney diseases has been demonstrated repeatedly [44]. Our kidney proteomics revealed a global dysregulation of cellular metabolism in obstructed kidneys, and mitochondrial abnormalities were central to these metabolic alterations. Moreover, CD38 increase in aging mice contributes to the development of mitochondrial dysfunction, by degrading cellular NAD+ and subsequently suppressing SIRT3 [45]. SIRT3 is one of the sirtuins localized in mitochondria that regulate key mitochondrial proteins important for oxidative homeostasis [46]. In our study, we observed a dramatic downregulation of SIRT3 in obstructed kidneys. Therefore, mitochondrial abnormalities in obstructive nephropathy might be attributed to CD38 elevation, NAD+ decline, and SIRT3 suppression. Targeting CD38 and NAD+ metabolism might restore mitochondrial function, which needs to be further investigated. In summary, we presented a proteomic landscape of obstructed kidneys from pediatric patients with UPJO and provided a rich resource of proteomic data to facilitate future study of obstructive nephropathy. We uncovered an aberrant NAD+ metabolism and a fibrosis-associated elevation of CD38 during obstructive nephropathy. CD38 deletion or inhibition and NAD+ supplementation mitigated UUO-induced renal fibrosis, partially through the mechanism of reducing kidney inflammation. Thus, our study emphasized the importance and therapeutic potential of CD38-mediated NAD+ metabolism in obstructive nephropathy. ## Limitations of This Study A limitation of the present study is the small size of the patient samples, for the reason that it is difficult to recruit a large number of pediatric patients. Besides, there are not technical or biological replicates for the DIA-based proteomics analysis. However, this can be partially compensated by a further in-depth exploration of the main findings. Since CD38 is barely expressed in tubular cells, further investigations are required to identify how CD38-expressing immune cells affect NAD+ levels and metabolic events in tubular cells. Moreover, the mechanisms by which CD38 and CD38-mediated NAD+ decline regulate kidney inflammation and fibrosis remain to be elucidated. ## Data Availability The MS raw data have been deposited to the ProteomeXchange Consortium via the iProx partner repository [47, 48] with the dataset identifier PXD039314. ## Supplemental data This article contains supplemental data. ## Conflict of interest The authors declare no competing interests. ## Supplemental Data Supplemental Data 1 Supplemental Data 2 Supplemental Data 3 Supplemental Fig. S1Proteomic landscape of human obstructed kidneys. A–C, heatmap indicating the expression of identified proteins related to collagens (A), phagocytosis, cytokine signaling in immune response (B), and interferon signaling (C) in human control and obstructed kidneys. D, immunoblot analysis of phosphorylated (p-) and total protein of SMAD2 in kidney samples from controls and patients with obstructive nephropathy. Represented results of three separate experiments. Supplemental Fig. S2. Mitochondrial dysfunction in human obstructed kidneys. Heatmap indicating the expression of significant down-regulated proteins (p value <0.05, q value <0.05, and Ob/Ctr fold change <0.5) related to mitochondria organization in human obstructed kidneys. Supplemental Fig. S3. Proteomic profiling of mouse obstructed kidneys. A, PCA analysis of sham kidney proteomes and UUO kidney proteomes ($$n = 6$$/6). B, heatmap showing all proteins identified by DIA-MS of kidneys from wild-type mice subjected to sham or UUO operation for 7 days. C and D, top 20 most significant terms enriched by significantly up-regulated (C) and down-regulated (D) proteins (p value <0.05, q value <0.05, and UUO/Sham fold change >2 or <0.5) in obstructed kidneys from UUO mice. Supplemental Fig. S4. Aberrant NAD+metabolism in human and mouse obstructed kidneys. A and B, expression levels of key enzymes catalyzing NAD+ synthesis and consumption in human (A) and mouse (B) kidney proteomics. All data represent the mean ± SEM. ∗$p \leq 0.05$, ∗∗$p \leq 0.01$, ∗∗∗$p \leq 0.01$, NS no significance. p values were calculated by two-tailed Student’s t-test. Supplemental Fig. S5. The expression patterns of CD38 in UUO kidneys. A, Uniform Manifold Approximation and Projection (UMAP) dimension reduction showing distinct cell types identified by unsupervised clustering. The scRNAseq data were from Gene Expression Omnibus (GEO) database (GSE140023). PCT, proximal convoluted tubule; Mac, macrophage; Neu, neutrophil; Mono, monocyte; DC, dendritic cell; EC, endothelial cell; NK, natural killer cell; Mes, mesenchymal cell; LoH, loop of Henle; CD, Collecting Duct. B and C, feature plots (B) and violin plots (C) showing the expression of Cd38 in each cell type in UUO kidneys. D, GO analysis showing the top 20 upregulated pathways in CD38+ macrophages. Supplemental Fig. S6. Uncropped scans of Western blots. ## Author contributions H. Z., F. Y., and X. L. conception and supervision; J. W. DIA-MS; N. L. Bioinformatic analysis; Y. T., X. L., D. L., and Y. L. experiments, data acquisition, analysis, and interpretation; X. Z., P. L., Y. Z., X. Z., L. M., and T. T. technical or material support; Y. T., H. Z., F. Y., and X. L. article writing and revising; all authors approved the final version of the manuscript. ## References 1. Vemulakonda V.M.. Ureteropelvic junction obstruction: diagnosis and management. Curr. Opin. Pediatr. (2021) 33 227-234. PMID: 33470672 2. Chevalier R.L.. Congenital urinary tract obstruction: the long view. Adv. Chronic Kidney Dis. (2015) 22 312-319. PMID: 26088076 3. Stevens S.. Obstructive kidney disease. Nurs. Clin. North Am. (2018) 53 569-578. PMID: 30388982 4. Humphreys B.D.. Mechanisms of renal fibrosis. Annu. Rev. Physiol. (2018) 80 309-326. PMID: 29068765 5. Jackson L., Woodward M., Coward R.J.. The molecular biology of pelvi-ureteric junction obstruction. Pediatr. Nephrol. (2018) 33 553-571. PMID: 28286898 6. Cachat F., Lange-Sperandio B., Chang A.Y., Kiley S.C., Thornhill B.A., Forbes M.S.. Ureteral obstruction in neonatal mice elicits segment-specific tubular cell responses leading to nephron loss. Kidney Int. (2003) 63 564-575. PMID: 12631121 7. Tang P.M., Nikolic-Paterson D.J., Lan H.Y.. Macrophages: versatile players in renal inflammation and fibrosis. Nat. Rev. Nephrol. (2019) 15 144-158. PMID: 30692665 8. Ralto K.M., Rhee E.P., Parikh S.M.. NAD(+) homeostasis in renal health and disease. Nat. Rev. Nephrol. (2020) 16 99-111. PMID: 31673160 9. Yoshino J., Baur J.A., Imai S.I.. NAD(+) intermediates: the biology and therapeutic potential of NMN and NR. Cell Metab. (2018) 27 513-528. PMID: 29249689 10. Katsyuba E., Romani M., Hofer D., Auwerx J.. NAD(+) homeostasis in health and disease. Nat. Metab. (2020) 2 9-31. PMID: 32694684 11. Covarrubias A.J., Perrone R., Grozio A., Verdin E.. NAD(+) metabolism and its roles in cellular processes during ageing. Nat. Rev. Mol. Cell Biol. (2021) 22 119-141. PMID: 33353981 12. Poyan Mehr A., Tran M.T., Ralto K.M., Leaf D.E., Washco V., Messmer J.. De novo NAD(+) biosynthetic impairment in acute kidney injury in humans. Nat. Med. (2018) 24 1351-1359. PMID: 30127395 13. Katsyuba E., Mottis A., Zietak M., De Franco F., van der Velpen V., Gariani K.. De novo NAD(+) synthesis enhances mitochondrial function and improves health. Nature (2018) 563 354-359. PMID: 30356218 14. Guan Y., Wang S.R., Huang X.Z., Xie Q.H., Xu Y.Y., Shang D.. Nicotinamide mononucleotide, an NAD(+) precursor, rescues age-associated susceptibility to AKI in a sirtuin 1-dependent manner. J. Am. Soc. Nephrol. (2017) 28 2337-2352. PMID: 28246130 15. Tran M.T., Zsengeller Z.K., Berg A.H., Khankin E.V., Bhasin M.K., Kim W.. PGC1alpha drives NAD biosynthesis linking oxidative metabolism to renal protection. Nature (2016) 531 528-532. PMID: 26982719 16. Chini E.N., Chini C.C.S., Espindola Netto J.M., de Oliveira G.C., van Schooten W.. The pharmacology of CD38/NADase: an emerging target in cancer and diseases of aging. Trends Pharmacol. Sci. (2018) 39 424-436. PMID: 29482842 17. Piedra-Quintero Z.L., Wilson Z., Nava P., Guerau-de-Arellano M.. CD38: an immunomodulatory molecule in inflammation and autoimmunity. Front. Immunol. (2020) 11 597959. PMID: 33329591 18. Covarrubias A.J., Kale A., Perrone R., Lopez-Dominguez J.A., Pisco A.O., Kasler H.G.. Senescent cells promote tissue NAD(+) decline during ageing via the activation of CD38(+) macrophages. Nat. Metab. (2020) 2 1265-1283. PMID: 33199924 19. Dominguez-Pantoja M., Lopez-Herrera G., Romero-Ramirez H., Santos-Argumedo L., Chavez-Rueda A.K., Hernandez-Cueto A.. CD38 protein deficiency induces autoimmune characteristics and its activation enhances IL-10 production by regulatory B cells. Scand. J. Immunol. (2018) 87 20. Ogura Y., Kitada M., Xu J., Monno I., Koya D.. CD38 inhibition by apigenin ameliorates mitochondrial oxidative stress through restoration of the intracellular NAD(+)/NADH ratio and Sirt3 activity in renal tubular cells in diabetic rats. Aging (Albany NY) (2020) 12 11325-11336. PMID: 32507768 21. Yang W., Tao Y., Wu Y., Zhao X., Ye W., Zhao D.. Neutrophils promote the development of reparative macrophages mediated by ROS to orchestrate liver repair. Nat. Commun. (2019) 10 1076. PMID: 30842418 22. Zhou Y., Zhou B., Pache L., Chang M., Khodabakhshi A.H., Tanaseichuk O.. Metascape provides a biologist-oriented resource for the analysis of systems-level datasets. Nat. Commun. (2019) 10 1523. PMID: 30944313 23. Wallace D.P.. Periostin in the kidney. Adv. Exp. Med. Biol. (2019) 1132 99-112. PMID: 31037629 24. Meng X.M., Nikolic-Paterson D.J., Lan H.Y.. Inflammatory processes in renal fibrosis. Nat. Rev. Nephrol. (2014) 10 493-503. PMID: 24981817 25. Lv W., Booz G.W., Wang Y., Fan F., Roman R.J.. Inflammation and renal fibrosis: recent developments on key signaling molecules as potential therapeutic targets. Eur. J. Pharmacol. (2018) 820 65-76. PMID: 29229532 26. Schunk S.J., Floege J., Fliser D., Speer T.. WNT-beta-catenin signalling - a versatile player in kidney injury and repair. Nat. Rev. Nephrol. (2021) 17 172-184. PMID: 32989282 27. Kang H.M., Ahn S.H., Choi P., Ko Y.A., Han S.H., Chinga F.. Defective fatty acid oxidation in renal tubular epithelial cells has a key role in kidney fibrosis development. Nat. Med. (2015) 21 37-46. PMID: 25419705 28. Jiang M., Bai M., Lei J., Xie Y., Xu S., Jia Z.. Mitochondrial dysfunction and the AKI-to-CKD transition. Am. J. Physiol. Renal Physiol. (2020) 319 F1105-F1116. PMID: 33073587 29. Miguel V., Tituana J., Herrero J.I., Herrero L., Serra D., Cuevas P.. Renal tubule Cpt1a overexpression protects from kidney fibrosis by restoring mitochondrial homeostasis. J. Clin. Invest. (2021) 131 30. Chevalier R.L., Forbes M.S., Thornhill B.A.. Ureteral obstruction as a model of renal interstitial fibrosis and obstructive nephropathy. Kidney Int. (2009) 75 1145-1152. PMID: 19340094 31. Kuppe C., Ibrahim M.M., Kranz J., Zhang X., Ziegler S., Perales-Paton J.. Decoding myofibroblast origins in human kidney fibrosis. Nature (2021) 589 281-286. PMID: 33176333 32. Tan R.J., Zhou D., Zhou L., Liu Y.. Wnt/beta-catenin signaling and kidney fibrosis. Kidney Int. Suppl. (2011) (2014) 4 84-90. PMID: 26312156 33. Conway B.R., O'Sullivan E.D., Cairns C., O'Sullivan J., Simpson D.J., Salzano A.. Kidney single-cell atlas reveals myeloid heterogeneity in progression and regression of kidney disease. J. Am. Soc. Nephrol. (2020) 31 2833-2854. PMID: 32978267 34. Aksoy P., White T.A., Thompson M., Chini E.N.. Regulation of intracellular levels of NAD: a novel role for CD38. Biochem. Biophys. Res. Commun. (2006) 345 1386-1392. PMID: 16730329 35. Zhang X., Li L., Zhang Q., Wei Q., Lin J., Jia J.. CD38 causes autophagic flux inhibition and cardiac dysfunction through a transcriptional inhibition pathway under hypoxia/ischemia conditions. Front. Cell Dev. Biol. (2020) 8 191. PMID: 32363189 36. Tarrago M.G., Chini C.C.S., Kanamori K.S., Warner G.M., Caride A., de Oliveira G.C.. A potent and specific CD38 inhibitor ameliorates age-related metabolic dysfunction by reversing tissue NAD(+) decline. Cell Metab. (2018) 27 1081-1095.e10. PMID: 29719225 37. Hsieh C.L., Niemi E.C., Wang S.H., Lee C.C., Bingham D., Zhang J.. CCR2 deficiency impairs macrophage infiltration and improves cognitive function after traumatic brain injury. J. Neurotrauma (2014) 31 1677-1688. PMID: 24806994 38. Qian Y., Chen C., Ma L., Wang Z., Wang L.F., Zuo L.. CD38 deficiency promotes inflammatory response through activating Sirt1/NF-kappaB-mediated inhibition of TLR2 expression in macrophages. Mediators Inflamm. (2018) 2018 8736949. PMID: 29977153 39. Papadimitriou A., Silva K.C., Peixoto E.B., Borges C.M., Lopes de Faria J.M., Lopes de Faria J.B.. Theobromine increases NAD(+)/Sirt-1 activity and protects the kidney under diabetic conditions. Am. J. Physiol. Renal Physiol. (2015) 308 F209-225. PMID: 25411384 40. Chen Y., Liang Y., Hu T., Wei R., Cai C., Wang P.. Endogenous Nampt upregulation is associated with diabetic nephropathy inflammatory-fibrosis through the NF-kappaB p65 and Sirt1 pathway; NMN alleviates diabetic nephropathy inflammatory-fibrosis by inhibiting endogenous Nampt. Exp. Ther. Med. (2017) 14 4181-4193. PMID: 29104634 41. Zheng M., Cai J., Liu Z., Shu S., Wang Y., Tang C.. Nicotinamide reduces renal interstitial fibrosis by suppressing tubular injury and inflammation. J. Cell. Mol. Med. (2019) 23 3995-4004. PMID: 30993884 42. Shi B., Wang W., Korman B., Kai L., Wang Q., Wei J.. Targeting CD38-dependent NAD(+) metabolism to mitigate multiple organ fibrosis. iScience (2021) 24 101902. PMID: 33385109 43. Schneider M., Schumacher V., Lischke T., Lucke K., Meyer-Schwesinger C., Velden J.. CD38 is expressed on inflammatory cells of the intestine and promotes intestinal inflammation. PLoS One (2015) 10 44. Bhargava P., Schnellmann R.G.. Mitochondrial energetics in the kidney. Nat. Rev. Nephrol. (2017) 13 629-646. PMID: 28804120 45. Camacho-Pereira J., Tarrago M.G., Chini C.C.S., Nin V., Escande C., Warner G.M.. CD38 dictates age-related NAD decline and mitochondrial dysfunction through an SIRT3-dependent mechanism. Cell Metab. (2016) 23 1127-1139. PMID: 27304511 46. Shen Y., Wu Q., Shi J., Zhou S.. Regulation of SIRT3 on mitochondrial functions and oxidative stress in Parkinson's disease. Biomed. Pharmacother. (2020) 132 110928. PMID: 33128944 47. Ma J., Chen T., Wu S., Yang C., Bai M., Shu K.. iProX: an integrated proteome resource. Nucleic Acids Res. (2019) 47 D1211-D1217. PMID: 30252093 48. Chen T., Ma J., Liu Y., Chen Z., Xiao N., Lu Y.. iProX in 2021: connecting proteomics data sharing with big data. Nucleic Acids Res. (2022) 50 D1522-D1527. PMID: 34871441
--- title: Deep learning-assisted diagnosis of chronic atrophic gastritis in endoscopy authors: - Yanting Shi - Ning Wei - Kunhong Wang - Jingjing Wu - Tao Tao - Na Li - Bing Lv journal: Frontiers in Oncology year: 2023 pmcid: PMC10025314 doi: 10.3389/fonc.2023.1122247 license: CC BY 4.0 --- # Deep learning-assisted diagnosis of chronic atrophic gastritis in endoscopy ## Abstract ### Background Chronic atrophic gastritis (CAG) is a precancerous condition. It is not easy to detect CAG in endoscopy. Improving the detection rate of CAG under endoscopy is essential to reduce or interrupt the occurrence of gastric cancer. This study aimed to construct a deep learning (DL) model for CAG recognition based on endoscopic images to improve the CAG detection rate during endoscopy. ### Methods We collected 10,961 endoscopic images and 118 video clips from 4,050 patients. For model training and testing, we divided them into two groups based on the pathological results: CAG and chronic non-atrophic gastritis (CNAG). We compared the performance of four state-of-the-art (SOTA) DL networks for CAG recognition and selected one of them for further improvement. The improved network was called GAM-EfficientNet. Finally, we compared GAM-EfficientNet with three endoscopists and analyzed the decision basis of the network in the form of heatmaps. ### Results After fine-tuning and transfer learning, the sensitivity, specificity, and accuracy of GAM-EfficientNet reached $93\%$, $94\%$, and $93.5\%$ in the external test set and $96.23\%$, $89.23\%$, and $92.37\%$ in the video test set, respectively, which were higher than those of the three endoscopists. ### Conclusions The CAG recognition model based on deep learning has high sensitivity and accuracy, and its performance is higher than that of endoscopists. ## Introduction According to the latest World Cancer Report released by International Agency for Research on Cancer (IARC), 1,089,103 new cases of gastric cancer and 768,793 deaths were reported worldwide in 2020, accounting for $5.6\%$ and $7.7\%$ of total cancer incidence and deaths, ranking fifth and fourth respectively [1]. Most gastric cancers are gastric adenocarcinomas and CAG is the most common stage of progression to gastric adenocarcinoma [2, 3]. Studies have shown that the 5-year incidence of gastric cancer in patients with CAG is $1.9\%$ [3]. Some studies in China and Japan have shown that the prevalence of CAG is higher than $50\%$ (4–6). The collection of biopsies during gastroscopy and pathological analysis is the “gold standard” for diagnosing CAG. This depends significantly on the endoscopist’s ability to collect biopsies [7]. Studies have shown that CAG can be detected by white-light endoscopy but with poor accuracy [3]. The sensitivity of endoscopic diagnosis of atrophic gastritis is $61.5\%$ in the antrum and $46.8\%$ in the body of the stomach [8]. The manual operation of physicians to identify lesions with the naked eye renders it difficult to exclude missed diagnoses owing to fatigue and inexperience. Therefore, seeking an objective and accurate method to identify CAG is very important to slow down or interrupt gastric cancer progression and reduce endoscopists’ workload. In recent years, artificial intelligence (AI) techniques, represented by deep learning (DL), have been widely used in various medical imaging fields. Examples include disease detection [9, 10], disease prediction [11, 12], and organ detection [13, 14]. AI techniques have also shown excellent performance for the diagnosis of digestive diseases. Ueyama et al. [ 15] constructed a DL computer-aided diagnosis system based on narrow-band imaging to diagnose early gastric cancer, and the accuracy and sensitivity were $98.7\%$ and $98\%$, respectively. Shichijo et al. [ 16] constructed a convolutional neural network (CNN) [17, 18] and evaluated its ability to diagnose Helicobacter pylori infection, and the accuracy and sensitivity were $87.7\%$ and $88.9\%$, respectively. Zhao et al. [ 19] developed a DL-based assisted diagnostic system for the localization of colon polyps, with a sensitivity of $98.4\%$ in prospective validation. AI also showed excellent ability in CAG diagnosis. For example, Guimarães et al. [ 20] and Mu et al. [ 21] automatically extracted endoscopic image features to identify CAG by DL techniques with an accuracy of $93\%$ and $95\%$, respectively. CAG is endoscopically visible as a red-white mucosal, predominantly white, exposed section of the mucosal blood vessels, and it can be accompanied by mucosal granules or nodules [22, 23]. We aim to capture these visible or subvisible image features for CAG recognition using a new DL model, providing more evidence for the feasibility of AI-aided diagnosis of CAG. CNN is the mainstream algorithm used in the field of image recognition. Since 2017, transformers [24] have shown powerful capabilities in image classification [25, 26], semantic segmentation [27, 28], and object detection [29, 30]. In this study, we selected four SOTA DL networks for comparison: two CNNs and two transformers. A new CAG recognition model was constructed based on one of them. The results showed that the model’s accuracy, sensitivity, and specificity were better than those of endoscopic experts. ## Data collection and preprocessing In this study, three datasets were collected: [1] an internal image dataset from Zibo Central Hospital, which was used for the training and internal testing of the model; [2] an external test set, an image dataset from the Zhangdian Maternal and Child Health Care Hospital; and [3] video test set, a video dataset from Zibo Central Hospital. The pathology results support all images and videos. Images and videos from Zibo Central Hospital were captured using an Olympus GIF-HQ290 or GIF-H290Z (Olympus, Tokyo, Japan). Images from Zhangdian Maternal and Child Health Care Hospital were captured by Pentax EG29-i10 (Pentax, Tokyo, Japan). All images and videos were captured in the normal white imaging mode. The resolution of the original image was 1920 × 1080 pixels and the format was BMP. The resolution of the original video was 1920 × 1080 pixels, the encoding method was MJPEG, and the frame rate was 25 frames per second (fps). We labeled each image as CAG or CNAG. For videos, one label per video was equivalent to patient-specific labeling. Internal image dataset: We reviewed the images of patients who underwent gastroscopy at Zibo Central Hospital between June 2020 and June 2022 and had pathological results of chronic gastritis. To reduce interference, we excluded images based on the following conditions: poor quality, inadequate preparation of the digestive tract, altered gastric anatomy because of gastric surgery, and other diseases. The final dataset included 10,361 images from 3,718 patients. Based on the pathology results, we divided the dataset into two categories, CAG and CNAG. The CAG dataset contained images of 1933 patients with CAG, 1114 men and 819 women, with a mean age of 57.15 (± 10.49), 921 with mild atrophy, 984 with moderate atrophy, and 28 with severe atrophy. The CNAG dataset contained images of 1,785 patients with CNAG, 883 men and 902 women, with a mean age of 43.9 (± 13.23). A total of 5219 CAG images were obtained, including 3,826 images of the gastric sinuses, 1184 images of the gastric horns, and 209 images of the gastric body. A total of 5,142 CNAG images were obtained, including 3,545 images of the gastric sinuses, 980 images of the gastric horns, and 617 images of the gastric body. External test set: We reviewed images of patients who underwent gastroscopy at Zhangdian Maternal and Child Health Care Hospital between January 2022 and October 2022 and had pathological results of chronic gastritis. The inclusion and exclusion processes were identical for the internal image dataset. Finally, 300 images from 116 patients with CAG and 300 images from 98 patients with CNAG were included. Video test set: We collected video clips of patients who underwent gastroscopy at the Zibo Central Hospital between September 2022 and October 2022. The videos were deliberate scans of the entire gastric region performed by the endoscopist during endoscopy. The exclusion criteria were the same as those used for the internal image dataset. Fifty-three patients with CAG and 65 patients with CNAG were included in the study. The mean duration of the 118 video clips was 50.16 ± 9.57 seconds. Before training with the DL model, we processed the images involved in the training. First, we removed the invalid parts of the image and scaled the image to 512 × 512 pixels. We adopted image-enhancement strategies during model training, such as random rotation, flipping, and color dithering, to improve the model’s generalisation ability. The processing method is illustrated in Figure 1. Figure 1: Image preprocessing. (A) original image; (B) invalid area removed, resize to 512512 pixels; (C) rotate 90°counterclockwise; (D) flip vertically; (E) color random dither.* This study was approved by the ethics committees of the two hospitals involved (No. 202201016, Zibo Central Hospital. No. 202210019, Zhangdian Maternal and Child Health Care Hospital). The patients in the video test set provided written informed consent prior to participation. The ethics committee waived the requirement for informed consent for patients involved in retrospective imaging. ## Deep learning method The experimental hardware environment included an Intel i9 12900 K CPU, an Nvidia GeForce GTX 3090 GPU, and 32 GB of RAM. The experimental software environment included Ubuntu 22.04, CUDA 11.3, Anaconda 4.14, and PyTorch 1.12.1. We selected four SOTA DL networks for inclusion in this study: EfficientNetV2 [31], ConvNeXt [32], ViT [25], and Swin [26]. Their commonly used versions, EfficientNetV2-M [31], ConvNeXt-L [32], ViT-B [25], and Swin-B [26], were selected based on the computing power of GPU for training. EfficientNetV2 and ConvNeXt are representative CNN. ViT and Swin are representative transformer networks that have emerged in recent years. We used a pretrained model on ImageNet [33, 34] for transfer learning [35, 36]. A significant problem in medical image analysis is that the datasets are relatively small, resulting in less-accurate trained models. Thus, transfer learning can effectively solve this problem. In recent years, transfer learning has achieved good results in medical image analysis [37, 38], which can improve the accuracy of models and accelerate training [39, 40]. The training process for the four networks is illustrated in Figure 2. The accuracy of each network in the validation set increased with an increase in the number of iterations. After 150 epochs, the accuracy of EfficientNetV2-M, ConvNeXt-L, and ViT-B on the validation set stabilized and Swin-B oscillated in an interval. In summary, EfficientNetV2-M outperformed the other three networks; therefore, we selected it for further optimization. Figure 2: Training process of the four networks. We introduced the global attention mechanism (GAM) [41] module based on EfficientNetV2-M to enable the network to focus more on the critical information in the CAG region. The improved network is called GAM-EfficientNet and its structure is shown in Figure 3. The GAM module mainly consists of a channel attention submodule (CAM) and spatial attention submodule (SAM), and its structure is shown in Figure 4. The CAM first performs dimensional conversion for the input feature map. Following dimensional conversion, the feature map is input into a two-layer multilayer perceptron (MLP). MLP is an encoder-decoder structure that magnifies cross-dimensional channel-spatial dependencies. Subsequently, it is converted to the original dimension and output by sigmoid processing. In SAM, two convolution kernels of 7 × 7 are used for spatial information fusion. GAM amplifies the global dimension-interactive features by adding an element-wise multiplication operation between CAM and SAM. Figure 3: GAM-EfficientNet architecture. Conv represents convolution; Pooling is average pooling layer; FC is full connection layer; Ä represents element-wise multiplication; ×n is repeat times. Figure 4: Channel attention and spatial attention submodule. MPL represents multi-layer perceptron; r represents reduction ratio; W, H, and C represent the feature map’s width, height, and number of channels. With other fixed parameters, we selected the optimal parameters of the network using four cross-validations. The following were the final parameters: batch-size was 16, optimizer was AdamW algorithm, initial learning rate was 0.001, learning rate decay strategy was cosine decay [42], and weight decay coefficient was 0.01. After the parameters were determined, the network was retrained to obtain the final CAG recognition model. ## Model evaluation We invited three endoscopists with more than 10 years of experience in endoscopic operations to participate in the test. They diagnosed randomly ordered images/videos in the three test sets without being aware of the pathological findings. The test results were compared with those of GAM-EfficientNet. The CAG identification in this study was a binary classification problem. We evaluated the model’s performance by calculating the sensitivity (recall), specificity, precision, accuracy, and F1-score, and by plotting the receiver operating characteristic (ROC) curve. The relevant formulas are as follows: where TP, FP, FN and TN represent the numbers of true positives, false positives, false negatives, and true negatives. All statistical analyses were performed using GraphPad Prism 9.3.1 (GraphPad Software, Inc., San Diego, CA, USA). However, although AI has demonstrated excellent diagnostic performance (43–45), it has a significant drawback that is difficult to explain [46]. Explainable AI is an important research direction in medical AI [47]. One effective method is to generate heatmaps of the images. We used a gradient-weighted class activation map (Grad-CAM) [46] to create heatmaps showing the regions in which the model predicts the CAG. This can assess whether the model identification process is correct and provide aid to the endoscopist for diagnosis. ## Dataset A total of 10,961 endoscopic images and 118 video clips from 4,050 patients were included in the study. To train the final model, the internal image dataset was randomly divided into training, validation, and internal test sets at a ratio of 3:1:1. The training and validation sets were used to train the model. The internal test set was not involved in the training and was only used to evaluate the diagnostic capability of the model. The details of the internal image dataset, external test dataset, and video test set are listed in Table 1. Table 1 \| Category \| Internal image dataset \| Internal image dataset.1 \| Internal image dataset.2 \| Internal image dataset.3 \| vExternal validation set \| vExternal validation set.1 \| Video validation set \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Category \| Patients \| Trian Set \| Validation Set \| Test Set \| Patients \| Image \| Patients \| \| CAG \| 1933 \| 3131 \| 1044 \| 1044 \| 116 \| 300 \| 53 \| \| CNAG \| 1785 \| 3085 \| 1028 \| 1029 \| 98 \| 300 \| 65 \| \| Total \| 3718 \| 6216 \| 2072 \| 2073 \| 214 \| 600 \| 118 \| ## Model comparison We tested five DL models on our internal test set, and the test results are listed in Table 2. The main performance metrics of the GAM-EfficientNet are higher than those of the other four networks. The combined performance metric F-Score of GAM-EfficientNet is $94.26\%$, higher than the second-best performer, EfficientNetV2-M, by 1.4 percentage points. Supplementary File 1 contains complete training and testing process data for the models. Table 2 \| Diagnosed by \| Sensitivity \| Specificity \| Precision \| Accuracy \| F1-score \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| ConvNeXt-L \| 92.91 \| 90.09 \| 90.49 \| 91.5 \| 91.68 \| \| ViT-B \| 93.4 \| 90.38 \| 90.78 \| 91.9 \| 92.07 \| \| Swin-B \| 92.53 \| 89.12 \| 89.61 \| 90.83 \| 91.05 \| \| EfficientNetV2-M \| 92.91 \| 92.71 \| 92.82 \| 92.81 \| 92.87 \| \| GAM-EfficientNet \| 94.35 \| 94.07 \| 94.17 \| 94.21 \| 94.26 \| ## Performance evaluation Three endoscopists tested the internal and external test sets without being aware of the pathology results. The test results of GAM-EfficientNet and endoscopists are listed in Table 3. The ROC curves are shown in Figure 5. The area under the curve (AUC) of the GAM-EfficientNet was $98.79\%$ ($95\%$ CI: 0.98 to 0.99) for the internal test set and $99.46\%$ ($95\%$ CI: 0.99- 1.00) for the external test set. No overlap occurred between GAM-EfficientNet and the endoscopists. This indicated that AUC was statistically significantly different between GAM-EfficientNet and the endoscopists. The performance of GAM-EfficientNet in diagnosing CAG was significantly higher than that of the endoscopists. We attempted to determine the reasons for these model prediction errors. In the internal test set, 120 images were incorrectly predicted, including 59 FN images from 48 patients and 61 FP images from 51 patients. Three endoscopists collaboratively diagnosed the 120 misidentified images. If endoscopists disagree on the diagnosis, they resolved through discussion. Of the 59 FN images, only four were diagnosed correctly. Nine of the 61 FP images were correctly identified. We found that it was difficult for experienced endoscopists to evaluate the images that the model incorrectly predicted. We analyzed the causes of these prediction errors. This study used pathology results as the gold standard and endoscopists selected the images in the dataset. However, the following two cases cannot be excluded: the endoscopist did not take the atrophy site when taking the pathology, and the image did not contain the area of pathology. In addition, different light intensities and angles could also affect the judgment of the model. The Grad-CAM heatmap highlights the regions of interest of the GAM-EfficientNet in red, yellow, and green. As shown in Figure 6, the endoscopists labeled some of the images with atrophied regions and compared them with the heatmaps generated by the model. The endoscopists’ annotations were generally consistent with the areas of concern for the model. In summary, GAM-EfficientNet can focus on meaningful regions in endoscopic images for CAG predictions. Heatmaps can also provide endoscopists with a visual basis for diagnosis. Figure 6: Feature heatmaps of the GAM-EfficientNet. To improve the quality of the heatmaps, we cropped the original endoscopic image and did not compress it. ## Video verification As the images used for model training, internal testing, and external testing were selected by the endoscopist, GAM-EfficientNet’s adaptation to more complex real-time endoscopic environments need further validation. To test the model, we collected video clips of 118 patients who underwent real-time endoscopy. A separate model was trained to recognize blurred frames in the videos. Blurred frames were ignored during the GAM-EfficientNet diagnosis. To prevent the model from misdiagnosis owing to one frame, we specified that five consecutive frames were diagnosed as CAG. Otherwise, the patient was diagnosed as having CNAG. Three endoscopists independently diagnosed the videos based on their experience. The diagnostic results of the model and endoscopists are listed in Table 3. Supplementary Video 1 shows an example video of the model diagnostic process, which was converted to 10 fps to provide a better view of the diagnostic process. In the video test set, the F1-score and AUC of GAM-EfficientNet were $91.89\%$ and $92.73\%$, respectively, still higher than those of the endoscopist. ## Discussion Gastric cancer is the fifth most prevalent type of cancer and the fourth most common cause of cancer-related deaths worldwide [1]. Gastric mucosal atrophy is a critical stage in gastric cancer progression; the higher the degree of mucosal atrophy, the higher the risk of cancer [3]. If we can improve the recognition of CAG and timely intervention, the incidence of gastric cancer and the mortality rate will be reduced. In recent years, DL technology has achieved considerable success in image recognition, and its application in assisted gastroscopy diagnosis is of great significance in improving disease recognition rates. In this study, we created a new DL network, GAM-EfficientNet, based on EfficientNetV2-M by adding a GAM module. GAM introduces spatial and channel attention mechanisms that allow the network to focus more on valuable regions in the image. GAM-EfficientNet outperforms the other networks mentioned in the paper in terms of recognition ability, with sensitivity, specificity, and accuracy of $93\%$, $94\%$, and $93.5\%$, respectively, on the external test set. CAG is a precancerous disease that requires a high recall rate to reduce the number of missed diagnoses. The precision and recall rates of the GAM-EfficientNet were $93.94\%$ and $93\%$, respectively, on the external test set, which suggests that the model has high precision and a low probability of missing diagnosis. The comprehensive evaluation indices F1-score and AUC were $93.47\%$ and $99.46\%$, respectively, indicating the high value of the model as an aid to diagnosis. It provides a visual diagnostic basis for the endoscopist by generating a heatmap showing the areas of attention where the GAM-EfficientNet makes decisions. To further validate the model’s performance, we tested it on a video test set and compared it with that of the three endoscopists. The results showed that the F1-score and AUC of GAM-EfficientNet were $91.89\%$ and $92.73\%$, respectively, 1.58 and 6.73 percentage points lower than those on the external test set, respectively. The performance of all three endoscopists on the video test set improved, but remained lower than that of GAM-EfficientNet. The model has limitations and scope for further improvement, notably the following: [1] Multi-center study. The training set images used in this study were obtained from one hospital, and they were high-quality images screened by endoscopists. However, the images may require more diversity. In a real environment, different devices and parameter settings can affect endoscopic image imaging, and factors such as the angle, light, food residue, and digestive fluid can affect the evaluation of the model. In the future, we will include multiple centers to collect more images of different models of endoscopic devices for training, to improve the generalization ability of the model. [ 2] The classification was further refined according to the severity of atrophy. In this study, we identified only CAG and did not distinguish its severity. The degree of atrophy may differ from the pathological findings at different positions in the same patient. In the future, we will collect endoscopic images strictly according to the locations where the pathological biopsy was conducted, and classify and train according to the pathological results showing the degree of atrophy, to improve the recognition effect of the model. [ 3] Labeling of atrophy sites. Our classification model only provides diagnostic results for the image, although the heatmap can provide some indications of the atrophy site. Ideally, the area of atrophy should be outlined precisely in the image, providing a more visual aid to the endoscopists. However, this is a challenging task. [ 4] The performance of AI may have been overestimated. All images in the experiment were selected by the endoscopists, which may have led to overfitting of the model. Although we used videos to simulate a real environment, the videos did not include other lesions. With the inclusion of other lesions, the recognition ability of the model requires further validation, which is one of our future works. In this study, we constructed a deep learning-based CAG recognition model with higher diagnostic performance than that of endoscopists. This can provide an objective and reliable diagnostic basis for endoscopists. ## Data availability statement The original contributions presented in the study are included in the article/Supplementary Material. Our code can be get from GitHub, URL: https://github.com/flyingfatpig/GAM-EfficientNet. Further inquiries can be directed to the corresponding authors. ## Ethics statement This study was approved by the ethics committees of the two relevant hospitals involved (No. 202201016, Zibo Central Hospital. No. 202210019, Zhangdian Maternal and Child Health Hospital). Patients in the video test group provided written informed consent before participation. The ethics committee waived informed consent for patients who participated in retrospective imaging. 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 YS and BL conceived the idea. BL built and trained the model. YS wrote the manuscript with support from the rest of the authors. NW, NL, and JW collected the data, analyzed the experimental results, and produced the figures. HW validated the experimental data. TT advised the project and revised 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/fonc.2023.1122247/full#supplementary-material ## References 1. Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin (2021) 71. DOI: 10.3322/caac.21660 2. Correa P. A human model of gastric carcinogenesis. Cancer Res (1988) 48 3. Banks M, Graham D, Jansen M, Gotoda T, Coda S, di Pietro M. British Society of gastroenterology guidelines on the diagnosis and management of patients at risk of gastric adenocarcinoma. Gut (2019) 68. DOI: 10.1136/gutjnl-2018-318126 4. Weck MN, Brenner H. Prevalence of chronic atrophic gastritis in different parts of the world. Cancer Epidemiol Biomarkers Prev (2006) 15. DOI: 10.1158/1055-9965.EPI-05-0931 5. Asaka M, Kato M, Kudo M, Katagiri M, Nishikawa K, Koshiyama H. Atrophic changes of gastric mucosa are caused by helicobacter pylori infection rather than aging: studies in asymptomatic Japanese adults. Helicobacter (1996) 1. DOI: 10.1111/j.1523-5378.1996.tb00008.x 6. You WC, Zhang L, Gail MH, Li JY, Chang YS, Blot WJ. Precancerous lesions in two counties of China with contrasting gastric cancer risk. Int J Epidemiol (1998) 27. DOI: 10.1093/ije/27.6.945 7. Lahner E, Zagari RM, Zullo A, Di Sabatino A, Meggio A, Cesaro P. Chronic atrophic gastritis: Natural history, diagnosis and therapeutic management. a position paper by the Italian society of hospital gastroenterologists and digestive endoscopists [AIGO], the Italian society of digestive endoscopy [SIED], the Italian society of gastroenterology [SIGE], and the Italian society of internal medicine [SIMI]. Digestive Liver Dis (2019) 51. DOI: 10.1016/j.dld.2019.09.016 8. Eshmuratov A, Nah JC, Kim N, Lee HS, Lee HE, Lee BH. The correlation of endoscopic and histological diagnosis of gastric atrophy. Dig Dis Sci (2010) 55. DOI: 10.1007/s10620-009-0891-4 9. Kanawong R, Obafemi-Ajayi T, Ma T, Xu D, Li S, Duan Y. Automated tongue feature extraction for ZHENG classification in traditional Chinese medicine. Evid Based Complement Alternat Med (2012) 2012. DOI: 10.1155/2012/912852 10. Gargeya R, Leng T. Automated identification of diabetic retinopathy using deep learning. Ophthalmology (2017) 124. DOI: 10.1016/j.ophtha.2017.02.008 11. Choi E, Schuetz A, Stewart WF, Sun J. Using recurrent neural network models for early detection of heart failure onset. J Am Med Inform Assoc (2017) 24. DOI: 10.1093/jamia/ocw112 12. Medved D, Ohlsson M, Höglund P, Andersson B, Nugues P, Nilsson J. Improving prediction of heart transplantation outcome using deep learning techniques. Sci Rep (2018) 8 3613. DOI: 10.1038/s41598-018-21417-7 13. Helwan A, Uzun Ozsahin D. Sliding window based machine learning system for the left ventricle localization in MR cardiac images. Appl Comput Intell Soft Computing (2017) 2017. DOI: 10.1155/2017/3048181 14. Yan Z, Zhan Y, Peng Z, Liao S, Shinagawa Y, Zhang S. Multi-instance deep learning: Discover discriminative local anatomies for bodypart recognition. IEEE Trans Med Imaging (2016) 35. DOI: 10.1109/TMI.2016.2524985 15. Ueyama H, Kato Y, Akazawa Y, Yatagai N, Komori H, Takeda T. Application of artificial intelligence using a convolutional neural network for diagnosis of early gastric cancer based on magnifying endoscopy with narrow-band imaging. J Gastroenterol Hepatol (2021) 36. DOI: 10.1111/jgh.15190 16. Shichijo S, Nomura S, Aoyama K, Nishikawa Y, Miura M, Shinagawa T. Application of convolutional neural networks in the diagnosis of helicobacter pylori infection based on endoscopic images. EBioMedicine (2017) 25. DOI: 10.1016/j.ebiom.2017.10.014 17. LeCun Y, Boser B, Denker JS, Henderson D, Howard RE, Hubbard W. Backpropagation applied to handwritten zip code recognition. Neural Comput (1989) 1. DOI: 10.1162/neco.1989.1.4.541 18. Krizhevsky A, Sutskever I, Hinton GE. ImageNet classification with deep convolutional neural networks. Commun ACM (2017) 60 84-90. DOI: 10.1145/3065386 19. Zhao S-B, Yang W, Wang S-L, Pan P, Wang R-D, Chang X. Establishment and validation of a computer-assisted colonic polyp localization system based on deep learning. WJG (2021) 27. DOI: 10.3748/wjg.v27.i31.5232 20. Guimarães P, Keller A, Fehlmann T, Lammert F, Casper M. Deep-learning based detection of gastric precancerous conditions. Gut (2020) 69 4-6. DOI: 10.1136/gutjnl-2019-319347 21. Mu G, Zhu Y, Niu Z, Li H, Wu L, Wang J. Expert-level classification of gastritis by endoscopy using deep learning: a multicenter diagnostic trial. Endosc Int Open (2021) 9. DOI: 10.1055/a-1372-2789 22. Tytgat GN. The Sydney system: endoscopic division. endoscopic appearances in gastritis/duodenitis. J Gastroenterol Hepatol (1991) 6. DOI: 10.1111/j.1440-1746.1991.tb01469.x 23. Fang J-Y, Du YQ, Liu WZ, Ren JL, Li YQ, Chen XY. Chinese Society of gastroenterology, Chinese medical association. Chinese consensus on chronic gastritis (2017, shanghai). J Dig Dis (2018) 19 182-203. DOI: 10.1111/1751-2980.12593 24. Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, Gomez AN. Attention is all you need. (2017) 25. Dosovitskiy A, Beyer L, Kolesnikov A, Weissenborn D, Zhai X, Unterthiner T. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv [Preprint] (2021). DOI: 10.48550/arXiv.2010.11929 26. Liu Z, Lin Y, Cao Y, Hu H, Wei Y, Zhang Z. Swin transformer: Hierarchical vision transformer using shifted windows. (2021). DOI: 10.1109/ICCV48922.2021.00986 27. Li S, Sui X, Luo X, Xu X, Liu Y, Goh R. Medical image segmentation using squeeze-and-Expansion transformers. (2021). DOI: 10.24963/ijcai.2021/112 28. Xie E, Wang W, Wang W, Sun P, Xu H, Liang D. Segmenting Transparent Objects in the Wild with Transformer. (2021). DOI: 10.24963/ijcai.2021/165 29. Ren S, He K, Girshick R, Sun J. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks. IEEE Trans Pattern Anal Mach Intell (2017) 39. DOI: 10.1109/TPAMI.2016.2577031 30. Lin T-Y, Goyal P, Girshick R, He K, Dollar P. Focal Loss for Dense Object Detection. (2017). DOI: 10.1109/ICCV.2017.324 31. Tan M, Le Q. EfficientNetV2: Smaller models and faster training. (2021) 32. Liu Z, Mao H, Wu C-Y, Feichtenhofer C, Darrell T, Xie S. A ConvNet for the 2020s. (2022). DOI: 10.1109/CVPR52688.2022.01167 33. Deng J, Dong W, Socher R, Li L-J, Kai L, Li F-F. ImageNet: A large-scale hierarchical image database. (2009). DOI: 10.1109/CVPR.2009.5206848 34. Russakovsky O, Deng J, Su H, Krause J, Satheesh S, Ma S. ImageNet Large scale visual recognition challenge. Int J Comput Vis (2015) 115. DOI: 10.1007/s11263-015-0816-y 35. Pan SJ, Yang Q. A survey on transfer learning. IEEE Trans Knowl Data Eng (2010) 22. DOI: 10.1109/TKDE.2009.191 36. Day O, Khoshgoftaar TM. A survey on heterogeneous transfer learning. J Big Data (2017) 4 29. DOI: 10.1186/s40537-017-0089-0 37. Liang G, Zheng L. A transfer learning method with deep residual network for pediatric pneumonia diagnosis. Comput Methods Programs BioMed (2020) 187 104964. DOI: 10.1016/j.cmpb.2019.06.023 38. Huang C, Lv W, Zhou C, Mao L, Xu Q, Li X. Discrimination between transient and persistent subsolid pulmonary nodules on baseline CT using deep transfer learning. Eur Radiol (2020) 30. DOI: 10.1007/s00330-020-07071-6 39. Ridnik T, Ben-Baruch E, Noy A, Zelnik-Manor L. ImageNet-21K pretraining for the masses. arXiv [Preprint] (2021). DOI: 10.48550/arXiv.2104.10972 40. Cao R. Towards Accelerated and Robust Rreinforcement Learning with Transfer Learning. (2022). DOI: 10.1109/BDICN55575.2022.00069 41. Liu Y, Shao Z, Hoffmann N. Global attention mechanism: Retain information to enhance channel-spatial interactions. arXiv[Preprint] (2021). DOI: 10.48550/arXiv.2112.05561 42. He T, Zhang Z, Zhang H, Zhang Z, Xie J, Li M. Bag of Tricks for Image Classification with Convolutional Neural Networks. (2019). DOI: 10.1109/CVPR.2019.00065 43. Aggarwal R, Sounderajah V, Martin G, Ting DSW, Karthikesalingam A, King D. Diagnostic accuracy of deep learning in medical imaging: a systematic review and meta-analysis. NPJ Digit Med (2021) 4 65. DOI: 10.1038/s41746-021-00438-z 44. Haenssle HA, Fink C, Schneiderbauer R, Toberer F, Buhl T, Blum A. Man against machine: diagnostic performance of a deep learning convolutional neural network for dermoscopic melanoma recognition in comparison to 58 dermatologists. Ann Oncol (2018) 29. DOI: 10.1093/annonc/mdy166 45. Sollini M, Antunovic L, Chiti A, Kirienko M. Towards clinical application of image mining: a systematic review on artificial intelligence and radiomics. Eur J Nucl Med Mol Imaging (2019) 46. DOI: 10.1007/s00259-019-04372-x 46. Barredo Arrieta A, Díaz-Rodríguez N, Del Ser J, Bennetot A, Tabik S, Barbado A. Explainable artificial intelligence (XAI): Concepts, taxonomies, opportunities and challenges toward responsible AI. Inf Fusion (2020) 58 82-115. DOI: 10.1016/j.inffus.2019.12.012 47. Tjoa E, Guan C. A survey on explainable artificial intelligence (XAI): Toward medical XAI. IEEE Trans Neural Netw Learn Syst (2021) 32. DOI: 10.1109/TNNLS.2020.3027314
--- title: Comprehensive analysis of the prognosis, tumor microenvironment, and immunotherapy response of SDHs in colon adenocarcinoma authors: - Han Nan - Pengkun Guo - Jianing Fan - Wen Zeng - Chonghan Hu - Can Zheng - Bujian Pan - Yu Cao - Yiwen Ge - Xiangyang Xue - Wenshu Li - Kezhi Lin journal: Frontiers in Immunology year: 2023 pmcid: PMC10025334 doi: 10.3389/fimmu.2023.1093974 license: CC BY 4.0 --- # Comprehensive analysis of the prognosis, tumor microenvironment, and immunotherapy response of SDHs in colon adenocarcinoma ## Abstract ### Background Succinate dehydrogenase (SDH), one of the key enzymes in the tricarboxylic acid cycle, is mainly found in the mitochondria. SDH consists of four subunits encoding SDHA, SDHB, SDHC, and SDHD. The biological function of SDH is significantly related to cancer progression. Colorectal cancer (CRC) is one of the most common malignant tumors globally, whose most common histological subtype is colon adenocarcinoma (COAD). However, the correlation between SDH factors and COAD remains unclear. ### Methods The data on pan-cancer was obtained from The Cancer Genome Atlas (TCGA) database. Kaplan-Meier survival analysis showed the prognostic ability of SDHs. The cBioPortal database reflected genetic variations of SDHs. The correlation analysis was conducted between SDHs and mitochondrial energy metabolism genes (MMGs) and the protein-protein interaction (PPI) network was built. Consequently, Univariate and Multivariate Cox Regression Analysis on SDHs and other clinical characteristics were conducted. A nomogram was established. The ssGSEA analysis visualized the association between SDHs and immune infiltration. Immunophenoscore (IPS) explored the correlation between SDHs and immunotherapy, and the correlation between SDHs and targeted therapy was investigated through Genomics of Drug Sensitivity in Cancer. Finally, qPCR and immunohistochemistry detected SDHs’ expression. ### Results After assessing SDHs differential expression in pan-cancer, we found that SDHB, SDHC, and SDHD benefit COAD patients. The cBioPortal database demonstrated that SDHA was the top gene in mutation frequency rank. Correlation analysis mirrored a strong link between SDHs and MMGs. We formulated a nomogram and found that SDHB, SDHC, SDHD, and clinical characteristics correlated with COAD patients’ survival. For T helper cells, Th2 cells, and Tem, SDHA, SDHB, SDHC, and SDHD were significantly enriched in the high expression group. Moreover, COAD patients with high SDHA expression were more suitable for immunotherapy. And COAD patients with different SDHs’ expression have different sensitivity to targeted drugs. Further verifying the gene and protein expression levels of SDHs, we found that the tissues were consistent with the bioinformatics analysis. ### Conclusions Our study analyzed the expression and prognostic value of SDHs in COAD, explored the pathway mechanisms involved, and the immune cell correlations, indicating that SDHs might be biomarkers for COAD patients. ## Introduction Colon adenocarcinoma (COAD) is the most common histological subtype, accounting for more than $90\%$ of CRC [1]. According to statistics, Colorectal cancer (CRC) ranks third among incident cases in both men and women and the third most lethal cancer worldwide in 2022 [2]. New treatments for COAD have developed with advances in surgery and medicine, but long-term survival rates of patients remain considerably lower [3]. Therefore, searching for potential cancer biomarkers and developing new-targeted drugs for immunotherapy may become essential research directions in COAD. In recent years, tumor immunotherapy has been used in high-incidence malignancies such as colon cancer [4], non-small-cell lung [5], and triple-negative breast cancer [6], which would activate immunologic cells to attack the tumor by cell metabolic signaling pathway. In contrast to normal differentiated cells, which rely primarily on mitochondrial oxidative phosphorylation to generate the energy needed for cellular processes, most cancer cells instead rely on aerobic glycolysis, a phenomenon termed “the Warburg effect”. Mitochondria are the powerhouses of cells, and aerobic glycolysis is considered the primary metabolic phenotype of tumor cells, which meet the challenges of high energy demand for rapid cancer cell division and migration by enhancing glycolysis exhibited under aerobic conditions [7]. In terms of tumor metabolism, enhanced glycolysis phenotype reflects the progression of tumor development [8]. To illustrate, enhanced glycolysis regulates pancreatic cancer metastasis [9], and colorectal cancer metastasis [10]. Additionally, a pan-cancer analysis of glycolysis with TCGA database regarded increased tumor glycolytic activity as inferior survival in various cancers [11]. A report indicated that succinate dehydrogenases (SDHs) are closely related to mitochondria and are primarily involved in the occurrence and progression of tumors [12, 13]. Moreover, succinate, which accumulates as a result of SDH inhibition, inhibits HIF-α prolyl hydroxylases in the cytosol, stabilizes and activates Hypoxia-inducible factor-1 (HIF-1) [14]. HIF-1, a transcription factor involves in hypoxic induction of glycolysis, leads to malignant transformation [15]. SDH, also known as mitochondrial complex II, is composed of four subunits encoding SDHA, SDHB, SDHC, and SDHD [16]. The structure of the protein comprises a hydrophilic head and a hydrophobic tail. The hydrophilic head protrudes into the mitochondrial matrix, and the hydrophobic tail anchors the protein to the mitochondrial inner membrane [17]. SDH, functioning as the catalytic core, the head portion is composed of the flavoprotein SDHA and the iron sulphur (Fe-S) containing protein SDHB. The membrane domain comprises the SDHC and SDHD subunits, containing a bound heme moiety and a binding site for ubiquinone [17, 18]. There exists a close relationship between malignancies and the expression of succinate and SDH, including SDH mutations, regulation of mRNA expression, and cancer immunosurveillance [16]. SDH mutations have been found in familial paragangliomas and pheochromocytomas (19–24), renal carcinomas [25], and gastrointestinal stromal tumors [26]. Some rare SDH-wt cases have shown that the occurrence of the Carney triad-related gastrointestinal stromal tumors (GISTs) (27–29) or paragangliomas (PGLs) [30] correlated with a decreased mRNA expression of the SDHC subunits. It is reported that SDHC is correlated with increased metastasis-free survival in malignant pheochromocytoma/paraganglioma [31]. Additionally, it is found that there is decreased expression of SDHD in gastric cancer [32]. Nevertheless, SDH factors are rarely reported in COAD, which indicates that the correlation between SDH factors and COAD remains to be explored. In this study, we investigated SDHs’ expression in pan-cancer and prediction in the prognosis of COAD patients. Furthermore, the associations among SDHs, immune infiltration, and immunotherapy are explored. To sum up, our results prompted that SDHs may become novel cancer biomarkers in COAD, which act as an immunomodulatory derivative from the tricarboxylic acid cycle, participating in the occurrence and development of COAD. ## Data collection and variation analysis Fragments per Kilobase Million (FPKM) normalized expression profile data of pan-cancer, including 33 cancers of The Cancer Genome Atlas (TCGA) database, were downloaded from Genomic Data Commons (GDC) database (https://portal.gdc.cancer.gov/) and merged into an expression matrix. According to human gene annotations (Homo_sapiens. GRCh38.101.CRH.GTF), the Ensemble IDs were transformed into gene symbols. Then, the clinical data of patients with 36 Cholangiocarcinoma (CHOL), 453 Colon adenocarcinoma (COAD), 370 Liver hepatocellular carcinoma (LIHC), 165 Rectum adenocarcinoma (READ), and 370 Stomach adenocarcinoma (STAD) were downloaded and combined into another matrix, respectively. The expression matrix of COAD was stored in Table S1, and the clinical characteristics of COAD patients were documented in Table S2. ## Expression and prognostic significance of SDHs in COAD To investigate the difference in gene expression between cancer tissues and normal tissues, we first compared the raw data (Counts) of differentially expressed genes (DEGs) between normal tissues and CHOL, COAD, LIHC, READ, and STAD, respectively, with a threshold of false discovery rate (FDR) < 0.05 by R package “limma”. 168 mitochondrial energy metabolism genes (MMGs) were obtained from KEGG PATHWAY database (https://www.kegg.jp/kegg/pathway.html) [33] and 1476 HIF-1α related genes were downloaded in the GeneCards database (https://www.genecards.org/). After intersecting with MMGs and HIF-1α related genes, a total of 8 DEGs overlapped were recognized. Then, 8 DEGs were used to build the protein-protein interaction (PPI) network by the Search Tool for the Retrieval of Interacting Genes (STRING) 11.0 and visualized in Cytoscape 3.8.2. The expression of 8 DEGs was visualized by R package “pheatmap”. Then, Kaplan-Meier survival analysis, which applied two-sided log-rank tests with a threshold of $p \leq 0.05$, was performed on patients with CHOL, COAD, LIHC, READ, and STAD based on 8 DEGs with R package “survminer”. Additionally, a gene expression omnibus (GEO) dataset, GSE14333, which contained the microarray-based of 226 COAD patients and corresponding clinical data, respectively, were downloaded from GEO website. The workflow of our study was shown in Figure 1. Figure 1: Flowchart of the study process. We first compared the raw data (Counts) of differentially expressed genes (DEGs) between normal tissues and CHOL, COAD, LIHC, READ, and STAD, respectively, with a threshold of false discovery rate (FDR) < 0.05. Ultimately, a total of 2854 DEGs were identified (Figure 2A). Then, the intersection of 2854 DEGs, 168 mitochondrial energy metabolism genes (MMGs), and 1476 HIF-1α related genes included 8 DEGs (ACAT1, HADHA, PFKFB, PPARA, SDHA, SDHB, SDHC, and SDHD) (Figure 2B). 168 MMGs were obtained from KEGG PATHWAY database, and 1476 HIF-1α related genes were downloaded from GeneCards database. The DEGs, MMGs, and HIF-1α-related genes were listed in Table S4. In addition, a protein-protein interaction (PPI) network with 8 DEGs was constructed through the Search Tool for the Retrieval of Interacting Genes (STRING) (Figure 2C, Table S5). According to the PPI network, there exists a strong relationship among SDHA, SDHB, SDHC, and SDHD. After reviewing the literature, we found that SDHA, SDHB, SDHC, and SDHD belong to the family of succinate dehydrogenase (SDH) [16]. Figure 2: Prognostic significance of SDHs. (A) Venn diagram of DEGs in CHOL, COAD, LIHC, READ, and STAD. (B) Venn diagram of DEGs, MMGs, and HIF-1α-related genes. (C) The network for 8 DEGs intersected. (D) Heatmap of SDHA, SDHB, SDHC, and SDHD between 41 normal tissues and 453 COAD patients. (E-H) Kaplan-Meier overall survival of SDHA, SDHB, SDHC, and SDHD in COAD. Figure 2D demonstrates the expression of 8 DEGs in CHOL, COAD, LIHC, READ, and STAD between normal tissues and pathological tissues, and the volcano figures were stored in Supplementary Figure 1. Especially for SDHs in COAD, the expression of SDHA, SDHB, SDHC, and SDHD in normal tissues is higher than that in COAD patients. To have a comprehensive insight into the prognostic value of 8 DEGs, Kaplan-Meier survival analysis was applied to patients with CHOL, COAD, LIHC, READ, and STAD. The results were shown in Supplementary Figure 2 and 3. It’s revealed that SDHB ($$p \leq 0.026$$), SDHC ($$p \leq 0.026$$), and SDHD ($$p \leq 0.018$$) were significantly associated with the prognosis of COAD in Figures 2E–H. The survival time in the high expression group of SDHB, SDHC, and SDHD was longer than that in the low expression group, which indicates that high expression of SDHB, SDHC, and SDHD benefits COAD patients. Additionally, the consistent results obtained from GSE14333 make the conclusion more convincing (Supplementary Figure 4). ## Genetic variations of SDHs in COAD To explore genetic variations of succinate dehydrogenases (SDHs), cBioPortal (http://www.cbioportal.org), a database for cancer genomics data including mutations, and copy number alternations (CNA) from GISTIC, was applied. The mutation profiles of SDHs came from Colorectal Adenocarcinoma (TCGA, PanCancer Atlas) with 526 patients. To explore genetic variations of SDHs, cBioPortal was applied. The mutation profiles of SDHs came from Colorectal Adenocarcinoma (TCGA, PanCancer Atlas) with 526 patients. As shown in Figure 3A, a high mutation rate of SDHs was observed in COAD patients. Among all SDHs, SDHA is regarded as the top gene in mutation frequency rank in COAD patients ($4\%$). Furthermore, the correlation between SDHs copy number alternations (CNA) and expression of mRNA was presented in Figures 3B–E, pointing out that a positive correlation was found between SDHs copy number and mRNA expression in COAD. Figure 3: Somatic mutation of SDHs. (A) Genetic mutation analysis of SDHs. (B-E) Relationship between CNA in SDHs and expression of mRNA. ## Correlation, functional enrichment based on MMGs With the RNAseq data of COAD from TCGA, correlation analysis between SDHs and MMGs was visualized by R package “pheatmap”. Additionally, 4 SDHs as well as 168 MMGs were used to build the PPI network by the STRING and visualized in Cytoscape, which involves 41 genes. With the criteria of FDR < 0.05, Gene Ontology (GO) enrichment and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis were performed utilizing R package “clusterProfiler” based on 41 genes and described by R package “ggplot2”. Considering that SDHs have a strong connection with energy metabolism, correlation analysis was conducted between SDHs and MMGs. The results were shown in Figure 4A, and the detailed data was demonstrated in Table S6, indicating the significant correlations between 4 SDHs and 168 MMGs. In addition, a PPI network with 4 SDHs as well as 168 MMGs was constructed through the STRING, involving 41 elements (Figure 4B, Table S7). Additionally, a correlation analysis between 4 SDHs was shown in Figure 4C, reflecting that SDHs have a strong correlation except for SDHA. Furthermore, Gene Ontology (GO) enrichment and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis were performed to predict the functions and pathways of 41 genes (Table S8). It is demonstrated that these SDHs-related genes involve electron transport chain, respiratory electron transport chain, mitochondrial ATP synthesis coupled electron transport, and respiratory chain complex in GO enrichment analysis (Figure 4D). Additionally, according to KEGG analysis, it is mirrored that these genes are relative to oxidative phosphorylation (OXPHOS), citrate cycle (TCA cycle), and glycolysis/Gluconeogenesis (Figure 4E). The results showed that SDHs-related genes were enriched in electron transport chain, respiratory electron transport chain, cellular respiration, respiratory chain, oxidative phosphorylation, carbon metabolism, and TCA cycle. Figure 4: Correlation, functional enrichment based on MMGs. (A) The correlation between MMGs and SDHs in COAD. (B) The network for 41 genes is based on SDHs and MMGs with the highest correlation. (C) The correlation between different SDHs in COAD. (D, E) The functions and pathways based on 41 genes were predicted by the analysis of GO and KEGG. ## Relationship between the expression of SDHs and the clinical characteristics of patients with COAD To figure out SDHs’ link with the clinical characteristics of patients with COAD, we analyzed the correlation between the expression levels of SDHs and various clinical characteristics, including T stage, N stage, M stage, age, and lymphatic invasion. Furthermore, Univariate and Multivariate Cox Regression Analysis were conducted to test whether SDHs can be considered independent prognostic factors. R package “rms” and “survival” were employed to formulate a nomogram, which is used to individualize the survival probability for 1-year, 3-year, and 5-year overall survival (OS). Then, time-dependent ROC analysis and Calibration curve were applied to evaluate the nomogram’s discrimination and calibration [34]. To figure out the connection between SDHs and the clinical characteristics of patients with COAD, including T stage, N stage, M stage, age, and lymphatic invasion, the violin diagram were drawn in Figures 5A–E. To illustrate, the T stage, N stage, and M stage represent the extent of primary cancer, the regional lymph node involvement, and the distant metastasis, respectively, based on evidence obtained from clinical assessment parameters determined prior to treatment. Additionally, Lymphatic invasion is a yes/no indicator to ask if malignant cells are present in small or thin-walled vessels suggesting lymphatic involvement. It’s pointed out that SDHA and SDHB were low expressed in higher N and M stages (Figures 5B, C). There exists no positive result in age (Figure 5D), however, it is demonstrated that SDHD expression levels were all lower in lymphatic invasion samples (Figure 5E). Summarily, there is a close relationship between SDHs and clinical characteristics. Figure 5: SDH factor expression and the clinical characteristics of patients with COAD. (A-E) SDHA, SDHB, SDHC, and SDHD are concerned with clinical characteristics involving T stage, N stage, M stage, age, and lymphatic invasion. (F) A nomogram to predict the overall survival rate of COAD patients. (G-J) Time-dependent ROC analysis and Calibration curve for the overall survival nomogram model in the discovery group. A dashed diagonal line represents the ideal nomogram. p < 0.05; p < 0.01 and p < 0.001; ns, not significant. To testify whether SDHs can be regarded as independent prognostic factors, Univariate and Multivariate Cox Regression Analysis were employed in COAD patients. The results were demonstrated in Table 1, and the risk score of the nomogram and the coefficient of clinical characteristics were documented in Table S9. It’s revealed that SDHB, SDHC, SDHD, and clinical characteristics involving T stage, N stage, M stage, age, and lymphatic invasion were correlated with the survival of COAD patients. A nomogram is formulated based on independent prognostic factors to predict the survival probability individually (Figure 5F). For each COAD patient,1-, 3-, and 5-year survival rates would be predicted by the total points in the nomogram accor to 8 indicators. To assess the sensitivity and specificity of this nomogram, time-dependent receiver operating characteristic (ROC) analysis was adopted. The ROC area under the curve (AUC) is 0.798 for 1-year, 0.780 for 3-year, and 0.705 for 5-year survival, representing an efficient predictive efficacy (Figure 5G). Then, the Calibration curve was applied to evaluate the nomogram’s discrimination and calibration, reflecting an ideal capacity of the nomogram for effectively predicting the prognosis of COAD patients (Figures 5H–J). Table 1 \| Characteristics \| Total (N) \| Univariate analysis \| Univariate analysis.1 \| \| --- \| --- \| --- \| --- \| \| Characteristics \| Total (N) \| Hazard ratio (95% CI) \| P value \| \| T stage \| 452 \| \| \| \| T1 \| 11 \| Reference \| \| \| T2 \| 77 \| 0.453 (0.088-2.347) \| 0.346 \| \| T3 \| 308 \| 1.326 (0.325-5.409) \| 0.694 \| \| T4 \| 56 \| 3.826 (0.893-16.394) \| 0.071 \| \| N stage \| 453 \| \| \| \| N0 \| 266 \| Reference \| \| \| N1 \| 105 \| 1.635 (0.991-2.695) \| 0.054 \| \| N2 \| 82 \| 3.997 (2.549-6.266) \| <0.001 \| \| M stage \| 396 \| \| \| \| M0 \| 332 \| Reference \| \| \| M1 \| 64 \| 4.327 (2.763-6.776) \| <0.001 \| \| Age \| 453 \| \| \| \| <=65 \| 188 \| Reference \| \| \| >65 \| 265 \| 1.649 (1.077-2.526) \| 0.021 \| \| Lymphatic invasion \| 410 \| \| \| \| NO \| 247 \| Reference \| \| \| YES \| 163 \| 2.315 (1.520-3.525) \| <0.001 \| \| SDHA \| 453 \| \| \| \| Low \| 226 \| Reference \| \| \| High \| 227 \| 0.818 (0.554-1.209) \| 0.314 \| \| SDHB \| 453 \| \| \| \| Low \| 226 \| Reference \| \| \| High \| 227 \| 0.637 (0.429-0.948) \| 0.026 \| \| SDHC \| 453 \| \| \| \| Low \| 226 \| Reference \| \| \| High \| 227 \| 0.640 (0.432-0.947) \| 0.026 \| \| SDHD \| 453 \| \| \| \| Low \| 226 \| Reference \| \| \| High \| 227 \| 0.620 (0.418-0.921) \| 0.018 \| \| High \| 227 \| 1.207 (0.816-1.786) \| 0.345 \| ## Association between SDHs and immune infiltration To characterize the immune microenvironment of patients with COAD, based on the expression matrix of SDHs, ssGSEA analysis was performed to visualize the correlation between SDHs and immune infiltration level of 24 immune cell types through R package “GSVA”. Correlation analysis was applied to clarify the SDHs expression in connection with the expressions of immune-related genes. The Tumor Immune Single-Cell Hub (TISCH) database (http://tisch.comp-genomics.org/home/), a scRNA-seq database focusing on the tumor microenvironment, was employed to analyze the correlations between SDHs expression and infiltrating immune cells [35]. Gene expression data was gained from the GEO database (GSE146771), including 10468 single cells from 10 patients. The expression of SDHs in different cell types based in GSE146771 was visualized using TISCH. To investigate the connection between SDHs and immune cells, the ssGESA analysis was performed (Supplementary Figures 5A-D, Table S10). Among 24 immune cells, there exists a close relationship between SDHs and Tem (Effective Memory T Cell), Tcm (Central Memory T cell), T helper cells, Th2 (T helper 2) cells, NK CD56bright cells, and NK cells (Figures 6A–D). Surprisingly, for T helper cells and Th2 cells, SDHs were significantly enriched in the high expression group. As for Tem, SDHs were significantly enriched in the low expression group. Additionally, for Tcm, SDHA and SDHB were significantly enriched in the low expression group, while SDHC and SDHD were enriched in the high expression group. For NK cells, SDHs were significantly enriched in the low expression group except for SHDA. However, SDHA was significantly enriched in the high expression group, while SDHC and SDHD were enriched in the low expression group for NK CD56bright cells. We also examined the correlations between the SDHs expression and the expressions of mark genes of immune cells in Figure 6E. It’s indicated that SDHs, especially SDHB, were negatively related to CD56, which is the marker gene of NK cells. In addition, there exists a strong correlation between SDHD and MBD2, a marker gene of Tcm. For CD44 and IL15RA, two genes related to Tcm and Tem, all SDHs are positively correlated with them, especially SDHA, SDHB, and SDHD. To figure out SDHs’ expression in different immune cell types, we analyzed single-cell sequencing datasets of GSE146771 from the Tumor Immune Single-Cell Hub (TISCH) database. In Supplementary Figure 6, GSE146771 was divided into 13 cell types. Focusing on the lower left corner of UMAP plots, we can see that SDHs mainly enriched in CD4Tconv cells, CD8T cells, CD8Tex cells, Treg cells, Tprolif cells, and NK cells, which is consistent with ssGSEA results. Figure 6: Association between SDHs and immune infiltration. (A-D) The ssGSEA analysis based on SDH factor expression for COAD and different types of immune cells. (E) The correlations between the SDHs expression and the expressions of mark genes of immune cells. ## Immunotherapy outcomes prediction The correlation heatmap between SDHs and each immunosuppressive and immunostimulatory gene was visualized by R package “pheatmap”. A total of 18 immunosuppressive genes including ADORA2A, BTLA, CD244, CD274, CD96, CSF1R, CTLA4, HAVCR2, IL10RB, KDR, LAG3, LGALS9, PDCD1, PDCD1LG2, PVRL2, TGFB1, TGFBR1, and TIGIT were selected. A total of 18 MHC molecules including B2M, HLA-A, HLA-B, HLA-C, HLA-DMA, HLA-DOA, HLA-DPA1, HLA-DPB1, HLA-DQA1, HLA-DQA2, HLA-DQB1, HLA-DRA, HLA-DRB1, HLA-E, HLA-F, HLA-G, TAP1, TAP2, and TAPBP were selected. A total of 43 immunostimulatory genes including C10orf54, CD27, CD276, CD28, CD40, CD40LG, CD48, CD70, CD80, CD86, CXCL12, CXCR4, ENTPD1, HHLA2, ICOS, ICOSLG, IL2RA, IL6, IL6R, KLRC1, KLRK1, LTA, MICB, NT5E, PVR, RAET1E, TMEM173, TMIGD2, TNFRSF13B, TNFRSF13C, TNFRSF14, TNFRSF18, TNFRSF25, TNFRSF4, TNFRSF8, TNFRSF9, TNFSF13, TNFSF13B, TNFSF14, TNFSF15, TNFSF4, and TNFSF9 were selected. The Cancer Immunome Atlas (https://tcia.at/) characterized the intratumoral immune landscapes and the cancer antigenomes from 20 solid cancers. The immunophenoscore (IPS) data of COAD patients was extracted for the following analysis to predict the response to immunotherapy, including the anti-PD-1/PD-L1 treatment and anti-CTLA-4 treatment scores. The microsatellite instability (MSI) was downloaded from cBioPortal and the consensus molecular subtypes (CMS) was obtained from a previous study [36]. To deepen the understanding of the value of SDHs for COAD treatment, the relationships between SDHs and marker genes of immunostimulation, MHC, and immunosuppression were listed in Figures 7A–C and Table S11-S13, respectively. It turns out that SDHs are significantly correlated with these immune-related genes. Unlike other SDHs, SDHA had different correlations with immune genes. Interestingly, some immunosuppressants showed uniform correlations. The results indicated that ADORA2A, CSF1R, CTLA4, KDR, PDCD1LG2, TGFBR1, TGFB1, and CXCL12 had significant negative correlations with SDHs while CD244, IL10RB, KLRC1, and RAET1E had significant positive correlations with SDHs. As for genes of MHC, SDHA was positively correlated with almost all genes, especially HLA-E and TAPBP. IPS is a machine learning-based scoring system that could predict patients’ responses to immunotherapy, including anti-PD-1/PD-L1 and anti-CTLA-4 treatment [37]. Combined analysis of the expression SDHs and IPS score proved that COAD patients with high SDHA expression are more suitable for immunotherapy such as anti-PD-1/PD-L1 ($$p \leq 3.4$$×10-7) and anti-CTLA-4 ($$p \leq 5.6$$×10-6) treatment (Figures 7D, E, Table S14). Furthermore, we explored how the microsatellite instability (MSI) and consensus molecular subtypes (CMS) effect the patients’ possibility to respond to immunotherapy with different SDHA expression. Microsatellite instability (MSI) distribution of patients was displayed in Figure 7F. Specifically, patients in microsatellite stability (MSS) with high SDHA expression are more suitable for immunotherapy such as anti-PD-1/PD-L1 ($$p \leq 9.2$$×10-6) and anti-CTLA-4 ($$p \leq 2.9$$×10-5) treatment (Figures 7G, H). Then, we explored how different CMS effect the possibility to respond immunotherapy in patients in MSS with different SDHA expression. Proportions of CMS in patients in MSS were demonstrated in Figure 7I. Patients in CMS3 and CMS4 with high SDHA expression have a higher possibility to respond to immunotherapy (Figures 7J, K). Figure 7: Immunotherapy outcomes prediction. (A-C) The correlation between SDHs and immunostimulatory, MHC, and immunosuppressive genes. (D, E) The association between SDHs expression and the relative probabilities of responding to immunotherapy, including anti-PD-1/PD-L1 therapy and anti-CTLA-4 therapy. (F) Proportions of MSI and MSS in patients. (G, H) The possibility to respond to immunotherapy based on different SDHA expression and MSI. (I) Proportions of CMS1, CMS2, CMS3, and CMS4 in patients in MSS. (J, K) The possibility to respond to immunotherapy in patients in MSS based on different SDHA expression and CMS. ## Targeted drug therapy outcomes prediction To predict targeted drug therapy outcomes according to SHDs’ expression, R package “pRRophetic” was utilized in Axitinib, Cetuximab, GDC0941, and Gefitinib based on the Genomics of Drug Sensitivity in Cancer (GDSC). The natural log of the half-maximal inhibitory concentration (LN_IC50 value) of chemotherapy drugs was downloaded from the GDSC, using GDSC2 screening set. The box plots were drawn by R package “ggplot2”. To investigate the relationship between SDHs and targeted drug sensitivity, the Genomics of Drug Sensitivity in Cancer (GDSC) of Axitinib, Cetuximab, GDC0941, and Gefitinib was utilized. The result indicated significant differences in targeted drug sensitivity in COAD patients with different SDHs’ expression (Figures 8A–D, Table S15). Specifically, COAD patients with different SDHA expression have different responses to Axitinib, Cetuximab, GDC0941, and Gefitinib. Additionally, with higher SDHB expression, COAD patients are more sensitive to GDC0941 and Gefitinib. However, the drug sensitivity of COAD patients with high expression of SDHC and SDHD is opposite to that of COAD patients with high expression of SDHA. Figure 8: Targeted drug therapy outcomes prediction. (A-D) GDSC predicts the IC50 difference of four drugs between COAD patients with different SDHs expression. ## Validation of SDHs at gene and protein levels The 19 paired COAD tissues were collected from patients who underwent surgical resection for COAD at the Second Affiliated Hospital of Wenzhou Medical University (Wenzhou, China). The corresponding Paraffin section was collected from the Pathology Department of the Second Affiliated Hospital of Wenzhou Medical University (Wenzhou, China). It has passed the examination of the Ethics Committee at Wenzhou Medical University. The protein expression level of SDHs in COAD and normal tissue was verified by immunohistochemistry (IHC). Sections were dewaxed and rehydrated. The catalase blocker blocked endogenous peroxidase activity (ZSGB-BIO), and the antigen was repaired by sodium citrate buffer (pH 6.0). Then, the tissue sections were incubated overnight with rabbit monoclonal anti-SDHA antibody (1:100 dilution, Proteintech), rabbit monoclonal anti-SDHB antibody (1:100 dilution, Santa cruz), rabbit monoclonal anti-SDHC antibody (1:100 dilution, Proteintech), and rabbit monoclonal anti-SDHD antibody (1:100 dilution, Affbiotech) at 4°C, respectively. After the antibodies were washed, the slices were incubated for 30 minutes with goat anti-rabbit IgG at 37 °C. Then, we redyed with hematoxylin the slices, used neutral gum to seal the shee, and observed it under the optical microscope. Additionally, protein expression of SDHs was downloaded from the Proteomic Data Commons (https://proteomic.datacommons.cancer.gov/pdc/). The primers of SDHA, SDHB, SDHC, and SDHD can be found in Table S3. The total RNA was extracted using TRNzol Reagent and was reverse-transcribed with ReverTra Ace®qPCR RT Master Mix with gDNA Remover (TOYOBO, Japan). All qPCR reactions were performed with Hieff® Qpcr SYBR Green Master Mix(Yeasen Biotechnology (Shanghai)) in 20µl volume containing 10µl 2× SYBR Green RT-PCR Master Mix, 0.4µl of each 0.2µM forward and reverse primer, 1µl of cDNA sample, and nuclease-free water up to 20µl. Amplification was carried out according to the following conditions: initial denaturation at 95°C for 5 min, followed by 40 cycles of denaturation at 95°C for 10s, and annealing at 60°C for 30s. The relative expression of the gene was calculated by the 2^-△Ct method. To validate the consistency between the gene level and protein level of SDHs in COAD, we evaluated the protein expressions of SDHs in COAD through Proteomic Data Commons (PDC) database and Immunohistochemistry (IHC). Figure 9A indicated that SHDs, mainly located in the cytoplasm, were mainly expressed in glandular cells. Furthermore, the immunohistochemical staining intensity of SDHA, SDHB, SDHC, and SDHD in normal tissues was more substantial than in COAD tissues, demonstrating that these proteins were more significantly expressed in adjacent colon tissues than in COAD tissues. According to the PDC database, compared with normal tissues, SDHA (Figure 9B, $p \leq 2.2$×10-16), SDHB (Figure 9C, $$p \leq 1.1$$×10-13), SDHC (Figure 9D, $$p \leq 1.2$$ ×10-10), and SDHD (Figure 9E, $$p \leq 0.00028$$) were low expressed in colon cancer at protein level. Figure 9: Validation of SDHs at gene and protein levels. (A) IHC of SDHs in COAD and normal tissues. (B-E) The protein expression of SDHs in COAD in Proteomic Data Commons database. (F-I) The relative mRNA expression level of SDHs in COAD and adjacent normal tissues detected by qPCR. In addition, qPCR with 19 paired tumors and adjacent tissues was performed, suggesting that the mRNA expression of SDHs was significantly different from tumors and adjacent tissues (Figures 9F–I). These results showed that all SDHs have good consistency between gene and protein levels, which was highly expressed in colon tissues and low expressed in colon cancer. ## Discussion A growing body of research proved mitochondrial metabolism plays an essential role in tumorigenesis, metastasis, and treatment resistance (7, 38–42). Succinate dehydrogenase (SDH), a tumor metabolite, acts as an oncogenic signaling molecule in many cellular processes such as metabolic and epigenetic alterations, angiogenic stimulation, migration, invasion, and post-translational modification of proteins [43]. Consequently, we found that high expression of SDHB, SDHC, and SDHD has a better prognosis for COAD patients, reflecting that all of them can be defined as protective factors for COAD by TCGA and GSE14333 data analysis. Mutations in genes are known to be closely linked to the development of malignant tumors. The mutation of SDH in the development and prognosis of several cancers has been partially established (26–30, 44). In Carney triad (CT) patients, a high methylation level of SDHC was found, which was correlated to functional impairment of the SDH complex [29]. And the notable immunohistochemical loss of SDHA in gastrointestinal stromal tumors (GISTs) signals mutation of SDHA [45]. Therefore, we explored the genetic variations of SDHs in COAD through cBioPorta. For SDHs, mutations are positively correlated with mRNA expression. Interestingly, it’s found that in COAD, SDHA is the top gene in mutation frequency rank and is mainly involved in the missense mutation in COAD. COAD progression can be hindered by inhibiting mitochondrial OXPHOS through Lin28a/SDHA signaling pathway [46]. Additionally, SDHA inactivation results in the accumulation of succinate, which binds to and activates thioredoxin reductase 2, a reactive oxygen species-scavenging enzyme, to render chemotherapy resistance in COAD [47]. Therefore, we speculate that SDHA may be involved in the progression and treatment of COAD as a critical gene among the SDHs. To further explore the link between SDHs and energy metabolism in COAD, the correlation analysis and PPI between SDHs and MMGs were conducted, indicating the significant correlations between SDHs and MMGs. Additionally, the correlation analysis between 4 SDHs reflected that SDHs have a strong correlation except for SDHA. Meanwhile, we found that SDHs-related genes were enriched in electron transport chain, OXPHOS, carbon metabolism, and TCA cycle by correlation analysis and functional enrichment analysis. It’s important to note that previous studies have shown that the TCA cycle and carbon metabolism have a particular impact on the prognosis of patients with COAD [48, 49]. It has been reported that SDHB gene knockout in the human pheochromocytoma cell line (HPheo1) up-regulates genes involved in glycolysis and down-regulates genes involved in OXPHOS [50]. Glycolylysis-dependent impaired OXPHOS has also been shown in familial renal cancer patients with germline mutations of the SDHB gene [51]. Our analysis revealed that SDHs play a role in the TCA cycle and metabolism process pathway. In terms of tumor metabolism, the glycolysis/oxidative phosphorylation (OXPHOS) ratio is of great significance in tumorigenesis. In recent years, cancer immunotherapy has generally drawn the public’s attention, which was named 2013’s Breakthrough of the Year by Science [52]. Up to now, checkpoint inhibitors have been the most thoroughly investigated class of immunotherapy. So far, five PD-1 or PD-L1 inhibitors and one CTLA4 inhibitor have been approved to treat various cancers based on improvements in overall survival [53]. However, many patients do not respond to treatment with checkpoint inhibitors. The factors underlying responsiveness to checkpoint inhibitors are being intensely studied [54]. When activated, T cells express programmed cell death 1 (PD-1) for recognizing abnormal and cancerous cells [55, 56]. cytotoxic T lymphocyte antigen 4 (CTLA4), is a co-inhibitory molecule that regulates the extent of T cell activation. blocks the interaction between CTLA4 and these ligands, CD80 and CD86, and keeps T cells remain active, which can recognize and kill tumor cells [57]. It has been reported that succinic acid plays a role in the cancer microenvironment and regulates many metabolic pathways through G protein-coupled receptors [58]. It is thus clear that as an essential intermediate product of the tricarboxylic acid (TCA) cycle, succinate and SDHs extend beyond metabolism and enter anticancer immunity [59]. To investigate the connection between SDHs and immune infiltration, we explored the association between SDHs and immune infiltration. In our study, the degree of immune infiltration of T helper cells was closely related to the expression of SDHs, which may be caused by the enrichment of SDH in T helper cells leading to enhancement of mitochondrial activity. It’s known that T helper cells are essential for protective immunity and play a role in inflammatory responses to self-antigens or nonharmful allergens [60]. Metabolic inhibition decreased T-cell proliferation and activation or led to T-cell anergy or cell death (61–63). Moreover, the low expression level of SDHs was correlated to functional impairment of the SDH complex because of the Warburg effect [64]. Nevertheless, the specific function of the Warburg effect in activated T cells remains unclear [65]. The functional mechanism of energy metabolism of SDHs on COAD needs to be further explored. SDHB, SDHC, and SDHD showed high similarity in our correlation analysis between SDHs and marker genes of immunosuppression and immunostimulation. SDHA is regarded as a new target to mitigate T cell-mediated intestinal diseases including alloimmune gastrointestinal graft versus host disease (GI-GVHD), autoimmune inflammatory bowel disease (IBD), and iatrogenic CTLA-4Ig ICB-mediated colitis [66] because this reduction in SDHA caused an enhanced sensitivity of the intestinal epithelial cells (IECs) to T cell-mediated cytotoxicity [67, 68]. Our analysis proved that SDHA, positively correlated with most of these gene signatures, has a peculiar pattern regarding gene signatures compared to other SDHs. Additionally, our results indicated that SDHA is significantly associated with Lymphocyte activation gene 3 protein (LAG3), which provides a new direction for immunotherapy in patients with COAD. Highly correlated with LAG3, adoptive cell therapy using tumor-infiltrating lymphocytes (TILs) was a promising immunotherapy approach for COAD [69]. Through immunophenoscore (IPS), COAD patients with high SDHA expression are more suitable for immunotherapy such as anti-PD-1/PD-L1 and anti-CTLA-4 treatment. In fact, it’s known that the majority of COAD patients in microsatellite instability (MSS) were less sensitive to immune checkpoint inhibitors than the minority of COAD patients in microsatellite instability (MSI) [70]. In our study, MSS patients with different SDHA expression have different possibility to respond to immunotherapy. With high SDHA expression, MSS patients can benefit more from immunotherapy. Consensus molecular subtypes (CMS) groups CRC samples according to their gene-signature in four subtypes: CMS1 (MSI Immune), CMS2 (Canonical), CMS3 (Metabolic), and CMS4 (Mesenchymal) [36]. Patients in CMS3 are demonstrated enrichment for multiple metabolism signatures, while patients in CMS4 are likely to be diagnosed at more advanced stages and have poor survival [71]. For both CMS3 and CMS4 patients in MSS, higher SDHA expression was associated with better treatment outcomes, indicating that SDHA might become a new biomarker for predicting the outcomes of immune checkpoint blockades such as anti-PD-1/PD-L1 and anti-CTLA-4. Target drugs such as Axitinib [72], Cetuximab [73], GDC0941 [74], and Gefitinib [52] have been applied to clinical practice. However, the most recent adjuvant clinical trials have not shown any value for adding targeted agents, like cetuximab, to standard chemotherapies in stage III disease, despite improved outcomes in the metastatic setting [75]. Additionally, pathologic features [76], MSI [77], Mutations of BRAF, KRAS, and PIK3CA [78], supervised prognostic genomic signatures [79], and unsupervised gene expression molecular subtypes [80] all contribute to the definition of optimal adjuvant treatments for patients. Nevertheless, none of the gene signatures known to date can predict benefits from therapy in COAD [75]. In our study, the Wilcoxon rank sum test demonstrated the significant influence of SDHs’ expression level on targeted drug sensitivity, showing the great potential for SDH to predict benefit from therapy in COAD. With the help of SDHs’ expression level, we would predict targeted drug therapy outcomes more precisely. Furthermore, based on the properties of SDHA targeting immune checkpoints to regulate immune infiltration, we believe that SDHA may be a crucial gene in the SDHs family, which plays an essential role in the development of immunotherapy and targeted drug therapy of COAD. ## Conclusions To sum up, our study comprehensively assessed the expression and prognostic value of SDHs in COAD and explored the pathway mechanisms involved and the immune cell correlations. Our findings suggested that SDHs might be potential biomarkers indicating the prognosis and therapeutic efficacy for patients with COAD and were associated with COAD immune microenvironment. ## 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/s ## Ethics statement The studies involving human participants were reviewed and approved by Medical Ethics Committee of the Second Affiliated Hospital of Wenzhou Medical University. The patients/participants provided their written informed consent to participate in this study. ## Author contributions HN and KL designed the study, analyzed the data, and drafted the paper. KL, XX, and WL critically revised it for important intellectual content. HN, JF, WZ, YC, and YG assisted in data acquisition and analysis. RNA extraction, reverse transcription, and qPCR were performed by PG, and WZ. IHC was performed by HN, PG, CH, CZ, and BP. All authors revised the manuscript. HN, PG, and JF contributed equally to this work. 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.2023.1093974/full#supplementary-material ## References 1. Barresi V, Reggiani Bonetti L, Ieni A, Caruso RA, Tuccari G. Histological grading in colorectal cancer: New insights and perspectives. Histol histopathol (2015) 30. DOI: 10.14670/hh-11-633 2. Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A. Global cancer statistics 2020: Globocan estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin (2021) 71. DOI: 10.3322/caac.21660 3. Haraldsdottir S, Einarsdottir HM, Smaradottir A, Gunnlaugsson A, Halfdanarson TR. [Colorectal cancer - review]. Laeknabladid (2014) 100 75-82. DOI: 10.17992/lbl.2014.02.531 4. Dai Z, Peng X, Guo Y, Shen X, Ding W, Fu J. Metabolic pathway-based molecular subtyping of colon cancer reveals clinical immunotherapy potential and prognosis. J Cancer Res Clin Oncol (2022). DOI: 10.1007/s00432-022-04070-6 5. Monaco L, Gemelli M, Gotuzzo I, Bauckneht M, Crivellaro C, Genova C. Metabolic parameters as biomarkers of response to immunotherapy and prognosis in non-small cell lung cancer (Nsclc): A real world experience. Cancers (Basel) (2021) 13. DOI: 10.3390/cancers13071634 6. Gong Y, Ji P, Yang YS, Xie S, Yu TJ, Xiao Y. Metabolic-Pathway-Based subtyping of triple-negative breast cancer reveals potential therapeutic targets. Cell Metab (2021) 33 51-64.e9. DOI: 10.1016/j.cmet.2020.10.012 7. Weinhouse S. On respiratory impairment in cancer cells. Science (1956) 124. DOI: 10.1126/science.124.3215.267 8. Jia D, Park JH, Jung KH, Levine H, Kaipparettu BA. Elucidating the metabolic plasticity of cancer: Mitochondrial reprogramming and hybrid metabolic states. Cells (2018) 7. DOI: 10.3390/cells7030021 9. Yang J, Ren B, Yang G, Wang H, Chen G, You L. The enhancement of glycolysis regulates pancreatic cancer metastasis. Cell Mol Life Sci (2020) 77. DOI: 10.1007/s00018-019-03278-z 10. Zhao S, Guan B, Mi Y, Shi D, Wei P, Gu Y. Lncrna Mir17hg promotes colorectal cancer liver metastasis by mediating a glycolysis-associated positive feedback circuit. Oncogene (2021) 40. DOI: 10.1038/s41388-021-01859-6 11. Liu YC, Lin P, Zhao YJ, Wu LY, Wu YQ, Peng JB. Pan-cancer analysis of clinical significance and associated molecular features of glycolysis. Bioengineered (2021) 12. DOI: 10.1080/21655979.2021.1955510 12. Porporato PE, Filigheddu N, Pedro JMB, Kroemer G, Galluzzi L. Mitochondrial metabolism and cancer. Cell Res (2018) 28. DOI: 10.1038/cr.2017.155 13. Chen Z, Han Z, Nan H, Fan J, Zhan J, Zhang Y. A novel pyroptosis-related gene signature for predicting the prognosis and the associated immune infiltration in colon adenocarcinoma. Front Oncol (2022) 12. DOI: 10.3389/fonc.2022.904464 14. Gaude E, Frezza C. Defects in mitochondrial metabolism and cancer. Cancer Metab (2014) 2. DOI: 10.1186/2049-3002-2-10 15. Hu CJ, Iyer S, Sataur A, Covello KL, Chodosh LA, Simon MC. Differential regulation of the transcriptional activities of hypoxia-inducible factor 1 alpha (Hif-1alpha) and hif-2alpha in stem cells. Mol Cell Biol (2006) 26. DOI: 10.1128/MCB.26.9.3514-3526.2006 16. Dalla Pozza E, Dando I, Pacchiana R, Liboi E, Scupoli MT, Donadelli M. Regulation of succinate dehydrogenase and role of succinate in cancer. Semin Cell Dev Biol (2020) 98 4-14. DOI: 10.1016/j.semcdb.2019.04.013 17. Yankovskaya V, Horsefield R, Törnroth S, Luna-Chavez C, Miyoshi H, Léger C. Architecture of succinate dehydrogenase and reactive oxygen species generation. Sci (New York NY) (2003) 299. DOI: 10.1126/science.1079605 18. Sun F, Huo X, Zhai Y, Wang A, Xu J, Su D. Crystal structure of mitochondrial respiratory membrane protein complex II. Cell (2005) 121. DOI: 10.1016/j.cell.2005.05.025 19. Hao HX, Khalimonchuk O, Schraders M, Dephoure N, Bayley JP, Kunst H. Sdh5, a gene required for flavination of succinate dehydrogenase, is mutated in paraganglioma. Sci (New York NY) (2009) 325. DOI: 10.1126/science.1175689 20. Astuti D, Latif F, Dallol A, Dahia PL, Douglas F, George E. Gene mutations in the succinate dehydrogenase subunit sdhb cause susceptibility to familial pheochromocytoma and to familial paraganglioma. Am J Hum Genet (2001) 69 49-54. DOI: 10.1086/321282 21. Niemann S, Muller U. Mutations in sdhc cause autosomal dominant paraganglioma, type 3. Nat Genet (2000) 26. DOI: 10.1038/81551 22. Baysal BE, Ferrell RE, Willett-Brozick JE, Lawrence EC, Myssiorek D, Bosch A. Mutations in sdhd, a mitochondrial complex II gene, in hereditary paraganglioma. Science (2000) 287. DOI: 10.1126/science.287.5454.848 23. Bardella C, Pollard PJ, Tomlinson I. Sdh mutations in cancer. Biochim Biophys Acta (2011) 1807. DOI: 10.1016/j.bbabio.2011.07.003 24. Zong WX, Rabinowitz JD, White E. Mitochondria and cancer. Mol Cell (2016) 61. DOI: 10.1016/j.molcel.2016.02.011 25. Vanharanta S, Buchta M, McWhinney SR, Virta SK, Peczkowska M, Morrison CD. Early-onset renal cell carcinoma as a novel extraparaganglial component of sdhb-associated heritable paraganglioma. Am J Hum Genet (2004) 74. DOI: 10.1086/381054 26. Stratakis CA, Carney JA. The triad of paragangliomas, gastric stromal tumours and pulmonary chondromas (Carney triad), and the dyad of paragangliomas and gastric stromal sarcomas (Carney-stratakis syndrome): Molecular genetics and clinical implications. J Intern Med (2009) 266 43-52. DOI: 10.1111/j.1365-2796.2009.02110.x 27. Killian JK, Miettinen M, Walker RL, Wang Y, Zhu YJ, Waterfall JJ. Recurrent epimutation of sdhc in gastrointestinal stromal tumors. Sci Transl Med (2014) 6 268ra177. DOI: 10.1126/scitranslmed.3009961 28. Urbini M, Astolfi A, Indio V, Heinrich MC, Corless CL, Nannini M. Sdhc methylation in gastrointestinal stromal tumors (Gist): A case report. BMC Med Genet (2015) 16 87. DOI: 10.1186/s12881-015-0233-7 29. Haller F, Moskalev EA, Faucz FR, Barthelmeß S, Wiemann S, Bieg M. Aberrant DNA hypermethylation of sdhc: A novel mechanism of tumor development in Carney triad. Endocrine-related Cancer (2014) 21. DOI: 10.1530/erc-14-0254 30. Richter S, Klink B, Nacke B, de Cubas AA, Mangelis A, Rapizzi E. Epigenetic mutation of the succinate dehydrogenase c promoter in a patient with two paragangliomas. J Clin Endocrinol Metab (2016) 101. DOI: 10.1210/jc.2015-3856 31. Span PN, Rao JU, Oude Ophuis SB, Lenders JW, Sweep FC, Wesseling P. Overexpression of the natural antisense hypoxia-inducible factor-1alpha transcript is associated with malignant Pheochromocytoma/Paraganglioma. Endocrine-related Cancer (2011) 18. DOI: 10.1530/erc-10-0184 32. Habano W, Sugai T, Nakamura S, Uesugi N, Higuchi T, Terashima M. Reduced expression and loss of heterozygosity of the sdhd gene in colorectal and gastric cancer. Oncol Rep (2003) 10. DOI: 10.3892/or.10.5.1375 33. Ye Z, Zhang H, Kong F, Lan J, Yi S, Jia W. Comprehensive analysis of alteration landscape and its clinical significance of mitochondrial energy metabolism pathway-related genes in lung cancers. Oxid Med Cell Longevity (2021) 2021. DOI: 10.1155/2021/9259297 34. Hu T, Chen Z, Hou M, Lin K. Overall and cancer-specific survival in patients with breast paget disease: A population-based study. Exp Biol Med (Maywood) (2022) 247. DOI: 10.1177/15353702211056264 35. Han Y, Wang Y, Dong X, Sun D, Liu Z, Yue J. Tisch2: Expanded datasets and new tools for single-cell transcriptome analyses of the tumor microenvironment. Nucleic Acids Res (2023) 51. DOI: 10.1093/nar/gkac959 36. Guinney J, Dienstmann R, Wang X, de Reynies A, Schlicker A, Soneson C. The consensus molecular subtypes of colorectal cancer. Nat Med (2015) 21. DOI: 10.1038/nm.3967 37. Charoentong P, Finotello F, Angelova M, Mayer C, Efremova M, Rieder D. Pan-cancer immunogenomic analyses reveal genotype-immunophenotype relationships and predictors of response to checkpoint blockade. Cell Rep (2017) 18. DOI: 10.1016/j.celrep.2016.12.019 38. Ward PS, Thompson CB. Metabolic reprogramming: A cancer hallmark even warburg did not anticipate. Cancer Cell (2012) 21 297-308. DOI: 10.1016/j.ccr.2012.02.014 39. Viale A, Corti D, Draetta GF. Tumors and mitochondrial respiration: A neglected connection. Cancer Res (2015) 75. DOI: 10.1158/0008-5472.CAN-15-0491 40. LeBleu VS, O'Connell JT, Gonzalez Herrera KN, Wikman H, Pantel K, Haigis MC. Pgc-1alpha mediates mitochondrial biogenesis and oxidative phosphorylation in cancer cells to promote metastasis. Nat Cell Biol (2014) 16 992-1003, 1-15. DOI: 10.1038/ncb3039 41. Porporato PE, Payen VL, Perez-Escuredo J, De Saedeleer CJ, Danhier P, Copetti T. A mitochondrial switch promotes tumor metastasis. Cell Rep (2014) 8. DOI: 10.1016/j.celrep.2014.06.043 42. Li Y, Zheng JY, Liu JQ, Yang J, Liu Y, Wang C. Succinate/Nlrp3 inflammasome induces synovial fibroblast activation: Therapeutical effects of clematichinenoside ar on arthritis. Front Immunol (2016) 7. DOI: 10.3389/fimmu.2016.00532 43. Sciacovelli M, Frezza C. Oncometabolites: Unconventional triggers of oncogenic signalling cascades. Free Radic Biol Med (2016) 100. DOI: 10.1016/j.freeradbiomed.2016.04.025 44. Oh TS, Hutchins DC, Mainali R, Goslen KH, Quinn MA. Itaconate and its derivatives repress early myogenesis in vitro and in vivo. Front Immunol (2022) 13. DOI: 10.3389/fimmu.2022.748375 45. Miettinen M, Killian JK, Wang ZF, Lasota J, Lau C, Jones L. Immunohistochemical loss of succinate dehydrogenase subunit a (Sdha) in gastrointestinal stromal tumors (Gists) signals sdha germline mutation. Am J Surg Pathol (2013) 37. DOI: 10.1097/PAS.0b013e3182671178 46. Liu YD, Zhuang XP, Cai DL, Cao C, Gu QS, Liu XN. Let-7a regulates ev secretion and mitochondrial oxidative phosphorylation by targeting Snap23 in colorectal cancer. J Exp Clin Cancer Res (2021) 40 31. DOI: 10.1186/s13046-020-01813-6 47. Du Z, Liu X, Chen T, Gao W, Wu Z, Hu Z. Targeting a Sirt5-positive subpopulation overcomes multidrug resistance in wild-type kras colorectal carcinomas. Cell Rep (2018) 22. DOI: 10.1016/j.celrep.2018.02.037 48. Zhang Z, Zhu H, Li Q, Gao W, Zang D, Su W. Gene expression profiling of tricarboxylic acid cycle and one carbon metabolism related genes for prognostic risk signature of colon carcinoma. Front Genet (2021) 12. DOI: 10.3389/fgene.2021.647152 49. Fedorova MS, Krasnov GS, Lukyanova EN, Zaretsky AR, Dmitriev AA, Melnikova NV. The cimp-high phenotype is associated with energy metabolism alterations in colon adenocarcinoma. BMC Med Genet (2019) 20 52. DOI: 10.1186/s12881-019-0771-5 50. Tabebi M, Kumar Dutta R, Skoglund C, Soderkvist P, Gimm O. Loss of sdhb induces a metabolic switch in the Hpheo1 cell line toward enhanced oxphos. Int J Mol Sci (2022) 23. DOI: 10.3390/ijms23010560 51. Shuch B, Linehan WM, Srinivasan R. Aerobic glycolysis: A novel target in kidney cancer. Expert Rev Anticancer Ther (2013) 13. DOI: 10.1586/era.13.57 52. Chen L, Meng Y, Guo X, Sheng X, Tai G, Zhang F. Gefitinib enhances human colon cancer cells to trail-induced apoptosis of. Apoptosis (2016) 21. DOI: 10.1007/s10495-016-1287-5 53. Ellis PM, Vella ET, Ung YC. Immune checkpoint inhibitors for patients with advanced non-Small-Cell lung cancer: A systematic review. Clin Lung Cancer (2017) 18 444-59.e1. DOI: 10.1016/j.cllc.2017.02.001 54. Maleki Vareki S, Garrigos C, Duran I. Biomarkers of response to pd-1/Pd-L1 inhibition. Crit Rev Oncol Hematol (2017) 116. DOI: 10.1016/j.critrevonc.2017.06.001 55. Alsaab HO, Sau S, Alzhrani R, Tatiparti K, Bhise K, Kashaw SK. Pd-1 and pd-L1 checkpoint signaling inhibition for cancer immunotherapy: Mechanism, combinations, and clinical outcome. Front Pharmacol (2017) 8. DOI: 10.3389/fphar.2017.00561 56. Munn DH, Bronte V. Immune suppressive mechanisms in the tumor microenvironment. Curr Opin Immunol (2016) 39 1-6. DOI: 10.1016/j.coi.2015.10.009 57. Webb ES, Liu P, Baleeiro R, Lemoine NR, Yuan M, Wang YH. Immune checkpoint inhibitors in cancer therapy. J BioMed Res (2018) 32. DOI: 10.7555/JBR.31.20160168 58. Mossa AH, Velasquez Flores M, Cammisotto PG, Campeau L. Succinate, increased in metabolic syndrome, activates Gpr91 receptor signaling in urothelial cells. Cell Signal (2017) 37. DOI: 10.1016/j.cellsig.2017.05.014 59. Jiang S, Yan W. Succinate in the cancer-immune cycle. Cancer Lett (2017) 390. DOI: 10.1016/j.canlet.2017.01.019 60. Zhu J. T Helper cell differentiation, heterogeneity, and plasticity. Cold Spring Harb Perspect Biol (2018) 10. DOI: 10.1101/cshperspect.a030338 61. Endo Y, Asou HK, Matsugae N, Hirahara K, Shinoda K, Tumes DJ. Obesity drives Th17 cell differentiation by inducing the lipid metabolic kinase, Acc1. Cell Rep (2015) 12. DOI: 10.1016/j.celrep.2015.07.014 62. Raha S, Raud B, Oberdörfer L, Castro CN, Schreder A, Freitag J. Disruption of. Eur J Immunol (2016) 46. DOI: 10.1002/eji.201546152 63. Lee J, Walsh MC, Hoehn KL, James DE, Wherry EJ, Choi Y. Regulator of fatty acid metabolism, acetyl coenzyme a carboxylase 1, controls T cell immunity. J Immunol (Baltimore Md 1950) (2014) 192. DOI: 10.4049/jimmunol.1302985 64. Eng C, Kiuru M, Fernandez MJ, Aaltonen LA. A role for mitochondrial enzymes in inherited neoplasia and beyond. Nat Rev Cancer (2003) 3 193-202. DOI: 10.1038/nrc1013 65. Almeida L, Lochner M, Berod L, Sparwasser T. Metabolic pathways in T cell activation and lineage differentiation. Semin Immunol (2016) 28. DOI: 10.1016/j.smim.2016.10.009 66. Wu SR, Reddy P. Tissue tolerance: A distinct concept to control acute gvhd severity. Blood (2017) 129. DOI: 10.1182/blood-2016-09-740431 67. Fujiwara H, Seike K, Brooks MD, Mathew AV, Kovalenko I, Pal A. Mitochondrial complex ii in intestinal epithelial cells regulates T cell-mediated immunopathology. Nat Immunol (2021) 22. DOI: 10.1038/s41590-021-01048-3 68. Izquierdo JM. Sdha: A key player in T cell-mediated intestinal disease severity. Cell Mol Immunol (2022) 19. DOI: 10.1038/s41423-021-00820-7 69. Kitsou M, Ayiomamitis GD, Zaravinos A. High expression of immune checkpoints is associated with the til load, mutation rate and patient survival in colorectal cancer. Int J Oncol (2020) 57. DOI: 10.3892/ijo.2020.5062 70. Le DT, Uram JN, Wang H, Bartlett BR, Kemberling H, Eyring AD. Pd-1 blockade in tumors with mismatch-repair deficiency. N Engl J Med (2015) 372. DOI: 10.1056/NEJMoa1500596 71. Van Cutsem E, Labianca R, Bodoky G, Barone C, Aranda E, Nordlinger B. Randomized phase iii trial comparing biweekly infusional Fluorouracil/Leucovorin alone or with irinotecan in the adjuvant treatment of stage iii colon cancer: Petacc-3. J Clin Oncol (2009) 27. DOI: 10.1200/JCO.2008.21.6663 72. Rugo HS, Herbst RS, Liu G, Park JW, Kies MS, Steinfeldt HM. Phase I trial of the oral antiangiogenesis agent Ag-013736 in patients with advanced solid tumors: Pharmacokinetic and clinical results. J Clin Oncol (2005) 23. DOI: 10.1200/JCO.2005.04.192 73. Adams GP, Weiner LM. Monoclonal antibody therapy of cancer. Nat Biotechnol (2005) 23. DOI: 10.1038/nbt1137 74. Semaan C, Neilson KA, Brown B, Molloy MP. Pharmacological inhibition of casein kinase 2 enhances the effectiveness of Pi3k inhibition in colon cancer cells. Anticancer Res (2018) 38. DOI: 10.21873/anticanres.12973 75. Dienstmann R, Salazar R, Tabernero J. Personalizing colon cancer adjuvant therapy: Selecting optimal treatments for individual patients. J Clin Oncol (2015) 33. DOI: 10.1200/JCO.2014.60.0213 76. Roth AD, Delorenzi M, Tejpar S, Yan P, Klingbiel D, Fiocca R. Integrated analysis of molecular and clinical prognostic factors in stage Ii/Iii colon cancer. J Natl Cancer Inst (2012) 104. DOI: 10.1093/jnci/djs427 77. Hutchins G, Southward K, Handley K, Magill L, Beaumont C, Stahlschmidt J. Value of mismatch repair, kras, and braf mutations in predicting recurrence and benefits from chemotherapy in colorectal cancer. J Clin Oncol (2011) 29. DOI: 10.1200/JCO.2010.30.1366 78. Innocenti F, Ou FS, Qu X, Zemla TJ, Niedzwiecki D, Tam R. Mutational analysis of patients with colorectal cancer in Calgb/Swog 80405 identifies new roles of microsatellite instability and tumor mutational burden for patient outcome. J Clin Oncol (2019) 37. DOI: 10.1200/JCO.18.01798 79. Gray RG, Quirke P, Handley K, Lopatin M, Magill L, Baehner FL. Validation study of a quantitative multigene reverse transcriptase-polymerase chain reaction assay for assessment of recurrence risk in patients with stage II colon cancer. J Clin Oncol (2011) 29. DOI: 10.1200/JCO.2010.32.8732 80. Budinska E, Popovici V, Tejpar S, D'Ario G, Lapique N, Sikora KO. Gene expression patterns unveil a new level of molecular heterogeneity in colorectal cancer. J Pathol (2013) 231 63-76. DOI: 10.1002/path.4212
--- title: 'Helicobacter pylori infection increase the risk of subclinical hyperthyroidism in middle-aged and elderly women independent of dietary factors: Results from the Tianjin chronic low-grade systemic inflammation and health cohort study in China' authors: - Juanjuan Zhang - Xinghua Hai - Siyu Wang - Fan Zhu - Yeqing Gu - Ge Meng - Qing Zhang - Li Liu - Hongmei Wu - Shunming Zhang - Tingjing Zhang - Xing Wang - Shaomei Sun - Ming Zhou - Qiyu Jia - Kun Song - Kaijun Niu journal: Frontiers in Nutrition year: 2023 pmcid: PMC10025335 doi: 10.3389/fnut.2023.1002359 license: CC BY 4.0 --- # Helicobacter pylori infection increase the risk of subclinical hyperthyroidism in middle-aged and elderly women independent of dietary factors: Results from the Tianjin chronic low-grade systemic inflammation and health cohort study in China ## Abstract ### Background Prospective studies on the association between Helicobacter pylori (H. pylori) infection and subclinical hyperthyroidism are limited. We, therefore, designed a large-scale cohort study to explore the association between H. pylori infection and the risk of subclinical hyperthyroidism in women. ### Methods This prospective cohort study investigated 2,713 participants. H. pylori infection was diagnosed with the carbon 13 breath test. Subclinical hyperthyroidism was defined as serum thyroid-stimulating hormone levels are low or undetectable but free thyroxine and tri-iodothyronine concentrations are normal. Propensity score matching (PSM) analyses and Cox proportional hazards regression models were used to estimate the association between H. pylori infection and subclinical hyperthyroidism. ### Results A total of 1,025 PS-matched pairs of H. pylori infection women were generated after PSM. During 6 years of follow-up, the incidence rate of subclinical hyperthyroidism was $\frac{7.35}{1}$,000 person-years. After adjusting potential confounding factors (including iodine intake in food and three main dietary patterns score), the multivariable hazard ratio (HR; $95\%$ confidence intervals) of subclinical hyperthyroidism by H. pylori infection was 2.49 (1.36, 4.56). Stratified analyses suggested a potential effect modification by age, the multivariable HR ($95\%$ confidence intervals) was 2.85 (1.45, 5.61) in participants aged ≥ 40 years and 0.70 (0.08, 6.00) in participants aged < 40 years (P for interaction = 0.048). ### Conclusion Our prospective study first indicates that H. pylori infection is significantly associated with the risk of subclinical hyperthyroidism independent of dietary factors among Chinese women, especially in middle-aged and older individuals. Clinical Trial Registration:https://upload.umin.ac.jp/cgi-open-bin/ctr_e/ctr_view.cgi?recptno=R000031137, identifier UMIN000027174. ## Introduction Thyroid hormones (THs), including triiodothyronine (T3) and thyroxine (T4), are produced and released by the thyroid gland [1]. They are key regulators of basal metabolic rate and are essential for the normal growth and development of a body. Synthesis and secretion of THs are finely modulated by the hypothalamic–pituitary–thyroid (HPT) axis. Thyroid-stimulating hormone (TSH) is produced by the anterior pituitary, stimulating the synthesis and secretion of THs by negative feedback inhibition of the HPT axis [2]. Subclinical hyperthyroidism (SHyper) is defined biochemically: TSH concentrations are low or undetectable but free thyroxine (FT4) and tri-iodothyronine (FT3) concentrations are normal [3]. The common causes of subclinical hyperthyroidism include toxic multinodular goiter, toxic adenoma, Graves’ disease, and some exogenous causes [4]. The prevalence of subclinical hyperthyroidism in the general population is $1\%$–$3\%$ [5, 6], which varies by age, sex, race, genetic predisposition, iodine status, and the definition of subclinical hyperthyroidism [7]. Women, older individuals, and those living in iodine-deficient regions were more frequent [8]. In the past decade, the adverse effects of subclinical hyperthyroidism on health have been explored extensively, which has previously been associated with an increased risk for cardiovascular disease, bone loss, fractures, and dementia, even may progression to overt hyperthyroidism [4], which have become an important public health problem. Helicobacter pylori (H. pylori) is a major human pathogen that specifically colonizes the gastric epithelium and infects the stomach of $44.3\%$ ($50.8\%$ in developing countries vs. $34.7\%$ in developed countries) of the world population [9, 10]. Although most infected individuals do not develop obvious clinical sequelae, H. pylori is a Group I carcinogen according to the International Agency for Research on Cancer (IARC), with $89\%$ of all gastric cancers being attributable to this infection [9]. Helicobacter pylori infection can also involve some extra gastric diseases, such as respiratory (bronchiectasis and asthma), cardiovascular (atherosclerosis, myocardial infarction), Parkinson’s disease, metabolic syndrome, fatty liver disease, and immune-mediated allergic diseases [11, 12]. Evidence suggests that H. pylori infection plays an important role in the pathogenesis of thyroid autoimmune diseases due to its ability to mimic antigen distribution on thyroid cell membranes [13, 14]. In addition, De Luis et al. [ 15] showed that the titer of anti- H. pylori immunoglobulin G (IgG) antibody in patients with subclinical hyperthyroidism was significantly higher than that in the control group. Furthermore, a recent meta-analysis [16] involving 862 patients showed that H. pylori infection was significant in Graves’ disease but not in Hashimoto thyroiditis. To date, specific data linking H. pylori infection with subclinical hyperthyroidism in human populations has been limited and conflicting, and most of the thyroid diseases show woman predilection [17], therefore, we conducted a large prospective cohort study to investigate whether baseline H. pylori infections were associated with subclinical hyperthyroidism in woman adults. Moreover, evidence has illustrated that dietary factors may play a crucial role in subclinical hyperthyroidism [18], which also directly influences H. pylori colonization or virulence [19]. Therefore, while studying the association between H. pylori infection and subclinical hyperthyroidism, we must consider the interference of dietary factors in the association. Owing to it being a retrospective cohort, propensity score matching (PSM) was applied to reduce selection bias when comparing the clinical outcomes between patients with and without H. pylori infections. ## Participants Tianjin Chronic Low-grade Systemic Inflammation and Health (TCLSIH) Cohort Study is a prospective dynamic cohort study of participants over 18 years of age, focusing on the relationship between chronic low-grade systemic inflammation and health status [20, 21]. This is an ongoing study that was launched in 2007, where the participants underwent health examinations and completed a questionnaire survey to assess their diet and lifestyle factors till May 2013 [22]. Refer to previous reports for detailed information [23]. A total of 3,585 women had received at least one health examination and questionnaire survey, including blood tests, thyroid function tests, 13C-urea breath test (13C-UBT), and lifestyle factors. During the research period. Participants who did not complete the questionnaire survey at baseline ($$n = 85$$) were excluded. In addition, we excluded participants who had a history of cardiovascular disease ($$n = 262$$) or cancer ($$n = 48$$), given that cardiovascular disease and cancer can significantly affect the lifestyle of participants. We also excluded participants who had a history of thyroid disease ($$n = 175$$) at baseline, subjects with subclinical hyperthyroidism at baseline ($$n = 97$$), and 205 participants who were lost to follow-up examinations. The final cohort analysis comprised 2,713 participants (follow-up rate, $93.0\%$). The participant selection process is described in Figure 1. All participants have agreed to participate and provided written informed consent. The protocol of the study is in accordance with the Declaration of Helsinki and was approved by the Institutional Review Board of Tianjin Medical University. Figure 1: Flow diagram of study participant selection. ## Thyroid function tests After fasting for 8–12 h, fasting blood samples were collected by elbow vein puncture in the morning. Serum FT3, FT4, and TSH were measured by chemiluminescence immunoassay using the ADVIA Centaur FT3, FT4, and TSH3-Ultra analyzer (Siemens Healthcare Diagnostics, New York, NY). The ranges of measurement for FT3, FT4, and TSH were 0.3–30.8 and 1.3–155 pmol/L, and 0.001–150 mIU/L, respectively. Based on the previous reports and national guidelines [23, 24], the normal ranges of FT3, FT4, and TSH were 3.50–6.50 and 11.50–22.70 pmol/L, and 0.55–4.78 mIU/L, respectively. We used a uniform thyrotropin cutoff level for subclinical hyperthyroidism definition: TSH < 0.55 IU/mL and normal FT4 levels. ## Assessment of Helicobacter pylori infection The diagnosis of H. pylori infection was based on the result of fasting 13C-UBT, which is based on the simple principle that the abundantly expressed urease of H. pylori can rapidly hydrolyze isotopically labeled urea solutions. A baseline breath sample was obtained by blowing into a 20-mL container, and a capsule containing 75 mg of 13C-urea was given to patients with 100 ml of water. Two breath samples were collected within a 30-min interval. Patient samples were analyzed by gas chromatography and H. pylori infection was considered present if the difference between the 30-min value and baseline value divided by the baseline value exceeded 4.0‰. 13C-UBT is a mature diagnostic test for H. pylori infection due to its rapid, non-invasive, high sensitivity, and specificity [25]. ## Assessment of other variables Data on potential confounders-such as demographic characteristics, socioeconomic status, lifestyle factors, reproductive factors, and family history of CVD, hypertension, diabetes, and hyperlipidemia were obtained from a standardized health-related questionnaire survey. The total energy intake for each participant and dietary intake was assessed using a 100-item Food Frequency Questionnaire (FFQ). Spearman rank correlation coefficient for energy intake between two FFQs administered 3 months apart was 0.68, and for food groups (e.g., fruits, vegetables, and soft drinks) ranged from 0.62 to 0.79. Details of the FFQ have been described elsewhere [26]. Factor analysis with principal components based on the FFQ was used to derive three dietary patterns (fruit and sweet foods dietary pattern, vegetable dietary pattern, and animal foods dietary pattern; Supplementary Table 1) [27]. Anthropometric parameters, blood pressure, and blood tests are done by trained staff at the hospital. Body mass index (BMI) was calculated as weight/height2 (kg/m2). Metabolic syndrome (MetS) was defined according to the 2009 American Heart Association scientific criteria [28]. Details of the various tests have been described in a previous study [29]. ## Statistical analysis All statistical analyses were performed using SAS software, version 9.4 (SAS Institute Inc., Cary, NC, United States). We summarized baseline characteristics based on H. pylori infection using the mean and $95\%$ confidence interval (CI) for continuous variables and counts (with percentages) for categorical variables. Propensity scores were calculated using a logistic regression model and the main covariates (refer to Table 1). We used nearest neighbor matching to match H. pylori infection and H. pylori uninfected patients in a 1:1 ratio with a caliper distance of 0.2 of the standard deviation of the logit of the PS [30]. Standardized differences were used to assess the covariate balance after matching. An absolute standardized difference of less than 0.1 was considered negligible in the groups [31]. After the cases and controls were randomly sorted, the control with the closest propensity score was selected for the first case. If the propensity score difference between the two was within the caliper range, the matching was successful; if it was outside the caliper range, the matching failed. If successful, the case and control were removed from the original database and matched to the next case. If no suitable matched control was found for the last case, it was deleted. Table 1 \| Characteristics \| H. pylori infection status (before matching) \| H. pylori infection status (before matching).1 \| P b \| H. pylori infection status (after matching) \| H. pylori infection status (after matching).1 \| P b.1 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Characteristics \| No \| Yes \| No \| Yes \| \| \| \| No. of participants \| 1426 \| 1287 \| \| 1025 \| 1025 \| \| \| Age (y) \| 46.7 (46.2, 47.2) \| 48.2 (47.6, 48.7) \| <0.0001 \| 46.7 (46.2, 47.2) \| 46.5 (46.0, 47.1) \| 0.73 \| \| BMI (kg/m2) \| 23.6 (23.4, 23.7) \| 24.0 (23.8, 24.2) \| <0.001 \| 23.6 (23.4, 23.7) \| 23.7 (23.5, 23.8) \| 0.37 \| \| Total energy intake (kcal/day) \| 1875.9 (1848.5, 1903.7) \| 1835.4 (1807.6, 1863.6) \| 0.04 \| 1876.6 (1849.4, 1904.1) \| 1868.6 (1841.5, 1896.0) \| 0.68 \| \| High-sensitivity C-reactive protein (ug/L) \| 1.32 (1.21, 1.42) \| 1.16 (1.05, 1.26) \| 0.03 \| 1.31 (1.21, 1.42) \| 1.21 (1.10, 1.31) \| 0.14 \| \| PA (≥23MET × hour/week, %) \| 29.9 \| 28.9 \| 0.53 \| 29.8 \| 29.9 \| 0.94 \| \| Smoking status (%) \| Smoking status (%) \| Smoking status (%) \| Smoking status (%) \| Smoking status (%) \| Smoking status (%) \| Smoking status (%) \| \| Current smoker \| 2.35 \| 2.90 \| 0.34 \| 2.37 \| 2.67 \| 0.58 \| \| Ex-smoker \| 1.03 \| 0.72 \| 0.36 \| 0.91 \| 0.97 \| 0.86 \| \| Non-smoker \| 96.6 \| 96.4 \| 0.71 \| 96.7 \| 96.4 \| 0.57 \| \| Alcohol drinking status (%) \| Alcohol drinking status (%) \| Alcohol drinking status (%) \| Alcohol drinking status (%) \| Alcohol drinking status (%) \| Alcohol drinking status (%) \| Alcohol drinking status (%) \| \| Everyday drinker \| 0.74 \| 0.98 \| 0.45 \| 0.75 \| 0.80 \| 0.85 \| \| Sometime drinker \| 43.1 \| 44.8 \| 0.34 \| 43.15 \| 41.13 \| 0.23 \| \| Ex-drinker \| 8.12 \| 8.10 \| 0.98 \| 8.17 \| 9.30 \| 0.24 \| \| Non-drinker \| 48.0 \| 46.1 \| 0.27 \| 47.9 \| 48.6 \| 0.70 \| \| Educational level (≥college grade, %) \| 47.7 \| 40.6 \| <0.0001 \| 47.5 \| 47.1 \| 0.84 \| \| Occupation (%) \| Occupation (%) \| Occupation (%) \| Occupation (%) \| Occupation (%) \| Occupation (%) \| Occupation (%) \| \| Managers \| 32.6 \| 27.0 \| <0.001 \| 32.5 \| 33.0 \| 0.74 \| \| Professionals \| 11.7 \| 11.1 \| 0.59 \| 11.7 \| 12.0 \| 0.76 \| \| Other \| 55.7 \| 61.9 \| <0.001 \| 55.9 \| 55.0 \| 0.61 \| \| Depressive symptoms score (≥45, %) \| 13.2 \| 15.2 \| 0.10 \| 13.2 \| 13.4 \| 0.88 \| \| Metabolic syndrome \| 20.1 \| 21.3 \| 0.38 \| 20.2 \| 20.4 \| 0.90 \| \| Amenorrhea (Yes, %) \| 31.2 \| 34.3 \| 0.05 \| 31.2 \| 29.9 \| 0.40 \| \| Household income (≥10,000 Yuan, %) \| 41.6 \| 38.2 \| 0.04 \| 41.7 \| 41.5 \| 0.92 \| \| Family history of diseases (%) \| Family history of diseases (%) \| Family history of diseases (%) \| Family history of diseases (%) \| Family history of diseases (%) \| Family history of diseases (%) \| Family history of diseases (%) \| \| CVD \| 35.8 \| 36.9 \| 0.51 \| 35.9 \| 35.3 \| 0.70 \| \| Hypertension \| 56.6 \| 54.6 \| 0.24 \| 56.6 \| 57.0 \| 0.79 \| \| Hyperlipidemia \| 0.17 \| 0.06 \| 0.35 \| 0.00 \| 0.00 \| - \| \| Diabetes \| 27.9 \| 27.4 \| 0.74 \| 27.8 \| 27.8 \| 1.00 \| \| “Sweets” dietary pattern score \| −0.08 (−0.13, −0.04) \| −0.07 (−0.11, −0.02) \| 0.64 \| −0.08 (−0.13, −0.04) \| −0.08 (−0.12, −0.03) \| 0.82 \| \| “Vegetables” dietary pattern score \| 0.16 (0.12, 0.21) \| 0.13 (0.08, 0.17) \| 0.26 \| 0.16 (0.12, 0.21) \| 0.16 (0.12, 0.21) \| 1.00 \| \| “Animal foods” dietary pattern score \| −0.35 (−0.39, −0.32) \| −0.33 (−0.36, −0.29) \| 0.32 \| −0.35 (−0.39, −0.31) \| −0.34 (−0.38, −0.31) \| 0.77 \| \| Iodine in food (ug/d) \| 144.6 (139.6, 149.7) \| 146.1 (140.8, 151.3) \| 0.70 \| 144.7 (139.6, 149.8) \| 148.0 (142.9, 153.1) \| 0.37 \| Follow-up time was calculated from the date of measuring the baseline measurement of H. pylori infection to the date of the first diagnosis of subclinical hyperthyroidism, or the end of follow-up (31 December 2019), or loss to follow-up, whichever was earliest. Cox proportional hazards models were applied to calculate hazard ratios (HRs) and $95\%$ CIs for the association between H. pylori infection and the risk of subclinical hyperthyroidism. The incidence of subclinical hyperthyroidism was used as the dependent variable. Three models were developed. In Model 1, we adjusted for age, BMI, smoking status, drinking status, education levels, employment status, household income, depressive symptoms score, PA, family history of diseases (including CVD, hypertension, hyperlipidemia, and diabetes), total energy intake, high-sensitivity C-reactive protein (hsCRP), MetS, and amenorrhea status. The total iodine intake in food accounts for a high proportion of iodine intake, as high as $42.7\%$ in the Tianjin population [21], thus, in Model 2, we further adjusted the total iodine intake in food. A study [32] has shown that diet in daily life will have an impact on H. pylori infection, thus, we have adjusted dietary patterns in a final model. Potential interactions between H. pylori infection and main covariates were assessed by adding cross-product terms to the multivariable Cox models. We stratified the participants by potential effect modifiers including age, BMI, physical activity, amenorrhea status, education level, household income, and Mets status. To examine the robustness of our results, we conducted two sensitivity analyses: [1] we substituted adjustment for waist circumference (WC) was measured with BMI to investigate whether abdominal adiposity, compared with overall adiposity, changed the observed associations. [ 2] We included participants who had a history of cancer or CVD. All tests were two-tailed and $p \leq 0.05$ was defined as statistically significant. ## Results Characteristics of participants according to H. pylori infection status before and after PSM are shown in Table 1 and Figure 2. Among 2,713 participants who were available to be analyzed before PSM, $47.4\%$ were classified as infecting H. pylori. Participants with H. pylori infection status tended to be older, and have higher BMI. Moreover, they had lower total energy intake, hsCRP, educational level, and monthly household income, and they were more likely to be managers and had amenorrhea status. After PSM, there were 1,025 PS-matched pairs of H. pylori infection women were generated and showed no significant baseline differences in any characteristic. Between May 2013 and December 2019, a total of 2,050 participants completed the follow-up analysis for subclinical hyperthyroidism. In the total population, 55 first incident cases of subclinical hyperthyroidism occurred, the incidence rate of subclinical hyperthyroidism was 7.56 per 1,000 person-years across the 6-year follow-up period (range: 2–6 years; median: 3.55 years). Figure 2: Standardized mean differences of covariances before and after propensity score matching. Table 2 presents the associations between H. pylori infection and the risk of subclinical hyperthyroidism. In the analysis of H. pylori infection status before PSM, after adjustment for age, BMI, sociodemographic and lifestyle factors as well as for Mets status (Model 1), HR ($95\%$ CI) for subclinical hyperthyroidism was 2.07 (1.06, 4.03; $$p \leq 0.033$$). Further adjustments were made for total iodine intake in food, HRs ($95\%$ CIs) for subclinical hyperthyroidism were 2.05 (1.05, 4.02; $$p \leq 0.036$$), and we observed similar results. In the fully adjusted model further adjusted for the dietary quality (three major dietary patterns), these associations were not altered, and HR ($95\%$ CI) for subclinical hyperthyroidism was 2.06 (1.05, 4.03; $$p \leq 0.035$$). After H. pylori infection status PSM, we observed more obvious associations between H. pylori infection and the risk of subclinical hyperthyroidism. In Model 1, HR ($95\%$ CI) for subclinical hyperthyroidism was 2.34 (1.28, 4.27; $$p \leq 0.006$$). In Model 2, HR ($95\%$ CI) for subclinical hyperthyroidism was 2.41 (1.32, 4.40; $$p \leq 0.004$$). In the fully adjusted model further adjusted for the dietary quality (three major dietary patterns), these associations were not altered, HR ($95\%$ CI) for subclinical hyperthyroidism was 2.49 (1.36, 4.56; $$p \leq 0.003$$). Table 2 \| Unnamed: 0 \| H. pylori infection status (before matching) \| H. pylori infection status (before matching).1 \| P a \| H. pylori infection status (after matching) \| H. pylori infection status (after matching).1 \| P a.1 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| \| No \| Yes \| P a \| No \| Yes \| P a \| \| No. of subjects \| 1426 \| 1287 \| \| 1025 \| 1025 \| \| \| Number of cases \| 25 \| 37 \| \| 19 \| 36 \| \| \| Person-years \| 4124 \| 3828 \| \| 3640 \| 3635 \| \| \| Model 1b \| 1.00 (reference) \| 2.07 (1.06, 4.03)c \| 0.033 \| 1.00(reference) \| 2.34 (1.28, 4.27) \| 0.006 \| \| Model 2d \| 1.00(reference) \| 2.05 (1.05, 4.02) \| 0.036 \| 1.00(reference) \| 2.41 (1.32, 4.40) \| 0.004 \| \| Model 3e \| 1.00(reference) \| 2.06 (1.05, 4.03) \| 0.035 \| 1.00(reference) \| 2.49 (1.36, 4.56) \| 0.003 \| The analyses were stratified by age (<40 or ≥40 years), BMI (< 24.0 or ≥24.0), physical activity (<23 or ≥23 MET-h/wk), MetS (yes or no), amenorrhea status (yes or no), an education level (college graduate or not), and household income (<10,000 or ≥10,000 yuan). The associations between H. pylori infection status and the risk of subclinical hyperthyroidism were generally similar across all subgroups (Table 3). All interactions were not statistically significant (P for interaction > 0.05), except for age (P for interaction = 0.05). Stratified analyses indicated that H. pylori infection was associated with subclinical hyperthyroidism in individuals aged ≥ 40 but not in individuals aged < 40. Similar results were also found in the sensitivity analysis (Table 4). When we adjusted for WC instead of BMI, the full model-adjusted HRs $95\%$ CI for subclinical hyperthyroidism in women were 2.48 (1.36, 4.54; $$p \leq 0.003$$), and the associations did not substantially change. When we included participants with a history of cancer or CVD, the analyses yielded very similar results, the adjusted HRs ($95\%$ CIs) of subclinical hyperthyroidism were 2.03 (1.13, 3.64; $$p \leq 0.019$$). ## Discussion In this large-scale prospective study of a Chinese adult population, we have assessed the association between H. pylori infection and the incidence of subclinical hyperthyroidism in women. Considering the potential effect of iodine on subclinical hyperthyroidism, we conducted this cohort study in Tianjin, an iodine-replete area, and further adjusted for total iodine intake in food in Model 2. Meanwhile, dietary factors are not only associated with H. pylori infection [33] but also affect the risk of subclinical hyperthyroidism [34], however, no previous cohort study adjusted for dietary intake when exploring the associations between H. pylori infection and the incidence of subclinical hyperthyroidism, indicating that the risk effect of H. pylori infection on incident subclinical hyperthyroidism might be misestimated. In this large prospective cohort study, we adjusted for dietary factors in the form of three dietary patterns in Model 3, thus, can prevent the influence of dietary factors on the association between H. pylori infection and subclinical hyperthyroidism. We found that H. pylori infection was associated with an increased risk of incidents of subclinical hyperthyroidism independent of dietary factors in women, a series of sensitivity and subgroup analyses supported these results. To the best of our knowledge, this study is the first prospective cohort investigation regarding the association between H. pylori infection and subclinical hyperthyroidism independent of dietary factors in women. The present results suggested that H. pylori infection was significantly associated with subclinical hyperthyroidism in women, which seems to be in agreement with previous studies [16, 35]. Although the exact mechanisms of H. pylori infection in subclinical hyperthyroidism have not been elucidated, some evidence may explain the association. First, it was recently discovered that H. pylori strains can express fucosylated Lewis determinants, which are widely shared by different host tissues and may stimulate an autoimmune response that could potentially damage the thyroid gland [36]. Second, a previous study has shown that a homologous 11-residue peptide in both gastric parietal cell antigen and thyroid peroxidase suggests the existence of an epitope common to both antigens [37]. Therefore, antibodies are produced during H. pylori infection may cross-react with thyroid antigens, resulting in subclinical hyperthyroidism [38]. Furthermore, evidence that H. pylori infection can increase the risk of subclinical hyperthyroidism also comes from the strong correlation between IgG anti-H. pylori antibodies and thyroid autoantibodies, as well as the observation that thyroid auto-antibodies levels gradually decrease after eradication of H. pylori infection [39]. Taken together, these studies provide potential explanations linking H. pylori infection and subclinical hyperthyroidism. We found a significant association between H. pylori infection and subclinical hyperthyroidism in middle-aged and elderly women but not in younger women. This association may be caused by two reasons. First, the successful colony of H. pylori in the stomach requires age-related gastric physiology and special characteristics related to the host [40], and H. pylori-related diseases are increasing with increasing age [41]. Second, the thyroid gland would undergo important functional changes during aging, the clinical course of thyroid diseases is different between the elderly and the young. A previous study showed a degree of insensitivity in thyroid cells in the anterior pituitary gland in older adults, occurring with age, mainly manifested as the decrease of TSH secretion of thyrotropin-releasing hormone (TRH) in the elderly, and the serum TSH level is usually higher in older than in young people in response to the decrease of thyroid hormone concentration [42]. Further exploration is required to clarify this issue. The major strengths of this study are it first assessed the association between H. pylori infection and subclinical hyperthyroidism in a large-scale adult population from an iodine-replete area. It is a prospective dynamic study design, which allowed us to examine the long-term effect of H. pylori infection on thyroid function before the occurrence of subclinical hyperthyroidism. Furthermore, we applied the PSM in the H. pylori-infected and -uninfected so we could observe the balance of confounding factors between groups more intuitively in comparing matched groups. Thus, the distribution of covariates between the two groups was balanced and the bias in the observational study was reduced. Moreover, we performed several sensitivity analyses to confirm the robustness of the findings. Nonetheless, several biases and limitations should be considered in the present study. First, although our analyses adjusted for a considerable number of confounders, we could not rule out residual confounding by other unmeasured or unknown factors. Second, the clinical outcome of H. pylori infection may be determined by a combination of virulence among H. pylori strains, duration of infection, host genetic polymorphisms, specific host–microbe interactions, and environmental factors [43]. However, our study was based on data from health check-ups; H. pylori strains, eradication of H. pylori infection, and genome-wide association studies (GWAS) were not investigated. CagA-positive strains are endowed with an enhanced inflammatory potential [44] and affect patients with hyperthyroid Graves’s disease more frequently [45]. Previous studies supported the finding that eradicating H. pylori infection could lead to the restitution of the cellular immune response [46] and reduce thyroid autoantibodies [14]. In addition, one study found the presence of a gene encoding an endogenous peroxidase in the dissected chromosome of an H. pylori strain [47]. Therefore, the lack of data on H. pylori might modify the strength of the study results, underestimating the observed associations. Third, although subjects with a self-reported history of all types of thyroid disease, including positive thyroid autoantibodies, were excluded based on a detailed structured questionnaire, serum thyroid autoantibodies were not measured in the current study. Previous reviews [48] have shown that positive thyroid autoantibodies can predict the risk of future thyroid dysfunction, and studies have shown that thyroid autoantibodies reduce the risk of H. pylori infection after eradication [39], suggesting that the lack of data on thyroid autoantibodies may lead to underestimation of the risk of H. pylori infection to subclinical hyperthyroidism. Finally, H. pylori infection rates vary according to geographic region [49], and our study was performed in the general adult Chinese population, therefore, it is possible that our results cannot be generalized to other populations. Further studies are needed to verify the results in other populations. ## Conclusion In conclusion, the findings of this population-based cohort study suggested that H. pylori infection was associated with the risk of subclinical hyperthyroidism independent of dietary factors in adult women. More studies should be performed on different populations to confirm these 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 author/s. ## Ethics statement The studies involving human participants were reviewed and approved by UMIN Clinical Trials Registry. The patients/participants provided their written informed consent to participate in this study. ## Author contributions JZ and XH analyzed data and wrote the paper. SW, FZ, YG, GM, QZ, LL, HW, SZ, TZ, XW, SS, MZ, QJ, and KS conducted research. KN designed research and had primary responsibility for final content. All authors read and approved the final manuscript. ## Funding This study was supported by grants from the National Natural Science Foundation of China (No. 81941024, 81872611, 82103837, and 81903315), Tianjin Major Public Health Science and Technology Project (No. 21ZXGWSY00090), National Health Commission of China (No. SPSYYC 2020015), Food Science and Technology Foundation of Chinese Institute of Food Science and Technology (No. 2019-12), 2014 and 2016 Chinese Nutrition Society (CNS) Nutrition Research Foundation—DSM Research Fund (Nos. 2016-046, 2014-071 and 2016-023), China. ## 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.2023.1002359/full#supplementary-material ## References 1. Mullur R, Liu YY, Brent GA. Thyroid hormone regulation of metabolism. Physiol Rev (2014) 94 355-82. DOI: 10.1152/physrev.00030.2013 2. Chiamolera MI, Wondisford FE. Minireview: thyrotropin-releasing hormone and the thyroid hormone feedback mechanism. Endocrinology (2009) 150 1091-6. DOI: 10.1210/en.2008-1795 3. Cooper DS, Biondi B. Subclinical thyroid disease. Lancet (2012) 379 1142-54. DOI: 10.1016/S0140-6736(11)60276-6 4. Biondi B, Cooper DS. Subclinical hyperthyroidism. N Engl J Med (2018) 378 2411-9. DOI: 10.1056/NEJMcp1709318 5. Alwan H, Villoz F, Feller M, Dullaart RPF, Bakker SJL. Subclinical thyroid dysfunction and incident diabetes: a systematic review and an individual participant data analysis of prospective cohort studies. Eur J Endocrinol (2022) 187 S35-s46. DOI: 10.1530/EJE-22-0523 6. Wang X, Zhao X, Huang X. Association of Subclinical Thyroid Dysfunction with chronic kidney disease: a systematic review and meta-analysis. Endocr Res (2020) 45 41-9. DOI: 10.1080/07435800.2019.1645164 7. Vidili G, Delitala A, Manetti R. Subclinical hyperthyroidism: the cardiovascular point of view. Eur Rev Med Pharmacol Sci (2021) 25 3264-71. DOI: 10.26355/eurrev_202104_25735 8. Tsai K, Leung AM. Subclinical hyperthyroidism: a review of the clinical literature. Endocr Pract (2021) 27 254-60. DOI: 10.1016/j.eprac.2021.02.002 9. Hooi JKY, Lai WY, Ng WK, Suen MMY, Underwood FE, Tanyingoh D. Global prevalence of helicobacter pylori infection: systematic review and meta-analysis. Gastroenterology (2017) 153 420-9. DOI: 10.1053/j.gastro.2017.04.022 10. Zamani M, Ebrahimtabar F. Systematic review with meta-analysis: the worldwide prevalence of helicobacter pylori infection. Aliment Pharmacol Ther (2018) 47 868-76. DOI: 10.1111/apt.14561 11. Fiorini G, Zullo A, Castelli V, Lo Re G, Holton J, Vaira D. Role of helicobacter pylori infection in the thyroid diseases. J Gastrointestin Liver Dis (2013) 22 261-3. PMID: 24078981 12. Pellicano R, Ianiro G, Fagoonee S, Settanni CR, Gasbarrini A. Review: Extragastric diseases and helicobacter pylori. Helicobacter (2020) 25 e12741. DOI: 10.1111/hel.12636 13. Valtonen VV, Ruutu P, Varis K, Ranki M, Malkamäki M, Mäkelä PH. Serological evidence for the role of bacterial infections in the pathogenesis of thyroid diseases. Acta Med Scand (1986) 219 105-11. PMID: 3754083 14. Hou Y, Sun W, Zhang C, Wang T, Guo X, Wu L. Meta-analysis of the correlation between helicobacter pylori infection and autoimmune thyroid diseases. Oncotarget (2017) 8 115691-700. DOI: 10.18632/oncotarget.22929 15. de Luis DA, Varela C, de La Calle H, Cantón R, de Argila CM, San Roman AL. Helicobacter pylori infection is markedly increased in patients with autoimmune atrophic thyroiditis. J Clin Gastroenterol (1998) 26 259-63. DOI: 10.1097/00004836-199806000-00008 16. Shi WJ, Liu W, Zhou XY, Ye F, Zhang GX. Associations of helicobacter pylori infection and cytotoxin-associated gene a status with autoimmune thyroid diseases: a meta-analysis. Thyroid (2013) 23 1294-300. DOI: 10.1089/thy.2012.0630 17. Li H, Li J. Thyroid disorders in women. Minerva Med (2015) 106 109-14. PMID: 25668600 18. Ihnatowicz P, Drywień M, Wątor P, Wojsiat J. The importance of nutritional factors and dietary management of Hashimoto's thyroiditis. Ann Agric Environ Med (2020) 27 184-93. DOI: 10.26444/aaem/112331 19. Raei N, Behrouz B, Zahri S, Latifi-Navid S. Helicobacter pylori infection and dietary factors act synergistically to promote gastric cancer. Asian Pac J Cancer Prev (2016) 17 917-21. DOI: 10.7314/APJCP.2016.17.3.917 20. Fan L, Tan L, Chen Y, Du C, Zhu M, Wang K. Investigation on the factors that influence the prevalence of thyroid nodules in adults in Tianjin, China. J Trace Elem Med Biol (2018) 50 537-42. DOI: 10.1016/j.jtemb.2018.03.004 21. Hou C, Liu Z, Cui Y, Wang Y, Fu G, Liu H. An investigation of iodine intake and iodine nutritional status of adults in Tianjin City[in Chinese]. Chin J Endemiol (2016) 35 138-42 22. Zhang S, Fu J, Zhang Q, Liu L, Meng G, Yao Z. Association between nut consumption and non-alcoholic fatty liver disease in adults. Liver Int (2019) 39 1732-41. DOI: 10.1111/liv.14164 23. Gu Y, Zheng L, Zhang Q, Liu L, Meng G, Yao Z. Relationship between thyroid function and elevated blood pressure in euthyroid adults. J Clin Hypertens (2018) 20 1541-9. DOI: 10.1111/jch.13369 24. Chaker L, Bianco AC, Jonklaas J, Peeters RP. Hypothyroidism. Lancet (2017) 390 1550-62. DOI: 10.1016/S0140-6736(17)30703-1 25. Logan RP. Urea breath tests in the management of helicobacter pylori infection. Gut (1998) 43 S47-50. PMID: 9764040 26. Li H, Gu Y, Wu X, Rayamajhi S, Bian S, Zhang Q. Association between consumption of edible seaweeds and newly diagnosed non-alcohol fatty liver disease: the TCLSIH cohort study. Liver Int (2021) 41 311-20. DOI: 10.1111/liv.14655 27. Xia Y, Xiang Q, Gu Y, Jia S, Zhang Q, Liu L. A dietary pattern rich in animal organ, seafood and processed meat products is associated with newly diagnosed hyperuricaemia in Chinese adults: a propensity score-matched case-control study. Br J Nutr (2018) 119 1177-84. DOI: 10.1017/S0007114518000867 28. Alberti KG, Eckel RH, Grundy SM, Zimmet PZ, Cleeman JI, Donato KA. 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 1640-5. DOI: 10.1161/CIRCULATIONAHA.109.192644 29. Wu H, Liu M, Chi VTQ, Wang J, Zhang Q, Liu L. Handgrip strength is inversely associated with metabolic syndrome and its separate components in middle aged and older adults: a large-scale population-based study. Metab Clin Exp (2019) 93 61-7. DOI: 10.1016/j.metabol.2019.01.011 30. Austin PC. Optimal caliper widths for propensity-score matching when estimating differences in means and differences in proportions in observational studies. Pharm Stat (2011) 10 150-61. DOI: 10.1002/pst.433 31. Austin PC. Balance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity-score matched samples. Stat Med (2009) 28 3083-107. DOI: 10.1002/sim.3697 32. Xia Y, Meng G, Zhang Q, Liu L, Wu H, Shi H. Dietary patterns are associated with helicobacter pylori infection in Chinese adults: a cross-sectional study. Sci. Rep. (2016) 30 32334. DOI: 10.1038/srep32334 33. Monno R, De Laurentiis V, Trerotoli P, Roselli AM, Ierardi E, Portincasa P. Helicobacter pylori infection: association with dietary habits and socioeconomic conditions. Clin Res Hepatol Gastroenterol (2019) 43 603-7. DOI: 10.1016/j.clinre.2018.10.002 34. Zhang J, Gu Y, Meng G, Zhang Q, Liu L, Wu H. Association between dietary onion intake and subclinical hypothyroidism in adults: a population-based study from an iodine-replete area. Endocrine (2021) 74 616-24. DOI: 10.1007/s12020-021-02790-2 35. Figura N, Di Cairano G, Lorè F, Guarino E, Gragnoli A, Cataldo D. The infection by helicobacter pylori strains expressing CagA is highly prevalent in women with autoimmune thyroid disorders. J Physiol Pharmacol (1999) 50 817-26. PMID: 10695561 36. Wirth HP, Yang M, Karita M, Blaser MJ. Expression of the human cell surface glycoconjugates Lewis x and Lewis y by helicobacter pylori isolates is related to cagA status. Infect Immun (1996) 64 4598-605. DOI: 10.1128/iai.64.11.4598-4605.1996 37. Elisei R, Mariotti S, Swillens S, Vassart G, Ludgate M. Studies with recombinant autoepitopes of thyroid peroxidase: evidence suggesting an epitope shared between the thyroid and the gastric parietal cell. Autoimmunity (1990) 8 65-70. DOI: 10.3109/08916939008998434 38. Choi YM, Kim TY, Kim EY, Jang EK, Jeon MJ, Kim WG. Association between thyroid autoimmunity and helicobacter pylori infection. Korean J Intern Med (2017) 32 309-13. DOI: 10.3904/kjim.2014.369 39. Bertalot G, Montresor G, Tampieri M, Spasiano A, Pedroni M, Milanesi B. Decrease in thyroid autoantibodies after eradication of helicobacter pylori infection. Clin Endocrinol (2004) 61 650-2. DOI: 10.1111/j.1365-2265.2004.02137.x 40. Huang JY, Sweeney EG, Sigal M, Zhang HC, Remington SJ, Cantrell MA. Chemodetection and destruction of host urea allows helicobacter pylori to locate the epithelium. Cell Host Microbe (2015) 18 147-56. DOI: 10.1016/j.chom.2015.07.002 41. Goh KL, Chan WK, Shiota S, Yamaoka Y. Epidemiology of helicobacter pylori infection and public health implications. Helicobacter (2011) 16 1-9. DOI: 10.1111/j.1523-5378.2011.00874.x 42. Lewiński ASE, Karbownik M, Karasek M. Aging processes and the thyroid gland. Aging and age-related diseases: The basics (2006) 131-72 43. Shukla SK, Singh G, Ahmad S, Pant P. Infections, genetic and environmental factors in pathogenesis of autoimmune thyroid diseases. Microb Pathog (2018) 116 279-88. DOI: 10.1016/j.micpath.2018.01.004 44. Collodel G, Moretti E, Campagna MS, Capitani S, Lenzi C, Figura N. Infection by CagA-positive helicobacter pylori strains may contribute to alter the sperm quality of men with fertility disorders and increase the systemic levels of TNF-alpha. Dig Dis Sci (2010) 55 94-100. DOI: 10.1007/s10620-008-0704-1 45. Figura N, Di Cairano G, Moretti E. Helicobacter pylori infection and autoimmune thyroid diseases: the role of virulent strains. Antibiotics (2019) 9 12. DOI: 10.3390/antibiotics9010012 46. Astl J, Šterzl I. Activation of helicobacter pylori causes either autoimmune thyroid diseases or carcinogenesis in the digestive tract. Physiol Res (2015) 64 S291-301. DOI: 10.33549/physiolres.933118 47. Tomb JF, White O, Kerlavage AR, Clayton RA, Sutton GG, Fleischmann RD. The complete genome sequence of the gastric pathogen helicobacter pylori. Nature (1997) 388 539-47. DOI: 10.1038/41483 48. Prummel MF, Wiersinga WM. Thyroid peroxidase autoantibodies in euthyroid subjects. Best Pract Res Clin Endocrinol Metab (2005) 19 1-15. DOI: 10.1016/j.beem.2004.11.003 49. Crowe SE. Helicobacter pylori infection. N Engl J Med (2019) 380 1158-65. DOI: 10.1056/NEJMcp1710945
--- title: Serum neutrophil gelatinase-associated lipocalin as a potential biomarker for cognitive decline in spinal cord injury authors: - Qinghao Zhang - Ziteng Li - Liangyu Xie - Shengnan Cao - Zhonghao Cui - Bin Shi - Yuanzhen Chen journal: Frontiers in Neurology year: 2023 pmcid: PMC10025340 doi: 10.3389/fneur.2023.1120446 license: CC BY 4.0 --- # Serum neutrophil gelatinase-associated lipocalin as a potential biomarker for cognitive decline in spinal cord injury ## Abstract ### Objective Neutrophil gelatinase-associated lipoprotein (NGAL), a protein encoded by the lipocalcin-2 (LCN2) gene, has been reported to be involved in multiple processes of innate immunity, but its relationship with spinal cord injury (SCI) remains unclear. This study set out to determine whether NGAL played a role in the development of cognitive impairment following SCI. ### Methods At the Neck-Shoulder and Lumbocrural Pain Hospital, a total of 100 SCI patients and 72 controls were enrolled in the study through recruitment. Through questionnaires, baseline data on the participants' age, gender, education level, lifestyle choices (drinking and smoking) and underlying illnesses (hypertension, diabetes, coronary heart disease, and hyperlipidemia) were gathered. The individuals' cognitive performance was evaluated using the Montreal Cognitive Scale (MoCA), and their serum NGAL levels were discovered using ELISA. ### Results The investigation included 72 controls and 100 SCI patients. The baseline data did not differ substantially between the two groups, however the SCI group's serum NGAL level was higher than the control group's ($p \leq 0.05$), and this elevated level was adversely connected with the MoCA score ($p \leq 0.05$). According to the results of the ROC analysis, NGAL had a sensitivity of $58.24\%$ and a specificity of $86.72\%$ for predicting cognitive impairment following SCI. ### Conclusions The changes in serum NGAL level could serve as a biomarker for cognitive impairment in SCI patients, and this holds true even after taking in account several confounding variables. ## 1. Introduction Spinal cord injury (SCI) can occur at any level of the spinal cord and can result in temporary or permanent functional changes [1, 2]. Symptoms and prognosis vary depending on the location and severity of the injury [3, 4]. In most cases, the injury comes from physical trauma, and more than half of all injuries affect the cervical spine [5]. In the US, there are roughly 20,000 new instances of SCI each year, and the per-person lifetime economic cost can be as high as 3 million US dollars [6, 7]. The annual socioeconomic impact of SCI is projected to be 2.67 billion US dollars [8]. Therefore, early identification of potential biomarkers that can effectively predict disease severity and prognosis in SCI may be the key to treating cognitive decline after SCI. Neutrophil gelatinase-associated lipocalcin (NGAL), also known as lipocalcin-2 (LCN2) or oncogene 24p3 or, is an immune protein encoded by the LCN2 gene with a molecular weight of 25 kDa, and its immunomodulatory mechanism may be by limiting the utilization of iron by bacteria, it limits bacterial growth (9–11). In vivo, NGAL is mainly expressed in neutrophils, which is generally regarded as a biomarker of renal impairment [12, 13]. In 1989, NGAL was first isolated from SV-40-infected mouse kidney cells by Hraba-Renevey et al. [ 14]. Human NGAL contains a 20 amino acid signal peptide and a “lipidin” domain at the N-terminus of the protein, which exerts biological effects by binding to corresponding ligands [15]. Human NGAL is up to $98\%$ homologous to chimpanzees, but $62\%$ and $63\%$ homologous to mice and rats, respectively [15, 16]. The role of NGAL in neuroplasticity and its effects on cognitive function have been documented in recent years, but its precise mechanism is still poorly understood [17]. Our purpose of this study is to further verify whether NGAL is involved in cognitive decline after SCI, in order to provide new biomarker targets for the prevention and treatment of cognitive impairment after SCI. ## 2.1. Study population Patients with SCI and healthy controls who were hospitalized for Neck-Shoulder and Lumbocrural Pain Hospital between September 2020 and August 2022 made up the study population. The most recent recommendations provide the basis for SCI diagnosis. Congenital spinal abnormalities, severe systemic disorders, a history of spinal cord surgery, cognitive impairment, transfer to another hospital, and unwillingness to cooperate are the exclusion criteria for SCI. Additionally, a control group was drawn from the general population. All participants signed informed consent, and our study was approved by the hospital ethics committee (No. 2022012). The detailed flowchart is shown in Figure 1. Figure 1: The flow chart of study implementation. ## 2.2. Baseline data We collected baseline data including age, gender, education level, living habits (smoking and drinking), and underlying diseases (hypertension, diabetes, coronary heart disease, and hyperlipidemia). These data are obtained through questionnaires and recorded, counted and analyzed by specialized personnel. ## 2.3. Cognitive function In this study, the MoCA scale, a popular measure for assessing cognitive function, was utilized to identify the cognitive function of SCI patients. The MoCA scale, developed by Montrealer Ziad Nasreddine, has been used by researchers and physicians worldwide since it was first released in 1996. It has been translated into 46 languages. For doctors, free. People were given 10 min to respond to 30 questions; the total score was 30, with one point deducted for each incorrect response, and a score of < 26 being judged cognitively deficient [18]. Our study was approved by MoCA Test Inc. The evaluators were specially trained and blinded to the baseline data of the test subjects. ## 2.4. Serum NGAL level Venous blood was collected immediately after fasting for 8 h after enrollment in all participants. Venous blood was allowed to stand at room temperature for 10 min, then centrifuged at 1200 g for 15 min, and the upper serum was separated and aliquoted and stored at −80°C for future use [19]. ThermoFisher, Wilmington, DE, USA, provided the ELISA kit that was utilized to measure the presence of NGAL in the serum of SCI patients. ## 2.5. Statistical analysis The statistical analysis was performed using SPSS 26.0. The measurement data was represented by the mean ± standard deviation (SD), while the enumeration data was represented by number (N). The link between serum NGAL and MoCA was discovered using P for trend. ROC analysis further evaluated the sensitivity and specificity of serum NGAL in predicting cognitive function in SCI. All statistical cutoffs were set at 0.05 and $p \leq 0.05$ was considered statistically significant. ## 3.1. Clinical baseline data of the study population We recruited a total of 100 SCI patients and 72 healthy controls for this study. The baseline demographic and clinical information for the complete study population, stratified by clinical traits, is shown in Table 1. Age, sex, education level, smoking, drinking, hypertension, coronary heart disease, diabetes, and hyperlipidemia were not statistically significantly different between the two groups, as shown in the table ($p \leq 0.05$). Table 1 \| Unnamed: 0 \| Controls (n = 72) \| SCI (n = 100) \| p-value \| \| --- \| --- \| --- \| --- \| \| Age, years \| 59.3 ± 7.1 \| 60.6 ± 7.8 \| 0.265 \| \| Gender, male/female \| 56/16 \| 88/12 \| 0.073 \| \| Education level, n (%) \| \| \| 0.798 \| \| Low \| 33 \| 51 \| \| \| Middle \| 26 \| 33 \| \| \| High \| 13 \| 16 \| \| \| Smoking, n (%) \| \| \| 0.520 \| \| Never \| 26 \| 38 \| \| \| Former \| 10 \| 17 \| \| \| Current \| 36 \| 45 \| \| \| Drinking, n (%) \| 28 \| 42 \| 0.682 \| \| Hypertension, n (%) \| 20 \| 32 \| 0.552 \| \| Diabetes, n (%) \| 11 \| 14 \| 0.815 \| \| Coronary heart disease, n (%) \| 9 \| 13 \| 0.923 \| \| Hyperlipidemia, n (%) \| 17 \| 28 \| 0.518 \| \| NGAL, pg/ml \| 125.4 ± 12.3 \| 196.7 ± 23.8 \| < 0.001 \| \| MoCA, points \| 27.5 ± 1.2 \| 24.6 ± 1.7 \| < 0.001 \| ## 3.2. Serum NGAL level and MoCA score The serum NGAL level for the control group was (125.4 ± 12.3 pg/ml, as reported in Table 1, while it was (196.72 ± 3.8 pg/ml) for the SCI group. The SCI group's serum NGAL levels were noticeably greater than those of the control group ($p \leq 0.001$). The SCI group's MoCA score was (24.6 ± 1.7) points, compared to the control group's (27.5 ± 1.2) points. When compared to the control group, the MoCA score of the SCI group was considerably lower ($p \leq 0.001$). Figure 2 compares the MoCA ratings and serum NGAL concentrations between the two groups. The results showed that the serum NGAL level in SCI group was significantly higher than that in control group, while the MoCA score was significantly lower than that in control group. Figure 2: The comparison of serum NGAL levels and MoCA scores between SCI and controls. SCI, spinal cord injury. p < 0.05.* ## 3.3. Correlation analysis between serum NGAL level and MoCA assessment We separated the SCI patients into 4 groups based on the quartile levels of the serum NGAL and looked at the connection between those groups and the MoCA score. Table 2 displays the correlation analysis between the serum NGAL level and MoCA score. From Q1 to Q4, MoCA scores were (25.8 ± 1.9), (24.9 ± 1.8), (24.2 ± 1.5) and (23.5 ± 1.6), respectively. The findings indicated that the MoCA score decreased as blood NGAL level increased ($p \leq 0.001$), indicating that a high serum NGAL level may be a sign of cognitive impairment. Table 2 \| Variable \| Q1 \| Q2 \| Q3 \| Q4 \| P-values \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| MoCA scores \| 25.8 ± 1.9 \| 24.9 ± 1.8 \| 24.2 ± 1.5 \| 23.5 ± 1.6 \| < 0.001 \| ## 3.4. Multiple model regression analysis To explore the etiology affecting the MoCA score, we performed a multi-model regression analysis (Table 3). In model 1, after adjusting the confounding of age, sex and education level, it was suggested that serum NGAL was a risk factor for SCI-related cognitive impairment ($p \leq 0.05$); in model 2, we further adjusted the Smoking and drinking suggest that serum NGAL is also a risk factor for SCI-related cognitive impairment ($p \leq 0.05$); Serum NGAL was found to be an independent risk factor for cognitive impairment caused by SCI in model 3, which was based on model 2. After further adjusting for the underlying conditions (hypertension, diabetes, coronary heart disease, and hyperlipidemia), the same result was found ($$p \leq 0.047$$). Table 3 \| Unnamed: 0 \| MoCA scores \| MoCA scores.1 \| \| --- \| --- \| --- \| \| \| Regression coefficient \| P -values \| \| Model 1 \| 0.352 \| < 0.001 \| \| Model 2 \| 0.271 \| < 0.001 \| \| Model 3 \| 0.218 \| 0.047 \| ## 3.5. ROC curve analysis In order to further verify the accuracy of serum NGAL level in diagnosing cognitive impairment after SCI, we performed ROC curve analysis, and the results are shown in Figure 3. The sensitivity of serum NGAL in diagnosing cognitive impairment after SCI was $72.48\%$, and the specificity was $61.28\%$. Figure 3: ROC curve analysis for serum NGAL levels in SCI. ## 4. Discussion This is the first study of the relationship between cognitive impairment and serum NGAL levels in patients after SCI. Our results showed that the blood level of NGAL in SCI patients was significantly higher than that in healthy controls, and the quartile level was negatively correlated with MoCA score. In multi-model regression analysis, serum NGAL levels were considered to be independent risk factors for SCI-related cognitive impairment after adjusting for multiple confounding. Subsequent ROC analysis further proved that the serum NGAL level has a high accuracy in diagnosing SCI-related cognitive impairment. Our study suggests that serum NGAL levels may serve as a potential biomarker of SCI-related cognitive function. NGAL is mainly expressed and secreted by immune cells, liver cells or renal tubular cells, and it can capture and consume siderophore to play an antibacterial role [20, 21]. In addition to antibacterial, NGAL can also be used as a factor regulating cell growth and differentiation, mediating the biological activity of iron inside and outside cells [22]. Genomic studies have shown that NGAL is one of the most up-regulated genes in acute kidney injury, and it can regulate the secretion of a renal tubulin with a molecular weight of 25KDa, which quickly enters the body fluid after the onset of renal injury [23]. NGAL rises 24–48 h earlier than conventional serum creatinine, making it potentially a more effective biomarker. Studies have also shown that elevated levels of NGAL can predict the prognosis of acute kidney injury [24]. All of the above make NGAL the focus of clinical translational research. In addition to its involvement in acute kidney injury, a role for NGAL in neurological disorders has also been found. Zhao et al. [ 25] found that the expression of NGAL increased after traumatic brain injury, which was negatively correlated with the clinical score reflecting the severity of traumatic brain injury, and it has good sensitivity and specificity as a biomarker for diagnosing traumatic brain injury [25]. Peng et al. [ 26] found that the level of NGAL increases after cerebral ischemia, and the activation of EGF/EGFR can regulate the expression of NGAL by activating the JAK2/STAT3 pathway to improve neurological deficits [26]. Serra et al. [ 27] discovered that plasma NGAL levels were significantly higher in aneurysm patients than in the control group, indicating that NGAL may be involved in the pathophysiological process of aneurysms and that NGAL may be used as an indicator for assessing aneurysm rupture and prognosis in the future [27]. In recent years, studies on the involvement of NGAL in cognitive impairment have been found. The Dutch research team found that low levels of NGAL in serum and cerebrospinal fluid can be used as potential biomarkers to predict the conversion of mild cognitive impairment to Alzheimer's disease (AD), and affect the pathophysiological process of AD accompanied by depression [28, 29]. The same research team also found that NGAL was associated with cognitive impairment in patients with depression, and there were gender differences [30]. In addition, NGAL is also considered to be associated with the pathogenesis of Down syndrome. The research of NGAL in SCI has also come into the field of vision of researchers. Behrens, V found that NGAL was significantly increased in the spinal cord, brain, liver and serum in the SCI mouse model, while the absence of NGAL could significantly reduce the differentiation of glial cells, indicating that it may be involved in the inflammatory injury after SCI [31]. Rathore et al. [ 32] found that Lcn2 can regulate the inflammatory response after SCI, while the lack of NGAL can reduce the secondary injury after SCI and improve the recovery of motor function [32]. However, clinical studies of NGAL in SCI patients have not been reported. The first study to reveal cognitive damage in patients following NGAL involvement in SCI is ours. However, our study has certain flaws. Our study is a single-center, small-sample investigation with Chinese participants. It is debatable if the findings of this study apply to other geographic or racial groups. There is an urgent need for large-sample multi-center research to confirm this study. ## 5. Conclusions The results of the current study suggest that changes in serum NGAL could serve as a biomarker for cognitive impairment in SCI patients, and this finding holds true even after taking into account a number of confounding variables. To develop innovative methods for treating cognitive impairment caused by SCI, future scientific and clinical research must further investigate the underlying mechanism. ## 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 Ethics Committee of Neck-Shoulder and Lumbocrural Pain Hospital. The patients/participants provided their written informed consent to participate in this study. ## Author contributions QZ and YC designed this research and wrote the manuscript. ZL, LX, SC, ZC, and BS participated in data collection, experimental process, and data analysis. 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. Li Z, Yu S, Hu X, Li Y, You X, Tian D. Fibrotic scar after spinal cord injury: crosstalk with other cells, cellular origin, function, and mechanism. Front Cell Neurosci. (2021) 15 720938. DOI: 10.3389/fncel.2021.720938 2. Tran AP, Warren PM, Silver J. New insights into glial scar formation after spinal cord injury. Cell Tissue Res. (2021) 387 1-18. DOI: 10.1007/s00441-021-03477-w 3. Fouad K, Popovich PG, Kopp MA, Schwab JM. The neuroanatomical–functional paradox in spinal cord injury. Nat Rev Neurol. (2021) 17 53-62. DOI: 10.1038/s41582-020-00436-x 4. Krogh S, Aagaard P, Jønsson AB, Figlewski K, Kasch H. Effects of repetitive transcranial magnetic stimulation on recovery in lower limb muscle strength and gait function following spinal cord injury: a randomized controlled trial. Spinal Cord. (2022) 60 135-41. DOI: 10.1038/s41393-021-00703-8 5. Ahuja CS, Wilson JR, Nori S, Kotter M, Druschel C, Curt A. Traumatic spinal cord injury. Nat Rev Dis Primers. (2017) 3 1-21. DOI: 10.1038/nrdp.2017.18 6. Liu FJ, Xu HH, Yin Y, Chen YZ, Xie LY, Li HZ. Decreased adiponectin levels are a risk factor for cognitive decline in spinal cord injury. Disease Markers. (2022). DOI: 10.1155/2022/5389162 7. Castro-Marin F, Gaither JB, Rice AD, Blust R, Chikani V, Vossbrink A. Prehospital protocols reducing long spinal board use are not associated with a change in incidence of spinal cord injury. Prehospital Emerg Care. (2020) 24 401-10. DOI: 10.1080/10903127.2019.1645923 8. Cao Y, Krause JS. The association between secondary health conditions and indirect costs after spinal cord injury. Spinal Cord. (2021) 59 306-10. DOI: 10.1038/s41393-020-00567-4 9. Ye Z, Wang S, Yang Z, He M, Zhang S, Zhang W. Serum lipocalin-2, cathepsin S and chemerin levels and nonalcoholic fatty liver disease. Mol Biol Rep. (2014) 41 1317-23. DOI: 10.1007/s11033-013-2977-5 10. Millar SA, Anderson SI, O'Sullivan SE. Osteokines and the vasculature: a review of the in vitro effects of osteocalcin, fibroblast growth factor-23 and lipocalin-2. PeerJ. (2019) 7 e7139. DOI: 10.7717/peerj.7139 11. Zhang F, Liu E, Radaic A, Yu X, Yang S, Yu C. Diagnostic potential and future directions of matrix metalloproteinases as biomarkers in gingival crevicular fluid of oral and systemic diseases. Int J Biol Macromol. (2021) 188 180-96. DOI: 10.1016/j.ijbiomac.2021.07.165 12. Khawaja S, Jafri L, Siddiqui I, Hashmi M, Ghani F. The utility of neutrophil gelatinase-associated Lipocalin (NGAL) as a marker of acute kidney injury (AKI) in critically ill patients. Biomarker Res. (2019) 7 1-6. DOI: 10.1186/s40364-019-0155-1 13. Mishra J, Dent C, Tarabishi R, Mitsnefes MM, Ma Q, Kelly C. Neutrophil gelatinase-associated lipocalin (NGAL) as a biomarker for acute renal injury after cardiac surgery. Lancet. (2005) 365 1231-8. DOI: 10.1016/S0140-6736(05)74811-X 14. Garay-Rojas E, Harper M, Hraba-Renevey S, Kress M. An apparent autocrine mechanism amplifies the dexamethasone- and retinoic acid-induced expression of mouse lipocalin-encoding gene 24p3. Gene. (1996) 170 173-80. DOI: 10.1016/0378-1119(95)00896-9 15. Al Jaberi S, Cohen A, D'Souza C, Abdulrazzaq YM, Ojha S, Bastaki S. Lipocalin-2: structure, function, distribution and role in metabolic disorders. Biomed Pharmacother. (2021) 142 112002. DOI: 10.1016/j.biopha.2021.112002 16. Kjeldsen L, Cowland JB, Borregaard N. Human neutrophil gelatinase-associated lipocalin and homologous proteins in rat and mouse. BBA Protein Struct Mol Enzymol. (2000) 1482 272-83. DOI: 10.1016/S0167-4838(00)00152-7 17. Choi J, Lee H-W, Suk K. Increased plasma levels of lipocalin 2 in mild cognitive impairment. J Neurol Sci. (2011) 305 28-33. DOI: 10.1016/j.jns.2011.03.023 18. Wang Q, Wang K, Ma Y, Li S, Xu Y. Serum galectin-3 as a potential predictive biomarker is associated with poststroke cognitive impairment. Oxid Med Cell Longev. (2021) 2021 5827812. DOI: 10.1155/2021/5827812 19. Zhang F, Hou G, Hou G, Wang C, Shi B, Zheng Y. Serum irisin as a potential biomarker for cognitive decline in vascular dementia. Front Neurol. (2021) 12 755046. DOI: 10.3389/fneur.2021.755046 20. Schmidt-Ott KM, Mori K, Li JY, Kalandadze A, Cohen DJ, Devarajan P. Dual action of neutrophil gelatinase–associated lipocalin. J Am Soc Nephrology. (2007) 18 407-13. DOI: 10.1681/ASN.2006080882 21. Yang J, Gan X, Song X, Yuan R, Xiang Y. Apamer-based sensitive and label-free electrochemical detection of neutrophil gelatinase-associated lipocalin via recycling amplification cascades. Anal Chim Acta. (2022) 1233 340515. DOI: 10.1016/j.aca.2022.340515 22. Kim IY, Kim JH, Lee DW, Lee SB, Rhee H, Song SH. Plasma neutrophil gelatinase-associated lipocalin is associated with iron status in anemic patients with pre-dialysis chronic kidney disease. Clin Exp Nephrol. (2018) 22 28-34. DOI: 10.1007/s10157-017-1409-6 23. Virzì GM, Clementi A, De Cal M, Cruz DN, Ronco C. Genomics and biological activity of neutrophil gelatinase-associated lipocalin in several clinical settings. Blood Purif. (2013) 35 139-43. DOI: 10.1159/000346100 24. Haase M, Bellomo R, Devarajan P, Schlattmann P, Haase-Fielitz A. Accuracy of neutrophil gelatinase-associated lipocalin (NGAL) in diagnosis and prognosis in acute kidney injury: a systematic review and meta-analysis. Am J Kidney Dis. (2009) 54 1012-24. DOI: 10.1053/j.ajkd.2009.07.020 25. Zhao J, Chen H, Zhang M, Zhang Y, Qian C, Liu Y. Early expression of serum neutrophil gelatinase-associated lipocalin (NGAL) is associated with neurological severity immediately after traumatic brain injury. J Neurol Sci. (2016) 368 392-8. DOI: 10.1016/j.jns.2016.07.060 26. Peng D-H, Liu Y-Y, Chen W, Hu H-N, Luo Y. Epidermal growth factor alleviates cerebral ischemia-induced brain injury by regulating expression of neutrophil gelatinase-associated lipocalin. Biochem Biophys Res Commun. (2020) 524 963-9. DOI: 10.1016/j.bbrc.2020.02.025 27. Serra R, Volpentesta G, Gallelli L, Grande R, Buffone G, Lavano A. Metalloproteinase-9 and neutrophil gelatinase-associated lipocalin plasma and tissue levels evaluation in middle cerebral artery aneurysms. Br J Neurosurg. (2015) 29 1-5. DOI: 10.3109/02688697.2014.913777 28. Naudé PJ, Ramakers IH, van der Flier WM, Jiskoot LC, Reesink FE, Claassen JA. Serum and cerebrospinal fluid Neutrophil gelatinase-associated lipocalin (NGAL) levels as biomarkers for the conversion from mild cognitive impairment to Alzheimer's disease dementia. Neurobiol Aging. (2021) 107 1-10. DOI: 10.1016/j.neurobiolaging.2021.07.001 29. Dekens DW, Naudé PJ, Engelborghs S, Vermeiren Y, Van Dam D, Oude Voshaar RC. Neutrophil gelatinase-associated lipocalin and its receptors in Alzheimer's disease (AD) brain regions: differential findings in AD with and without depression. J Alzheimer's Dis. (2017) 55 763-76. DOI: 10.3233/JAD-160330 30. Naudé P, den Boer J, Comijs H, Bosker F, Zuidersma M, Groenewold N. Sex-specific associations between neutrophil gelatinase-associated Lipocalin (NGAL) and cognitive domains in late-life depression. Psychoneuroendocrinology. (2014) 48 169-77. DOI: 10.1016/j.psyneuen.2014.06.016 31. Behrens V, Voelz C, Müller N, Zhao W, Gasterich N, Clarner T. Lipocalin 2 as a putative modulator of local inflammatory processes in the spinal cord and component of organ cross talk after spinal cord injury. Mol Neurobiol. (2021) 58 5907-19. DOI: 10.1007/s12035-021-02530-7 32. Rathore KI, Berard JL, Redensek A, Chierzi S, Lopez-Vales R, Santos M. Lipocalin 2 plays an immunomodulatory role and has detrimental effects after spinal cord injury. J Neurosci. (2011) 31 13412-9. DOI: 10.1523/JNEUROSCI.0116-11.2011
--- title: 'Pharmacogenetic association of the NR1H3 promoter variant with antihypertensive response among patients with hypertension: A longitudinal study' authors: - Yu Chen - Yuqing Han - Yiyi Wu - Rutai Hui - Yunyun Yang - Yixuan Zhong - Shuyuan Zhang - Weili Zhang journal: Frontiers in Pharmacology year: 2023 pmcid: PMC10025344 doi: 10.3389/fphar.2023.1083134 license: CC BY 4.0 --- # Pharmacogenetic association of the NR1H3 promoter variant with antihypertensive response among patients with hypertension: A longitudinal study ## Abstract Background: The genetic factors in assessing therapeutic efficacy and predicting antihypertensive drug response are unclear. Therefore, this study aims to identify the associations between variants and antihypertensive drug response. Methods: A longitudinal study including 1837 hypertensive patients was conducted in Northern China and followed up for a median 2.24 years. The associations of 11 candidate variants with blood pressure changes in response to antihypertensive drugs and with the risk of cardiovascular events during the follow-up were examined. The dual-luciferase assay was carried out to assess the effect of genetic variants on gene transcriptional activity. Results: The variant rs11039149A>G in the promoter of nuclear receptor subfamily 1 group H member 3 (NR1H3) was associated with the change in systolic blood pressure (ΔSBP) in response to calcium channel blockers (CCBs) monotherapy. Patients carrying rs11039149AG genotype showed a significant increase of systolic blood pressure (SBP) at follow-up compared with AA carriers, and the difference of ΔSBP between AG and AA carriers was 5.94 mm Hg ($95\%$CI: 2.09–9.78, $$p \leq 0.002$$). In 1,184 patients with CCBs therapy, SBP levels decreased in AA carriers, but increased in AG carriers, the difference of ΔSBP between AG and AA carriers was 8.04 mm Hg ($95\%$CI: 3.28–12.81, $$p \leq 0.001$$). Further analysis in 359 patients with CCBs monotherapy, the difference of ΔSBP between AG and AA carriers was 15.25 mm Hg ($95\%$CI: 6.48–24.02, $$p \leq 0.001$$). However, there was no significant difference in ΔSBP between AG and AA carriers with CCBs multitherapy. The rs11039149A>G was not associated with the cardiovascular events incidence during the follow-up. Additionally, transcriptional factor forkhead box C1 (FOXC1) bound to the NR1H3 promoter containing rs11039149A and significantly increased the transcriptional activity, while rs11039149 A to G change led to a loss-of-function and disabled FOXC1 binding. For the other 10 variants, associations with blood pressure changes or risk of cardiovascular events were not observed. Conclusion: Hypertensive patients with rs11039149AG genotype in the NR1H3 gene have a significant worse SBP control in response to CCBs monotherapy compared with AA carriers. Our findings suggest that the NR1H3 gene might act as a promising genetic factor to affect individual sensitivity to antihypertensive drugs. ## 1 Introduction Hypertension is a major risk factor leading to cardiovascular diseases (Carey et al., 2022). About one-third of Chinese adults suffer from hypertension, while only $29.6\%$ of patients with antihypertensive treatment achieved the blood pressure goal (Lewington et al., 2016). Interindividual genetic variability might explain this disappointing outcome partly. However, the genetic factors in assessing therapeutic efficacy and predicting antihypertensive drug response are unclear. The blood pressure levels are influenced by environmental and genetic factors (Niu et al., 2021), and genetic factors have shown inheritability from $30\%$ to $50\%$ on blood pressure variation among Chinese and White individuals (Wu et al., 2011; Ehret & Caulfield, 2013; Munroe et al., 2013). Pharmacogenomic studies of hypertension have suggested that genetic variants related to blood pressure elevation showed effects on antihypertensive response in Chinese, Americans, Europeans and Hispanics, etc. ( Johnson, 2012; Gong et al., 2015; Iniesta et al., 2019; Citterio et al., 2021; Xiao et al., 2022). Previous genome-wide association studies and candidate gene strategy studies have identified multiple genes and variants related to antihypertensive response. For example, some genes are involved in ion channel function, such as calcium voltage-gated channel subunit alpha1 C (CACNA1C) that is found to be associated with antihypertensive response to calcium channel blockers (CCBs) (Beitelshees et al., 2009), and NEDD4 like E3 ubiquitin protein ligase (NEDD4L) that is associated with treatment response with beta-blockers or diuretics (Luo et al., 2009; Svensson-Färbom et al., 2011; McDonough et al., 2013). Some genes encode the components of the renin-angiotensin-aldosterone system (RAAS) such as angiotensinogen (AGT), angiotensin converting enzyme 2 (ACE2) and angiotensin II receptor type 1 (AGTR1), which are found to have effects on antihypertensive response to angiotensin converting enzyme inhibitors (ACEIs) or angiotensin receptor blockers (ARBs) (Liljedahl et al., 2004; Fan et al., 2007; Beitelshees et al., 2009; Luo et al., 2009; McDonough et al., 2013; de Denus et al., 2018). However, some genes and variants associated with antihypertensive response still have unclear biological mechanisms. Genetic variants can be used to explore the efficacy of antihypertensive drugs, but studies have shown that a risk allele for worse response to one drug class might exhibit a beneficial response to another drug class. For example, rs5051 in AGT is found to be associated with blood pressure lowering to beta-blockers in 115 Swedish patients with hypertension and left ventricular hypertrophy (Kurland et al., 2004) but with blood pressure lowering to ACEIs in 640 Chinese patients with essential hypertension (Kurland et al., 2004; Yu et al., 2014). Rs5186 in AGTR1 is related to different therapeutic efficacy to CCBs and ACEIs in 311 White Europeans with hypertension (Benetos et al., 1996) whereas associated with ARBs on systolic blood pressure (SBP) response in 1,049 Chinese patients with hypertension (Jiang et al., 2011). These inconsistent associations may have multiple reasons unrelated to the pharmacology, or due to methodological differences, or led by ethnics of participants. Other variants in CACNA1C (Beitelshees et al., 2009), ACE2 (Fan et al., 2007), NEDD4L (Luo et al., 2009; Svensson-Färbom et al., 2011; McDonough et al., 2013), adducin 1 (ADD1) (Turner et al., 2003; Vormfelde & Brockmöller, 2012), matrix metallopeptidase 3 (MMP3) (Sherva et al., 2011), nuclear receptor subfamily 1 group H member 3 (NR1H3) (Price et al., 2011), and protein tyrosine phosphatase receptor type D (PTPRD) (Gong et al., 2015) are also shown to have relationship with antihypertensive response or cardiovascular events in response to antihypertensive medications among the Chinese population or other population (including White Europeans, White Americans, Black Americans, Hispanics, etc.), but these results varied between populations and thus should be further explored. The investigations for genetic variants on antihypertensive response in Chinese population are limited. In this study, a total of 11 variants at 9 genes (AGT, AGTR1, ADD1, ACE2, CACNA1C, NEDD4L, NR1H3, PTPRD, MMP3) are chosen and genotyped in a Chinese prospective cohort study including 1837 patients with hypertension, aiming to explore the association of variants with the changes in blood pressure in response to antihypertensive drugs treatment and the risk of cardiovascular events during the follow-up. ## 2.1 Study population This prospective study was conducted in two communities, the Benxi County, Liaoning Province, and Hongxinglong County, Heilongjiang province in the northern region in China. A total of 1953 patients with hypertension (>40 years) were enrolled at two time periods according to the same criteria of inclusion and exclusion, of whom 463 patients from January to November in 2009 and 1,490 patients from January 2012 to December 2014. The definition of hypertension was based on the following criteria: SBP≥140 mm Hg and/or diastolic blood pressure (DBP)≥90 mm Hg, and/or receiving antihypertensive drugs, and/or history of hypertension. The enrolled patients were provided with antihypertensive drugs at free of charge, including CCBs, ARBs, ACEIs, thiazide-type diuretics (hydrochlorothiazide), or beta-blockers, unless intolerance was reported. The drug dosages and types were modulated by doctors according to the patients’ blood pressure levels. All patients were followed up face-to-face every 2 years by trained investigators, and the last visit was from May 1 to November 30 in 2016. The main endpoint was the composite of stroke, myocardial infarction, coronary revascularization, hospitalization for unstable angina or acute decompensated heart failure, and deaths from cardiovascular causes. Definitions of the endpoints were described in the Supplementary Material. Before data assessment, we excluded 98 patients lack of antihypertensive treatment data, 11 patients without genotyping data due to lack of blood samples, and 7 patients who were loss of follow-up due to immigration and lack of follow-up blood pressure. Thus, a total of 1837 patients with complete clinical data and genetic information were included in this study for analyzing the association of variants with blood pressure lowering response and risk of cardiovascular events. The flowchart of study was shown in Figure 1. FIGURE 1: Flowchart of the current study. Cardiovascular events including stroke, myocardial infarction, coronary revascularization, hospitalization for unstable angina or acute decompensated heart failure, and deaths from cardiovascular causes. ## 2.2 Data collection and blood pressure measurements Baseline demographic characteristics including age, sex, antihypertensive therapy, medical history, and lifestyles (smoking and alcohol status) of enrolled patients were collected through interviews, and a standardized questionnaire was utilized for data collection. Height and weight were measured by skilled nurses, and body mass index was computed as weight (kilograms) divided by the square of height (meters). Blood samples were drawn from the antecubital vein after overnight fast and total cholesterol, low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, triglycerides, fasting blood glucose and serum creatinine were examined by an automatic analyzer (Hitachi 7,060, Tokyo, Japan). All measurements were done at Beijing FuWai Clinical Laboratory and certified by the Centers for Disease Control and Prevention. Blood pressure was measured by trained nurses using validated oscillometric monitors with suitable-sized arm cuffs (large, medium, or small). All patients were requested to avoid smoke, alcohol, tea, coffee, and exercise for at least 30 min before the blood pressure was measured. The sitting blood pressure was measured 3 times at intervals of 1 min after at least 5 min of rest, the average of the second and third readings was recorded as the blood pressure level. The changes in blood pressure were derived from follow-up blood pressure minus baseline blood pressure. ΔSBP and ΔDBP were used for indicating the changes in SBP and DBP, respectively. ## 2.3 Gene variants selection and genotyping Eleven genetic variants linked to antihypertensive response were genotyped in this study. These variants were located at 9 genes, including 3 genes involved in the RAAS (AGT, AGTR1, ACE2), 2 genes involved in ion channels (CACNA1C, NEDD4L), 3 genes associated with vascular functions (NR1H3, PTPRD, MMP3), and ADD1 gene, which encodes adducin-1 and plays important roles in the stabilization of the membrane cortical cytoskeleton, cell-cell adhesions, and cell signal transduction (Manunta & Bianchi, 2006). Genetic variants were listed in Supplementary Table S1. Genomic DNA was extracted from the peripheral white blood cell with FlexGen Blood DNA Kit (Cowin Biotech Co., Beijing, China). Genotyping was performed using the Nascent Genotyping system with a custom-by-design 48-Plex SNPscan™ Kit (Genesky Biotechnologies Co., Shanghai, China) which was based on double ligation and multiplex fluorescence polymerase chain reaction (PCR) method. For each variant, two special primers (one for wild allele and the other for mutant allele) and one common primer were designed (Supplementary Table S2). The procedure was as the following: a ligation mixture was first prepared in 20 μl, containing 100 ng DNA sample, 1x ligase buffer, 1U ligase and 1 x primer mix. The ligation reaction was carried out in an ABI2720 thermal cycler under the following cycling program: 95°C for 5 min, 4 cycles of 94°C for 1 min, 58°C for 4 min, 94°C for 2 min, hold at 72°C. Two 48-multiplex fluorescence PCR reactions were performed for each ligation product. The PCR mixture was prepared in 20 μl, containing 1 μl ligation product, 1x primer mix and 1x PCR master mix. The PCR program was described as following: 95°C for 2 min; 9 cycles of 94°C for 20 s, 62°C–0.5°C/cycle for 40 s, and 72°C for 1.5 min; 25 cycles of 94°C for 20 s, 57°C for 40 s, and 72°C for 1.5 min; 68°C for 1 h; and hold at 4°C. PCR products were then detected by capillary electrophoresis on an ABI 3730XL sequencer. Genotyping data were analyzed by the software GeneMapper version 4.1. The genotyping method used in this study has been validated by single nucleotide extension with the Multiplex SNaPshot Kit (Applied Biosystems Inc., Foster City, CA, USA) in previous studies which have reported >$99\%$ concordance rates of validation (Chen et al., 2012; Lu et al., 2021). In our study, Genotyping for eleven variants had completion rates of $99.8\%$. For quality control, we chose $5\%$ of samples randomly and repeated genotyping, and the concordance rate of repeated samples was $100\%$. The phenotype status of the patients was masked throughout the genotyping. ## 2.4 Construction of plasmids Bioinformatic analysis (JASPAR database, http://jaspar.genereg.net/) was used to explore the potential binding sequence of transcriptional factors in relation to the genetic variants in this study. To verify whether rs11039149A>G influences the transcriptional activity of NR1H3, 200-bp NR1H3 promoter fragments containing rs11039149A or rs11039149G allele were amplified by PCR method using the following primers: forward 5′-GGCGGTACCCCTCAGCAGCTTGCCTCC-3′, reverse 5′- GGCACGCGTAGTGGCTGGG CTGGGATCAGA-3′, then cloned into the pGL3-promoter luciferase vector (Promega, WI, USA) at KpnI and MluI restriction sites, and two plasmids were generated: pGL3-rs11039149A and pGL3-rs11039149G. Forkhead box C1 (FOXC1) gene coding sequence was amplified using the following primers: forward 5′-GCGCGATCGCATGCAGGCGCGCTACTCC-3′, reverse 5′- GGCACGCGTCA AAACTTGCTACAGTCGTAG-3′, then cloned into eukaryotic expression vector pENTER (WZ Biosciences, Shandong, China) at AsisI and MluI restriction sites to generate the pENTER-FOXC1 plasmid. All plasmids were confirmed by DNA sequencing. ## 2.5 Cell culture, transfection, and dual-luciferase assays HEK293T cells were seeded at a density of 2.0 × 104 cells per well in a 96-well plate, and transfection was performed with lipofectamin3000 (Invitrogen, USA) according to the manufacturer’s protocol. Luciferase vector pGL3-rs11039149A or pGL3-rs11039149G were co-transfected with a transcription factor expression plasmid pENTER-FOXC1 or pENTER empty vector (each vector 20 ng per well). The Renilla luciferase plasmid pRL-TK (Promega, WI, USA) were co-transfected as an internal control (1 ng per well) to standardize the transfection efficiency. The transfected cells were harvested after 36 h, and the luciferase activity was measured using the dual-luciferase reporter assay system (Promega, WI, USA) by a luminometer (TD-20; Turner Designs). ## 2.6 Statistical analysis Hardy-Weinberg equilibrium for each variant was examined by chi-square test. We used fixed-effect model as the main analysis of genetic association, given that fixed-effect model is more appropriate for exploratory or predictive studies with large sample sizes greater than 500, and it has a smaller median absolute error ($4\%$) of the true marginal effect than random-effect model, as proposed by a simulation study (Dieleman & Templin, 2014). Univariable general linear model was used to explore the confounding variables that might affect the changes in blood pressure, and baseline characteristics of age, sex, body mass index, serum creatine, blood pressure, smoking status, alcohol intake and antihypertensive drugs were found to be significantly associated with the changes in blood pressure (Supplementary Table S3), therefore, these variables were adjusted as fixed effect. In addition, random-effect model was also used to re-analyze the main results, which further adjusted entry time and center as random effect. In this study, two parallel models were used to give more comprehensive assessments for the genetic association analysis. Generalized linear regression model was used to calculate the means, mean differences and $95\%$ confidence intervals (CIs) of the changes in blood pressure (ΔSBP and ΔDBP) among different genotypes of the variants. Multivariate linear regression model was used to obtain the correlation coefficient of genetic variants with the changes in blood pressure. The effects of antihypertensive treatments and sex on the correlation of variants with the changes in blood pressure were analyzed in further stratification. In addition, we performed epistasis analysis using the PLINK 1.9 (cog-genomics.org) to further explore the effects of interactions of the studied variants on the changes in blood pressure at the follow-up. The P-values for interaction were calculated by multiple linear regression models with adjustment for covariates as mentioned above. Cox proportional-hazards model was used to compute hazards ratios and $95\%$CIs for the association between variants and the risk of cardiovascular events. Person-years of follow-up began on the date of enrollment until the date of cardiovascular outcomes, death, or the end of follow-up (30 November 2016), whichever came first. False discovery rate (FDR) derived from the Benjamin & Hochberg method (Smeland et al., 2020) was calculated to correct for multiple comparisons of genetic associations. A two-tailed probability value of ≤0.05 was considered as significant. SPSS Statistics 25.0 (SPSS Inc., Chicago, USA) were used to conduct the analysis. ## 3.1 Baseline characteristics of patients with hypertension A total of 1837 patients were included in this study and the baseline characteristics of these patients were shown in Table 1. The mean age of the patients was 63.0 years, and $39.6\%$ were men. The means of baseline SBP and DBP of the patients were 156 mm Hg and 89.5 mm Hg, respectively. The patients received monotherapy or multitherapy of antihypertensive drugs: $64.5\%$ patients had CCBs, $53.2\%$ had ARBs, $12.0\%$ had ACEIs, $22.5\%$ had diuretics, and $20.7\%$ had beta-bockers. TABLE 1 \| Characteristics \| All patients (n = 1837) \| \| --- \| --- \| \| Age, years \| 63.0 ± 9.6 \| \| Men, no. (%) \| 727 (39.6) \| \| BMI, kg/m2 \| 26.3 ± 3.4 \| \| SBP, mm Hg \| 156 ± 22 \| \| DBP, mm Hg \| 89.5 ± 12.1 \| \| Lipids, mmol/L \| Lipids, mmol/L \| \| Total cholesterol \| 5.54 ± 1.04 \| \| Triglycerides \| 1.62 (1.12–2.34) \| \| HDL-C \| 1.36 ± 0.31 \| \| LDL-C \| 3.55 ± 0.83 \| \| Fasting serum glucose, mmol/L \| 5.69 (5.24–6.45) \| \| Serum creatinine, μmol/L \| 76.5 ± 20.4 \| \| Cigarette smoking, no. (%) \| 589 (32.1) \| \| Alcohol intake, no. (%) \| 581 (31.6) \| \| Medical history, no. (%) \| Medical history, no. (%) \| \| Coronary heart disease \| 513 (27.9) \| \| Diabetes \| 390 (21.2) \| \| Stroke \| 349 (19.0) \| \| Antihypertensive drugs, no. (%) \| Antihypertensive drugs, no. (%) \| \| Calcium channel blockers \| 1,184 (64.5) \| \| Angiotensin receptor blockers \| 977 (53.2) \| \| ACE inhibitors \| 221 (12.0) \| \| Diuretics \| 414 (22.5) \| \| Beta-blockers \| 380 (20.7) \| ## 3.2 NR1H3 gene variant rs11039149A>G was associated with the change in SBP The frequencies of 11 studied variants were shown in Supplementary Table S4 and the frequencies distribution of all variants did not deviate significantly from Hardy-Weinberg equilibrium (all $p \leq 0.05$). Means, mean differences and $95\%$CIs of blood pressure changes between genotypes were calculated by generalized linear regression model with adjustment for age, sex, body mass index, serum creatine, blood pressure, smoking status, alcohol intake and antihypertensive drugs. The results showed that NR1H3 gene variant rs11039149A>G was associated with ∆SBP. In this studied population, the minor allele frequency of rs11039149G was $3.0\%$, and the genotype frequencies of rs11039149 were respectively as AA ($94.0\%$) and AG ($6.0\%$). ∆SBP was 2.95 mm Hg ($95\%$CI: −1.08 to 6.98) for the NR1H3 variant rs11039149AG carriers and −2.99 mm Hg ($95\%$CI: −4.71 to −1.26) for the AA carriers; ∆DBP was −0.56 mm Hg ($95\%$CI: −2.68 to 1.57) for the AG carriers and −2.66 mm Hg ($95\%$CI: −3.57 to −1.57) for the AA carriers. The difference of ∆SBP between AG and AA carriers was 5.94 mm Hg ($95\%$CI: 2.09 to 9.78, $$p \leq 0.002$$, FDR = 0.02), but no significant difference of ∆DBP was found between the AG and AA carriers after multiple comparison correction by the Benjamin & Hochberg method (FDR = 0.44) (Table 2). TABLE 2 \| Changes in BP \| Genotype \| Mean (95%CI) a of changes in BP, mm Hg \| Mean difference (95%CI) a of changes in BP, mm Hg \| P a \| FDR b \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| NR1H3 rs11039149A>G \| NR1H3 rs11039149A>G \| \| \| \| \| \| ΔSBP, mm Hg \| AA (n = 1726) \| −2.99 (−4.71, −1.26) \| 5.94 (2.09, 9.78) \| 0.002 \| 0.02 \| \| ΔSBP, mm Hg \| AG (n = 111) \| 2.95 (−1.08, 6.98) \| 5.94 (2.09, 9.78) \| 0.002 \| 0.02 \| \| ΔDBP, mm Hg \| AA (n = 1726) \| −2.66 (−3.57, −1.75) \| 2.11 (0.08, 4.13) \| 0.04 \| 0.44 \| \| ΔDBP, mm Hg \| AG (n = 111) \| −0.56 (−2.68, 1.57) \| 2.11 (0.08, 4.13) \| 0.04 \| 0.44 \| For the NR1H3 variant rs11039149A>G, when stratified by sex, the difference of ∆SBP between AG and AA carriers was observed in both men and women (Supplementary Table S5). For the other 10 variants, the associations with blood pressure changes were not observed in either men, women or the whole patients (Supplementary Tables S5, S6). The correlation coefficients between variants and blood pressure changes were further assessed by multiple linear regression model. The coefficient beta was 5.77 ($$p \leq 0.003$$, FDR = 0.03) for ∆SBP and 2.08 ($$p \leq 0.04$$, FDR = 0.46) for ∆DBP, respectively (Supplementary Table S7). We also performed epistasis analysis to further explore the effects of interactions between rs11039149 and other 10 studied variants on blood pressure changes at the follow-up. No significant epistasis was found to affect blood pressure changes (all P for interaction>0.05) (Supplementary Table S8). Considering the potential effects of entry time and center, we further adjusted these two covariates as random effect in multiple linear mixing model. The results showed that rs11039149AG genotype was significantly associated with SBP increase during the follow-up ($$p \leq 0.01$$), which was consistent with the fixed-effect model (Supplementary Table S9). We compared baseline characteristics of the patients with two genotypes of rs11039149, showing that patients with AG genotype had a lower body mass index and a higher level of serum creatinine than those with AA genotype (Supplementary Table S10). In addition, patients with AG genotype had a higher level of SBP at the follow-up than those with AA genotype. There were no significant differences in other characteristics including age, sex, blood lipid, fasting serum glucose, baseline blood pressure, smoking status, alcohol intake, the usage of antihypertensive drugs and medical history of cardiovascular diseases. ## 3.3 NR1H3 gene variant rs11039149A>G was associated with SBP response to CCBs The association of NR1H3 variant rs11039149A>G with ∆SBP differed in antihypertensive therapy. In 1,184 patients with CCBs therapy, SBP levels decreased in AA carriers (ΔSBP: −2.49 mm Hg, $95\%$CI: −4.45 to −0.53), but increased in AG carriers (ΔSBP: 5.55 mm Hg, $95\%$CI: 0.64 to 10.64), the difference of ΔSBP between AG and AA carriers was 8.04 mm Hg ($95\%$CI: 3.28 to 12.81, $$p \leq 0.001$$, FDR = 0.01) (Table 3). However, there was no significant difference in ΔSBP between AG and AA carriers without CCBs therapy ($$p \leq 0.41$$). As for ARBs, ACEIs, diuretics and beta-blockers, the therapies of these drugs did not significantly affect the association between the NR1H3 variant rs11039149A>G and blood pressure changes (all $p \leq 0.05$) (Supplementary Table S11). TABLE 3 \| Genotype \| Mean (95%CI) a of ΔSBP, mm Hg \| Mean difference (95%CI) a of ΔSBP, mm Hg \| P a \| \| --- \| --- \| --- \| --- \| \| NR1H3 rs11039149A>G \| \| \| \| \| CCBs therapy (n = 1,184) \| CCBs therapy (n = 1,184) \| \| \| \| AA (n = 1,112) \| −2.49 (−4.45, −0.53) \| 8.04 (3.28, 12.81) \| 0.001 \| \| AG (n = 72) \| 5.55 (0.64, 10.46) \| 8.04 (3.28, 12.81) \| 0.001 \| \| CCBs monotherapy (n = 359) \| CCBs monotherapy (n = 359) \| \| \| \| AA (n = 335) \| −0.96 (−6.89, 4.96) \| 15.25 (6.48, 24.02) \| 0.001 \| \| AG (n = 24) \| 14.28 (4.18, 24.38) \| 15.25 (6.48, 24.02) \| 0.001 \| \| CCBs multitherapy (n = 825) \| CCBs multitherapy (n = 825) \| \| \| \| AA (n = 777) \| −3.30 (−5.73, −0.88) \| 4.51 (−1.69, 10.71) \| 0.15 \| \| AG (n = 48) \| 1.21 (−5.14, 7.55) \| 4.51 (−1.69, 10.71) \| 0.15 \| \| non CCBs therapy (n = 653) \| non CCBs therapy (n = 653) \| \| \| \| AA (n = 614) \| −3.39 (−5.47, −1.31) \| 2.05 (−4.42, 8.52) \| 0.41 \| \| AG (n = 39) \| −1.34 (−7.78, 5.10) \| 2.05 (−4.42, 8.52) \| 0.41 \| Among the 1,184 patients with CCBs therapy, 359 ($30.3\%$) patients had CCBs monotherapy and 825 ($69.7\%$) patients had multitherapy. Considering the potential effect of multitherapy on the association of the genotypes with CCBs therapy, we further assessed the differences of ΔSBP in patients with or without rs11039149G allele when receiving CCBs monotherapy and multitherapy. In patients with CCBs monotherapy, the difference of ΔSBP between patients with AG and AA genotype was 15.25 mm Hg ($95\%$CI: 6.48 to 24.02, $$p \leq 0.001$$). However, in patients with CCBs multitherapy, there was no significant difference in ΔSBP between the patients with two genotypes ($$p \leq 0.15$$). Compared with the patients receiving CCBs multitherapy, the difference in ∆SBP was more significant in patients with CCBs monotherapy (P genotype × CCBs therapy interaction = 0.002) (Table 3). Irregular drug taking (defined as the change in types of drugs, and/or the increase/decrease in the dosage, and/or stop the use of drugs without doctors’ prescription) and drug regimen change (defined as the change in types of drugs during the follow-up) may also affect the genetic association. Therefore, after 100 patients with irregular drug taking or 531 patients with drug regimen change were excluded, we further performed the sensitivity analysis. The results showed that the association of rs11039149 with SBP in response to CCBs still remained. ( Supplementary Tables S12, S13). With the use of multiple linear mixing model, entry time and center were further adjusted as random effect. The results showed that rs11039149A>G was significantly associated with SBP response to CCBs monotherapy (P genotype × CCBs therapy interaction = 0.004), which was consistent with the fixed-effect model (Supplementary Table S14). ## 3.4 No variants were associated with the risk of cardiovascular events All patients in this study were followed up for a median 2.24 years, and 159 events were recorded. There were no significant associations between the 11 variants and the risk of cardiovascular events (all $p \leq 0.05$, Supplementary Table S15). ## 3.5 NR1H3 variant rs11039149G allele disabled the transcriptional factor FOXC1 binding Bioinformatic analysis (JASPAR database, http://jaspar.genereg.net/) showed that the variant rs11039149 A to G change at the NR1H3 promoter region could affect the binding ability with transcriptional factor FOXC1, which changed the transcription activity of NR1H3 gene. Therefore, pENTER-FOXC1 and pGL3-promoter luciferase vectors pGL3-rs11039149A and pGL3-rs11039149G were constructed, and the luciferase activity representative of NR1H3 promoter activity was measured by the dual-luciferase assay carried out in HEK293T. The relative luciferase activity had no significant difference between the pGL3-rs11039149A and pGL3-rs11039149G when without any stimulation of transcriptional factors. As for the wild pGL3-rs11039149A having potential binding sequence with the FOXC1, its relative luciferase activity significantly increased by 2.1-fold comparing the presence of FOXC1 with the absence of FOXC1 ($p \leq 0.001$). After co-transfected with the FOXC1 gene, the wild pGL3-rs11039149A had a significantly higher relative luciferase activity by 2-fold than the mutant pGL3-rs11039149G ($p \leq 0.001$) (Figure 2). These results suggested that FOXC1 could promote the transcriptional activity of NR1H3 via binding to the wild rs11039149A allele, while the mutant rs11039149G allele disabled the FOXC1 binding and had no change in transcriptional activity. FIGURE 2: Transcription activity analysis of the NR1H3 promoter containing rs11039149A and rs11039149G alleles in HEK293T. After co-transfected with the FOXC1 gene, the wild pGL3-rs11039149A had a significantly higher relative luciferase activity by 2-fold than the mutant pGL3-rs11039149G (*p < 0.001). The pGL3-rs11039149A and pGL3-rs11039149G luciferase reporter vectors carried the NR1H3 gene promoter fragment containing rs11039149A or rs11039149G allele, respectively. Firefly luciferase activity was expressed relative to pGL3-rs11039149A co-transfected with pENTER empty plasmid. Renilla luciferase activity encoded by the co-transfected control plasmid pRL-TK to standardize the transfection efficiency. Data are presented as mean ± standard error (SE) from three independent experiments. ## 4 Discussion In this longitudinal study, a total of 11 variants at 9 genes (AGT, AGTR1, ADD1, ACE2, CACNA1C, NEDD4L, NR1H3, PTPRD, MMP3) were analyzed in 1837 Chinese patients with hypertension. Our data first provided evidence in Chinese patients with hypertension that NR1H3 promoter variant rs11039149A>G was related to the change in SBP and the SBP response to CCBs monotherapy. However, no significant association was found between NR1H3 variant rs11039149A>G and the risk of cardiovascular events. The genetic association analysis showed that SBP levels decreased in patients with rs11039149AA genotype, but increased in patients with AG genotype. In addition, bioinformatics analysis and dual luciferase assay showed that FOXC1 could bind to rs11039149A and promote the transcriptional activity of NR1H3. The rs11039149A to G change disabled FOXC1 binding with NR1H3. FOXC1 gene encodes fork head protein C1, acting as a transcription factor binding to gene promoter sequence and playing roles in the early development of the heart and blood vessels. NR1H3 gene encodes liver X receptor alpha (LXRA), one of the key transcription factors for cholesterol efflux and inflammatory gene responses in macrophages, and thus takes part in the process of atherosclerosis (Che et al., 2021; Savla et al., 2022). LXRA can bind to the promoter region of renin gene to regulate renin transcription (Morello et al., 2005), and activation of LXRA is reported to suppress the RAAS activation (Kuipers et al., 2010). Animal studies have shown that the LXR agonist GW3965 can reduce blood pressure in rats and have beneficial effects on vascular function (Han et al., 2018; Bal et al., 2019). Supporting, we found that rs11039149G allele is a loss-of-function allele that could disable the FOXC1 binding, which may partly explain the effects of rs11039149G allele on blood pressure changes. In addition, it has been reported that LXR deficiency can lead to atherosclerosis with increased monocyte entry, foam cell formation, and plaque inflammation (Endo-Umeda et al., 2022). A case-control study in the INVEST-GENES cohort showed that rs11039149G in NR1H3 is the protective allele for cardiovascular outcomes in White Americans and Hispanics (Price et al., 2011), while it was not found in our study. One reason may be the difference in ethnics, and another reason may be related to multiple complex factors on atherosclerosis and cardiovascular diseases. NR1H3 gene variant rs11039149A>G was significantly associated with ∆SBP but not with ∆DBP after multiple comparison correction by the Benjamin & Hochberg method. There are several potential explanations. First, the heritability for blood pressure is shown as 30–$50\%$ (Ehret & Caulfield, 2013; Munroe et al., 2013), and SBP has a higher heritability ($46\%$) than that of DBP ($30\%$) in Chinese population (Wu et al., 2011), indicating that SBP may be more susceptible to genetic factors. Second, NR1H3 regulates lipid homeostasis, inflammation and affects the process of arteriosclerosis (Jarvis et al., 2019), and moreover, atherosclerosis has a greater effect on the increase of SBP (Smulyan et al., 2016). Third, animal studies have shown that activation of NR1H3 can significantly reduce SBP in rats and mice (Negishi et al., 2018; Bal et al., 2019), and the possible mechanism may be that NR1H3 could regulate blood pressure through the RAAS (Joseph et al., 2002), which supported the results of the present study to some extent. Another finding of this study showed that NR1H3 variant rs11039149G was associated with SBP increase in patients with CCBs monotherapy. CCBs inhibit Ca2+ influx through L-type calcium channels in myocardium cells and vascular smooth muscle cell (VSMCs), thus relax vasoconstriction and reduce blood pressure (Laurent, 2017). Clunn et al. found that VSMCs transformation from a differentiated phenotype to a synthetic or dedifferentiated phenotype is associated with loss-of-function of L-type calcium channels and hence loss of potential responsiveness to CCBs (Clunn et al., 2010). The process of VSMCs phenotype switching is associated with lipid metabolism and inflammatory response (Shi et al., 2020). Davies et al. found that the direct activation of LXRs in VSMCs up-regulated FASE, SREBP1c, and SCD-1, thus promotes TG accumulation via de novo FA synthesis and SCD-1-mediated Δ9 desaturation (Davies et al., 2005). These studies indicate that NR1H3 gene may play a role in the loss of CCBs effect. In addition, we found that compared with patients receiving CCBs monotherapy, patients receiving CCBs multitherapy had a better blood pressure control. The results showed that the worse SBP control of CCBs in rs11039149AG carriers could be alleviated by combined use with other antihypertensive drugs, which supports the previous studies that combination therapy was more likely to achieve blood pressure goals (Ma et al., 2015; An et al., 2021). One advantage of the present study was that all patients are of Han ethnicity, which avoided the possibility of spurious association caused by population stratification. In addition, the study was a community-based study and all of the patients were recruited from the same geographic area, which minimized the influences of various lifestyles and environments. Several limitations of the present study need to be mentioned. First, drug adherence was not evaluated in this study, but the questionnaires were used to investigate whether the patients took drugs regularly according to the doctor’s prescription during the follow-up. After excluding 100 patients with irregular drug taking, we performed the sensitivity analysis and obtained a consistent result. Second, the adverse drug reactions were not collected in the database, and thus it was unable to assess the associations of genetic variants with the occurrence of adverse reactions or some grades of adverse reactions. However, in this study, the enrolled patients were provided with antihypertensive drugs (including CCBs, ARBs, ACEIs, thiazide-type diuretics, or beta-blockers, unless intolerance was reported), all of which are proved to be safe and most widely used antihypertensive drug types in China. Due to possible adverse effects, the study could not exclude some bias with regarding to the association of NR1H3 variant rs11039149A>G with blood pressure changes in response to CCBs. Third, the efficiency of CCBs was influenced by pharmacodynamic effects directly and correlated with its plasma concentration (Taburet et al., 1983), and the genetic variants may have effects on pharmacokinetics and pharmacodynamics (Guo et al., 2015; Wang et al., 2015). However, considering the long-term follow-up and large patient population in this study, the pharmacokinetic assessment of antihypertensive drugs was not included due to infeasibilities. The multiple-effects of genetic variants on antihypertensive drugs should be evaluated precisely in the future. Fourth, the data of antihypertensive drug dosage of each patient was not collected, and thus was unable to investigate the association in a dose-dependent manner. Given that the difference in drug dosages might affect the genetic association of NR1H3 variant rs11039149A>G with the changes in blood pressure in CCBs, future studies are needed to explore the genetic effects in a dose-dependent manner. Moreover, the results of the present study need to be strengthened by replication in another population, and need to be validated in other races to generalize to a larger population beyond the Han patients. In conclusion, the key finding of the study is that hypertensive patients with rs11039149AG genotype in the NR1H3 gene have a significantly worse SBP control in response to the monotherapy of CCBs, compared with patients with the wild-type rs11039149AA genotype. The underlying biological mechanisms may be related to the disabled binding ability of transcription factor FOXC1 to rs11039149 caused by A to G change. Our findings support that the NR1H3 gene might act as a promising genetic factor to affect individual sensitivity to antihypertensive drugs. Clinical trials in pharmacogenetics will be helpful to investigate genetically determined variations in response to drugs and thus benefit the individual treatment strategies. ## 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 the Ethics Committee of Fuwai Hospital. The patients/participants provided their written informed consent to participate in this study. ## Author contributions YC performed the experiment and wrote the manuscript. YH performed the data analyses and wrote the manuscript. YW, YZ, and SZ helped perform the analysis with constructive discussions. YY and SZ collected the data. RH and WZ conceived and designed the study and gave a critical review of this manuscript. All authors reviewed 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/fphar.2023.1083134/full#supplementary-material ## References 1. An J., Luong T., Qian L., Wei R., Liu R., Muntner P.. Treatment patterns and blood pressure control with initiation of combination versus monotherapy antihypertensive regimens. Hypertension (2021) 77 103-113. DOI: 10.1161/HYPERTENSIONAHA.120.15462 2. Bal N. B., Han S., Usanmaz S. E., Kiremitci S., Sadi G., Uludag O.. Activation of liver X receptors by GW3965 attenuated deoxycorticosterone acetate-salt hypertension-induced cardiac functional and structural changes. J. Cardiovasc Pharmacol. (2019) 74 105-117. DOI: 10.1097/FJC.0000000000000693 3. Beitelshees A. L., Navare H., Wang D., Gong Y., Wessel J., Moss J. I.. CACNA1C gene polymorphisms, cardiovascular disease outcomes, and treatment response. Circ. Cardiovasc Genet. (2009) 2 362-370. DOI: 10.1161/CIRCGENETICS.109.857839 4. Benetos A., Cambien F., Gautier S., Ricard S., Safar M., Laurent S.. Influence of the angiotensin II type 1 receptor gene polymorphism on the effects of perindopril and nitrendipine on arterial stiffness in hypertensive individuals. Hypertension (1996) 28 1081-1084. DOI: 10.1161/01.hyp.28.6.1081 5. Carey R. M., Moran A. E., Whelton P. K.. Treatment of hypertension: A review. JAMA (2022) 328 1849-1861. DOI: 10.1001/jama.2022.19590 6. Che X., Xiao Q., Song W., Zhang H., Sun B., Geng N.. Protective functions of liver X receptor α in established vulnerable plaques: Involvement of regulating endoplasmic reticulum-mediated macrophage apoptosis and efferocytosis. J. Am. Heart Assoc. (2021) 10 e018455. DOI: 10.1161/JAHA.120.018455 7. Chen X., Li S., Yang Y., Yang X., Liu Y., Liu Y.. Genome-wide association study validation identifies novel loci for atherosclerotic cardiovascular disease. J. Thromb. Haemost. (2012) 10 1508-1514. DOI: 10.1111/j.1538-7836.2012.04815.x 8. Citterio L., Bianchi G., Scioli G. A., Glorioso N., Bigazzi R., Cusi D.. Antihypertensive treatment guided by genetics: PEARL-HT, the randomized proof-of-concept trial comparing rostafuroxin with losartan. Pharmacogenomics J. (2021) 21 346-358. DOI: 10.1038/s41397-021-00214-y 9. Clunn G. F., Sever P. S., Hughes A. D.. Calcium channel regulation in vascular smooth muscle cells: Synergistic effects of statins and calcium channel blockers. Int. J. Cardiol. (2010) 139 2-6. DOI: 10.1016/j.ijcard.2009.05.019 10. Davies J. D., Carpenter K. L. H., Challis I. R., Figg N. L., McNair R., Proudfoot D.. Adipocytic differentiation and liver x receptor pathways regulate the accumulation of triacylglycerols in human vascular smooth muscle cells. J. Biol. Chem. (2005) 280 3911-3919. DOI: 10.1074/jbc.M410075200 11. de Denus S., Dubé M.-P., Fouodjio R., Huynh T., LeBlanc M.-H., Lepage S.. A prospective study of the impact of AGTR1 A1166C on the effects of candesartan in patients with heart failure. Pharmacogenomics (2018) 19 599-612. DOI: 10.2217/pgs-2018-0004 12. Ehret G. B., Caulfield M. J.. Genes for blood pressure: An opportunity to understand hypertension. Eur. Heart J. (2013) 34 951-961. DOI: 10.1093/eurheartj/ehs455 13. Endo-Umeda K., Kim E., Thomas D. G., Liu W., Dou H., Yalcinkaya M.. Myeloid LXR (liver X receptor) deficiency induces inflammatory gene expression in foamy macrophages and accelerates atherosclerosis. Arterioscler. Thromb. Vasc. Biol. (2022) 42 719-731. DOI: 10.1161/ATVBAHA.122.317583 14. Fan X., Wang Y., Sun K., Zhang W., Yang X., Wang S.. Polymorphisms of ACE2 gene are associated with essential hypertension and antihypertensive effects of Captopril in women. Clin. Pharmacol. Ther. (2007) 82 187-196. DOI: 10.1038/sj.clpt.6100214 15. Gong Y., McDonough C. W., Beitelshees A. L., El Rouby N., Hiltunen T. P., O'Connell J. R.. PTPRD gene associated with blood pressure response to atenolol and resistant hypertension. J. Hypertens. (2015) 33 2278-2285. DOI: 10.1097/HJH.0000000000000714 16. Guo C., Pei Q. I., Tan H., Huang Z., Yuan H., Yang G.. Effects of genetic factors on the pharmacokinetics and pharmacodynamics of amlodipine in primary hypertensive patients. Biomed. Rep. (2015) 3 195-200. DOI: 10.3892/br.2014.395 17. Han S., Bal N. B., Sadi G., Usanmaz S. E., Uludag M. O., Demirel-Yilmaz E.. The effects of LXR agonist GW3965 on vascular reactivity and inflammation in hypertensive rat aorta. Life Sci. (2018) 213 287-293. DOI: 10.1016/j.lfs.2018.10.042 18. Iniesta R., Campbell D., Venturini C., Faconti L., Singh S., Irvin M. R.. Gene variants at loci related to blood pressure account for variation in response to antihypertensive drugs between Black and white individuals. Hypertension (2019) 74 614-622. DOI: 10.1161/HYPERTENSIONAHA.118.12177 19. Jarvis S., Williamson C., Bevan C. L.. Liver X receptors and male (In)fertility. Int. J. Mol. Sci. (2019) 20 5379. DOI: 10.3390/ijms20215379 20. Jiang S., Hsu Y.-H., Venners S. A., Zhang Y., Xing H., Wang X.. Interactive effect of angiotensin II type 1 receptor (AGT1R) polymorphisms and plasma irbesartan concentration on antihypertensive therapeutic responses to irbesartan. J. Hypertens. (2011) 29 890-895. DOI: 10.1097/HJH.0b013e32834494f6 21. Johnson J. A.. Advancing management of hypertension through pharmacogenomics. Ann. Med. (2012) 1 S17-S22. DOI: 10.3109/07853890.2011.653399 22. Joseph S. B., McKilligin E., Pei L., Watson M. A., Collins A. R., Laffitte B. A.. Synthetic LXR ligand inhibits the development of atherosclerosis in mice. Proc. Natl. Acad. Sci. U. S. A. (2002) 99 7604-7609. DOI: 10.1073/pnas.112059299 23. Kuipers I., van der Harst P., Kuipers F., van Genne L., Goris M., Lehtonen J. Y.. Activation of liver X receptor-alpha reduces activation of the renal and cardiac renin-angiotensin-aldosterone system. Lab. Invest. (2010) 90 630-636. DOI: 10.1038/labinvest.2010.7 24. Kurland L., Liljedahl U., Karlsson J., Kahan T., Malmqvist K., Melhus H.. Angiotensinogen gene polymorphisms: Relationship to blood pressure response to antihypertensive treatment. Results from the Swedish irbesartan left ventricular hypertrophy investigation vs atenolol (SILVHIA) trial. Am. J. Hypertens. (2004) 17 8-13. DOI: 10.1016/j.amjhyper.2003.09.009 25. Laurent S.. Antihypertensive drugs. Pharmacol. Res. (2017) 124 116-125. DOI: 10.1016/j.phrs.2017.07.026 26. Lewington S., Lacey B., Clarke R., Guo Y., Kong X. L., Yang L.. The burden of hypertension and associated risk for cardiovascular mortality in China. JAMA Intern Med. (2016) 176 524-532. DOI: 10.1001/jamainternmed.2016.0190 27. Liljedahl U., Kahan T., Malmqvist K., Melhus H., Syvänen A.-C., Lind L.. Single nucleotide polymorphisms predict the change in left ventricular mass in response to antihypertensive treatment. J. Hypertens. (2004) 22 2321-2328. DOI: 10.1097/00004872-200412000-00014 28. Lu C., Zhang Y., Qin Y., Xu Q., Zhou R., Cui Y.. Human X chromosome exome sequencing identifies BCORL1 as contributor to spermatogenesis. J. Med. Genet. (2021) 58 56-65. DOI: 10.1136/jmedgenet-2019-106598 29. Luo F., Wang Y., Wang X., Sun K., Zhou X., Hui R.. A functional variant of NEDD4L is associated with hypertension, antihypertensive response, and orthostatic hypotension. Hypertension (2009) 54 796-801. DOI: 10.1161/HYPERTENSIONAHA.109.135103 30. Ma J., Wang X.-Y., Hu Z.-D., Zhou Z.-R., Schoenhagen P., Wang H.. Meta-analysis of the efficacy and safety of adding an angiotensin receptor blocker (ARB) to a calcium channel blocker (CCB) following ineffective CCB monotherapy. J. Thorac. Dis. (2015) 7 2243-2252. DOI: 10.3978/j.issn.2072-1439.2015.12.39 31. Manunta P., Bianchi G.. Pharmacogenomics and pharmacogenetics of hypertension: Update and perspectives–the adducin paradigm. J. Am. Soc. Nephrol. (2006) 17 S30-S35. DOI: 10.1681/ASN.2005121346 32. McDonough C. W., Burbage S. E., Duarte J. D., Gong Y., Langaee T. Y., Turner S. T.. Association of variants in NEDD4L with blood pressure response and adverse cardiovascular outcomes in hypertensive patients treated with thiazide diuretics. J. Hypertens. (2013) 31 698-704. DOI: 10.1097/HJH.0b013e32835e2a71 33. Morello F., de Boer R. A., Steffensen K. R., Gnecchi M., Chisholm J. W., Boomsma F.. Liver X receptors alpha and beta regulate renin expression. J. Clin. Invest. (2005) 115 1913-1922. DOI: 10.1172/JCI24594 34. Munroe P. B., Barnes M. R., Caulfield M. J.. Advances in blood pressure genomics. Circ. Res. (2013) 112 1365-1379. DOI: 10.1161/CIRCRESAHA.112.300387 35. Negishi E., Fukuda N., Otsuki T., Katakawa M., Komatsu K., Chen L.. Involvement of complement 3 in the salt-sensitive hypertension by activation of renal renin-angiotensin system in spontaneously hypertensive rats. Am. J. Physiol. Ren. Physiol. (2018) 315 F1747-F1758. DOI: 10.1152/ajprenal.00370.2018 36. Niu M., Zhang L., Wang Y., Tu R., Liu X., Wang C.. Lifestyle score and genetic factors with hypertension and blood pressure among adults in rural China. Front. Public Health (2021) 9 687174. DOI: 10.3389/fpubh.2021.687174 37. Price E. T., Pacanowski M. A., Martin M. A., Cooper-DeHoff R. M., Pepine C. J., Zineh I.. Liver X receptor α gene polymorphisms and variable cardiovascular outcomes in patients treated with antihypertensive therapy: Results from the INVEST-GENES study. Pharmacogenet Genomics (2011) 21 333-340. DOI: 10.1097/FPC.0b013e3283452fec 38. Savla S. R., Prabhavalkar K. S., Bhatt L. K.. Liver X receptor: A potential target in the treatment of atherosclerosis. Expert Opin. Ther. Targets (2022) 26 645-658. DOI: 10.1080/14728222.2022.2117610 39. Sherva R., Ford C. E., Eckfeldt J. H., Davis B. R., Boerwinkle E., Arnett D. K.. Pharmacogenetic effect of the stromelysin (MMP3) polymorphism on stroke risk in relation to antihypertensive treatment: The genetics of hypertension associated treatment study. Stroke (2011) 42 330-335. DOI: 10.1161/STROKEAHA.110.593798 40. Shi J., Yang Y., Cheng A., Xu G., He F.. Metabolism of vascular smooth muscle cells in vascular diseases. Am. J. Physiol. Heart Circ. Physiol. (2020) 319 H613-H631. DOI: 10.1152/ajpheart.00220.2020 41. Smeland O. B., Frei O., Shadrin A., O'Connell K., Fan C.-C., Bahrami S.. Discovery of shared genomic loci using the conditional false discovery rate approach. Hum. Genet. (2020) 139 85-94. DOI: 10.1007/s00439-019-02060-2 42. Smulyan H., Lieber A., Safar M. E.. Hypertension, diabetes type II, and their association: Role of arterial stiffness. Am. J. Hypertens. (2016) 29 5-13. DOI: 10.1093/ajh/hpv107 43. Svensson-Färbom P., Wahlstrand B., Almgren P., Dahlberg J., Fava C., Kjeldsen S.. A functional variant of the NEDD4L gene is associated with beneficial treatment response with β-blockers and diuretics in hypertensive patients. J. Hypertens. (2011) 29 388-395. DOI: 10.1097/HJH.0b013e3283410390 44. Taburet A. M., Singlas E., Colin J. N., Banzet O., Thibonnier M., Corvol P.. Pharmacokinetic studies of nifedipine tablet. Correlation with antihypertensive effects. Hypertension (1983) 5 II29-II33. DOI: 10.1161/01.hyp.5.4_pt_2.ii29 45. Turner S. T., Chapman A. B., Schwartz G. L., Boerwinkle E.. Effects of endothelial nitric oxide synthase, alpha-adducin, and other candidate gene polymorphisms on blood pressure response to hydrochlorothiazide. Am. J. Hypertens. (2003) 16 834-839. DOI: 10.1016/s0895-7061(03)01011-2 46. Vormfelde S. V., Brockmöller J.. The genetics of loop diuretic effects. Pharmacogenomics J. (2012) 12 45-53. DOI: 10.1038/tpj.2010.68 47. Wang X.-F., Yan L., Cao H.-M., Wei L.-M., Yang W.-H., Zhang S.-J.. *Effect of CYP3A41G, CYP3A53, POR28, and ABCB1 C3435T on the pharmacokinetics of nifedipine in healthy Chinese volunteers*. Int. J. Clin. Pharmacol. Ther.* (2015) 53 737-745. DOI: 10.5414/CP202211 48. Wu T., Snieder H., Li L., Cao W., Zhan S., Lv J.. Genetic and environmental influences on blood pressure and body mass index in han Chinese: A twin study. Hypertens. Res. (2011) 34 173-179. DOI: 10.1038/hr.2010.194 49. Xiao Z.-L., Yang M., Chen X.-B., Xie X.-M., Chen M.-F.. Personalized antihypertensive treatment guided by pharmacogenomics in China. Cardiovasc Diagn Ther. (2022) 12 635-645. DOI: 10.21037/cdt-22-154 50. Yu H., Lin S., Zhong J., He M., Jin L., Zhang Y.. A core promoter variant of angiotensinogen gene and interindividual variation in response to angiotensin-converting enzyme inhibitors. J. Renin Angiotensin Aldosterone Syst. (2014) 15 540-546. DOI: 10.1177/1470320313506481
--- title: Banzhilian formula alleviates psoriasis-like lesions via the LCN2/MMP-9 axis based on transcriptome analysis authors: - Meng Xing - Xiaoning Yan - Jiangtao Guo - Wenbin Li - ZhangJun Li - Chun Dong - Jiao Guo - Keshen Qu - Ying Luo journal: Frontiers in Pharmacology year: 2023 pmcid: PMC10025347 doi: 10.3389/fphar.2023.1055363 license: CC BY 4.0 --- # Banzhilian formula alleviates psoriasis-like lesions via the LCN2/MMP-9 axis based on transcriptome analysis ## Abstract Introduction: Oral Banzhilian formula (BZLF) is effective in the clinical treatment of psoriasis. However, the effectiveness and mechanism of different drug delivery routes deserve further study. Methods: First, we established the mouse model of psoriasis using imiquimod (IMQ), and high-performance liquid chromatography (HPLC) was used for the quality control of BZLF. Secondly, Total RNA Sequencing and bioinformatics analysis were used to explore the regulatory mechanism of BZLF in improving psoriatic lesions. Finally, further verification was based on animal experiments. Results: we externally applied BZLF for skin lesions in an imiquimod-induced psoriasis mouse model and found that BZLF alleviated psoriasis-like skin lesions while inhibiting the expression of Ki67 and inflammatory factors (Il17a, Tnf-α, S100a7 and Cxcl1) in skin lesions. Transcriptome sequencing results suggested that BZLF inhibited signalling pathways closely related to psoriatic inflammation, such as the IL-17 signalling pathway, chemokine signalling pathway, TNF signalling pathway, and NF-kappa B signalling pathway, and the protein-protein interaction (PPI) network identified LCN2 as one of the core target genes and screened out its regulated downstream gene MMP9. Discussion: Our findings suggest that the anti-psoriatic mechanism of BZLF involved in downregulating the LCN2/MMP-9 axis. ## 1 Introduction Psoriasis is a chronic inflammatory skin disease with a long course and a tendency to recur. Its clinical manifestations are mainly erythema and scales on the whole body. The prevalence of psoriasis is approximately $0.51\%$–$11.43\%$ (Michalek et al., 2017). The long course of the disease and the high risk of comorbidities, such as cardiovascular disease, tumors, and metabolic syndrome (Fernández-Armenteros et al., 2018; Lee et al., 2019), seriously affect patient health and quality of life, bringing a heavy burden to the social economy (Thomsen et al., 2019). Although biological agents such as interleukin (IL)-17/IL-23 have made certain achievements in recent years (Blauvelt et al., 2022; Thaçi et al., 2022), the treatment of psoriasis still requires the combination of multiple means, especially the application of external medicines. In the guidelines for psoriasis, external medication can be used for both mild psoriasis and for maintenance treatment of psoriasis alone, which is the cornerstone of psoriasis treatment. Traditional Chinese medicine (TCM) is an important part of complementary and alternative medicine, plays an important role in psoriatic prevention and treatment, and was highlighted in the 2018 edition of the Guidelines for the Diagnosis and Treatment of Psoriasis in China (Chinese Medical Association Dermatology Branch Psoriasis Professional Committee, 2019). TCM holds the viewpoint that the principle of external treatment is consistent with the principle of internal treatment. Therefore, a botanical drug in the clinic can be given both orally and externally or in medicated baths. Banzhilian formula (BZLF) is a common TCM prescription for psoriasis that consists of nine traditional Chinese medicinal materials: Scutellaria barbata D. Don, Dictamnus dasycarpus Turcz., Cnidium monnieri (L.) Cuss., Saposhnikovia divaricate (Turcz.) Schischk., Cicadae Periostracum, Dioscorea collettii var. hypoglauca (Palib.) S. J. Pei and C. T. Ting, Nepeta cataria L., Chrysanthemum indicum L., and Taraxacum mongolicum Hand. -Mazz. Previous clinical studies have confirmed that oral BZLF can effectively improve the PASI score and reduce serum levels of TNF-α and VEGF in patients with psoriasis (Li et al., 2014). Although BZLF has a curative effect on psoriasis without significant side effects, the efficacy and mechanism of its external application to treat psoriasis need to be further studied. RNA sequencing (RNA-seq) is an important tool for transcriptomic research and provides a new method for multicomponent and multitarget research in botanical drugs. Therefore, we used RNA-seq to study the potential mechanism by which BZLF protects against psoriasis by externally applying BZLF to mice with imiquimod (IMQ)-induced psoriasis to provide a more scientific basis for the subsequent development of external medicines for psoriasis. ## 2.1 Pharmaceutical composition and plant material of BZLF The BZLF consists of nine traditional Chinese medicinal materials: Scutellaria barbata D. Don [Lamiaceae; Scutellaria barbata], Dictamnus dasycarpus Turcz. [ Rutaceae; Cortex Dictamni], Cnidium monnieri (L.) Cusson [Umbelliferae; Fructus Cnidii], Saposhnikovia divaricata (Turcz.) Schischk. [ Umbelliferae; Saposhnikoviae Radix], Cicadae Periostracum [Cryptotympana pustulata Fabricius; Cicada Slough], Dioscorea collettii var. hypoglauca (Palib.) S. J. Pei and C. T. Ting [Dioscoreaceae; Dioscorea hypoglauca Palibin], Nepeta cataria L. [Lamiaceae; Schizonepeta], Chrysanthemum indicum L. [Compositae; wild chrysanthemum], Taraxacum mongolicum Hand. -Mazz. [ Asteraceae; dandelion]. The ratio is 6:10:5:5:5:5:5:5:5. All materials were extracted, concentrated, dried, and processed into granules. The granules of each material were purchased from Yueyang Hospital of Integrated Traditional Chinese and Western Medicine, Shanghai University of Traditional Chinese Medicine. The granules used in the current study were provided by Sichuan Neo-Green Pharmaceutical Technology Development Co., Ltd. (Sichuan, China). ## 2.2 HPLC HPLC used a Welch Ultimate PLUS C18 250 × 4.6 mm, 5 μm column; a DAD detector with a 220 nm detection wavelength; a flow rate of 1 mL/min; a sample size of 10 μL at 30 °C. Mobile phase A was $0.2\%$ phosphoric acid aqueous solution; and mobile phase B was $0.1\%$ trifluoroacetic acid acetonitrile. The standard application was as follows: caffeic acid (main active ingredient of Taraxacum mongolicum Hand.-Mazz., CAS No. 331-39-5, purity ≥$98\%$), baicalin (main active ingredient of Scutellaria barbata D. Don, CAS No. 27740-01-8, purity ≥$98\%$), obakunone (main active ingredient of Dictamnus dasycarpus Turcz., CAS No. 751-03-1, purity ≥$98\%$), fraxinellone (main active ingredient of Dictamnus dasycarpus Turcz., CAS No. 28808-62-0, purity ≥$98\%$), and osthole (main active ingredient of Cnidium monnieri (L.) Cusson, CAS No. 484-12-8, purity ≥$98\%$). ## 2.3 Animals Male specific-pathogen-free (SPF)-grade BALB/c mice (20–25 g body weight, 5–6 weeks old) were provided by the Shanghai Medical Experimental Animal Center (SCXK Shanghai 2013-0016, Shanghai, China). Mice were maintained in a controlled environment with room temperature at 22–23°C and a 12 h dark/light cycle. The fodder (Shanghai Pu Lu Tong Biological Technology Co., Ltd.) and sterile water were applied. All procedures were approved by and carried out in accordance with regulations of the Ethics Committee of Yueyang Hospital affiliated with the Shanghai University of Traditional Chinese Medicine (No. YYLAC-2021-107). ## 2.4 Experimental grouping and model establishment Mice were randomly divided into three groups after shaving their back hair (2 × 2 cm2). The mice psoriasis model was induced externally by imiquimod (IMQ) (Sichuan Mingxin Pharmaceutical Co., Ltd., Sichuan, China, Drug approval No. H20030128), and the control group was coated with petroleum jelly (Nanchang Baiyun Pharmaceutical Co., Ltd., Jiangxi, China, Drug Approval No. F20050006). The BZLF treatment was used as an external application at 20.4, 40.8, and 81.6 mg/cm2/day dosage. The groups were treated as follows: 1) control group: back was treated with 62.5 mg petroleum jelly. 2) IMQ + NS group: back was treated with 62.5 mg IMQ cream for 6 h, then externally covered with a $0.9\%$ NaCl solution and fixed with a layer of gauze and medical polyurethane film. 3) IMQ + BZLF group: back was treated with 62.5 mg IMQ cream for 6 h, then externally covered with $\frac{20.4}{40.8}$/81.6 mg/cm2 BZLF, respectively, and fixed with a layer of gauze and medical polyurethane film. All treatments were performed by applying the IMQ cream on day 0 and then once daily for 10 consecutive days. PASI scores were used to determine the severity of skin inflammation on the backs of the mice. On day 10, the mice were euthanized by inhalation of carbon dioxide. The lesions on the backs of the mice were collected as reserves. ## 2.5 Total RNA Sequencing Total RNA was extracted using the mirVana miRNA Isolation Kit (Ambion) following the manufacturer’s protocol. RNA integrity was evaluated using the Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, United States). The samples with RNA integrity number (RIN) ≥ 7 were subjected to the subsequent analysis. The libraries were constructed using TruSeq Stranded Total RNA with Ribo-Zero Gold according to the manufacturer’s instructions. Then, these libraries were sequenced on the Illumina sequencing platform (HiSeqTM 2500), and 150 bp/125 bp paired-end reads were generated. ## 2.6 mRNA quantitative and differential analysis Aligning the sequencing reads of each sample with the sequence of mRNA transcript sequences, known lncRNA sequences and lncRNA prediction sequences by Bowtie2, and using eXpress for quantitative gene analysis, the FPKM values and counts values (the number of reads for each gene in each sample) were obtained. The estimateSizeFactors function of the DESeq2R package was used to normalize the counts, and the nbinomTest function was used to calculate p-value and foldchange values for the difference comparison. Differential transcripts with p-values <0.05 and foldchange >2 were selected. ## 2.7 Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis The GO and KEGG pathway databases were used to analyze the functional interpretation and KEGG pathway of genes (Huang et al., 2009). The numbers of genes were counted according to GO terms and KEGG pathways. Fisher’s exact test was used to obtain the p-value, and multiple hypothesis testing was performed to obtain the q-value. The significantly enriched GO terms were defined as p-value <0.05. The same analytic approach was performed to identify significantly enriched KEGG terms of genes. ## 2.8 Construction of core targets for treatment of psoriasis-like lesions with BZLF The STRING database (https://string-db.org) was used to construct an active ingredient–target and target interaction network, while Cytoscape 3.8 software was used for visualization and network topology analysis. Using the average of degree centrality (DC), betweenness centrality (BC), and closeness centrality (CC) as the card values, the nodes meeting the three card values were selected as the key nodes of BZLF potential targets. ## 2.9 H&E solution and immunohistochemistry (IHC) On day 10, the mice were euthanized by inhaling carbon dioxide. The central lesions were fixed with $4\%$ formalin solution, dehydrated, embedded in paraffin, sectioned, and stained with H&E solution, and IHC was performed. Anti-Ki-67 antibody (1:50, ab16667, Abcam) was used in IHC. Quantitative methods for determining the epidermal thickness and positive cell rate have been described in previous studies and were the same as our previous study (Kuai et al., 2021). ## 2.10 RT-PCR The lesions were collected for RT-PCR on day 10. The experimental process included total RNA extraction, synthesis of cDNA from RNA by reverse transcriptase, and amplification and synthesis of target fragments using cDNA as a template under the action of DNA polymerase. The specific process and data statistical methods were the same as in our previous study (Kuai et al., 2018). Primer sequences are shown in Supplementary Table S1. ## 2.11 Western blot Briefly, after the samples were removed from the back of the mice, the total protein was extracted. Then, the samples were loaded for electrophoresis, followed by membrane transfer, sealing, and antibody incubation and detection. The antibodies for Western blotting included anti-LCN2 antibodies (26991-1-AP; Proteintech), anti-MMP9 antibodies (ab228402; Abcam), and β-actin (ab8226; Abcam). ## 2.12 Biochemical indicators On day 10, the mice were euthanized by inhaling carbon dioxide. Blood was collected to separate serum for detection of ALT, AST and CRE using a BECKMAN LX20 full automatic biochemical analyzer. ## 2.13 Statistical methods Data were analyzed via SPSS 24.0 (IBM, New York, United States) and described as mean ± standard deviation. The two groups were compared using the t-test. $p \leq 0.05$ was set as statistically significant. ## 3.1 Quality control of BZLF To conduct quality control of BZLF, we selected five representative main ingredients in BZLF for analysis. Caffeic acid, baicalin, and obakunone are the main active ingredients in Taraxacum mongolicum Hand. -Mazz., Scutellaria barbata D. Don, and Dictamnus dasycarpus Turcz., respectively, and all three ingredients have been reported to have some palliative effects on psoriasis (Yang et al., 2018; Wu L. et al., 2020; Wu X. et al., 2020; Dimitris et al., 2020). Furthermore, fraxinellone, a main active ingredient in Dictamnus dasycarpus Turcz., shows potent anti-inflammatory and immunomodulatory effects that are hepatoprotective and can treat proliferative diseases (Bailly and Vergoten, 2020). Additionally, osthole, a principal component of Cnidium monnieri (L.) Cusson, exerted inhibitory effects on hypoxic HCT116 cells that may be associated with eukaryotic initiation factor 2 alpha phosphorylation-mediated apoptosis and the translational repression of hypoxia-inducible factor-1 (HIF-1) (Peng and Chou, 2022). We previously confirmed that HIF-1α was highly expressed in psoriatic lesions and could be affected by pathways regulated by TCM (Qu et al., 2021). Collectively, we chose caffeic acid, baicalin, obakunone, fraxinellone, and osthole to preliminarily establish quality control for BZLF (Figure 1). FIGURE 1: Quality control of Banzhilian formula (BZLF) through caffeic acid, baicalin, obakunone, fraxinellone, and osthole. Caffeic acid, baicalin, obakunone, fraxinellone, and osthole were detected in both the positive control and the BZLF samples, but neither was found in the negative control. ## 3.2 BZLF relieved skin inflammation and keratinocyte (KC) proliferation in mice with IMQ-induced psoriasis To elucidate the effect of the external application of BZLF on psoriasis, we initially established a psoriasis mouse model with IMQ and determined that 40.8 mg/cm2 was the most effective dose of BZLF (Figure 2). The results showed that on the 10th day of BZLF treatment, the symptoms of psoriasis-like erythema and scaling were significantly improved and were accompanied by a significant decrease in the PASI score (Figures 3A, B). Next, to examine the influence of skin inflammation levels, we measured the expression of inflammatory cytokines related to psoriasis and confirmed that Il17a, Tnf-α, and Cxcl1 mRNA expression in IMQ-induced lesions decreased after treatment with BZLF (Figure 4A). Histopathology suggested that BZLF treatment could reduce epidermal thickness (Figures 4B, C) in IMQ-induced lesions. Moreover, excessive KC proliferation was prevented by BZLF (Figures 4B, D). On the other hand, we examined changes in body weight and the potential toxicity of BZLF in mice after the external application of BZLF and found that it had no significant effect on the body weight of psoriatic mice (Supplementary Figure S1A). Security index tests showed that BZLF did not influence liver or kidney function (Supplementary Figure S1B). In conclusion, we showed that external application of BZLF could alleviate psoriasis-like skin lesions with high safety. FIGURE 2: Effect of different doses of Banzhilian formula (BZLF) on IMQ-induced psoriasis-like skin lesions. (A) Appearance of back lesions in each group on day 10. (B) Psoriasis area severity index (PASI) score (0–12). (C) Representative H&E sections of skin lesions on the 10th day (×200). (D) Quantification of epidermal thickness in the back lesions. Scale bar: 100 µm. The data are expressed as mean ± SD. Four skin lesions in each group were included for analysis. p < 0.05, *p < 0.01, compared with the 20.4 mg/cm2 group. ns, not significant; 40.8 mg/cm2 group compared with the 81.6 mg/cm2 group. FIGURE 3: Banzhilian formula (BZLF) alleviates IMQ-induced psoriasis-like skin lesions in mice. (A) Appearance of the back lesions of the mice in each group on the 10th day. (B) Psoriasis area severity index (PASI) score (0–4) with scales, thickness, erythema, and total score. The data are expressed as mean ± SD. Four skin lesions from each group were included for analysis. # p < 0.05, ## p < 0.01, ### p < 0.001, compared with the control group. p < 0.05, p < 0.01, p < 0.001, compared with the IMQ + NS group. FIGURE 4: Banzhilian formula (BZLF) inhibits inflammation and epidermal proliferation in IMQ-induced psoriasis-like lesions. (A) mRNA expression of IL-17A, TNF-α, and CXCL1 in skin lesions of each group on the 10th day. (B) Representative H&E sections of skin lesions on the 10th day (×200) (left). Representative immunohistochemistry sections of Ki67 nuclear staining (brown) of the back lesions (×200) (right). (C) Quantification of epidermis thickness in back lesions. (D) Quantification of Ki67+ cells in skin lesions. Scale bar: 100 µm. The data are expressed as mean ± SD. Four skin lesions from each group were included for analysis. # p < 0.05, ## p < 0.01, ### p < 0.001, compared with the control group. p < 0.05, p < 0.01, p < 0.001, compared with the IMQ + NS group. ## 3.3.1 Differentially expressed genes (DEGs) following BZLF treatment To further examine the mechanism by which external application of BZLF can treat psoriasis, RNA-seq was performed on the control group and IMQ-induced lesions on day 10 of BZLF or NS treatment. A total of 86.53 G of clean data were obtained. The effective data amount in each sample ranged from 6.98 to 7.42 G, Q30 basic groups ranged from $94.79\%$ to $98.18\%$, and the average GC content was $51.598\%$. The FPKM values and the principal component analysis (PCA) are shown in Supplementary Figure S2. DESeq2 software was used to standardize the counts of each sample gene. Then, the DEGs were screened according to p-values <0.05 and foldchange >2. Between normal skin and psoriasis-like lesions, a total of 3,969 DEGs were identified, of which 1,719 were upregulated, and 2,250 were downregulated (Figure 5A). In comparing untreated psoriasis-like lesions versus lesions treated with BZLF, there were 535 DEGs, of which 121 were upregulated genes and 414 were downregulated genes (Figure 5B). To further identify the DEGs related to psoriasis following BZLF treatment, we examined the intersection of the upregulated and downregulated genes. Finally, 330 overlapping DEGs closely related to psoriasis were screened out, including 92 upregulated genes and 238 downregulated genes (Figure 6A). Next, we verified the top four DEGs (Figure 6B). FIGURE 5: Differentially expressed genes (DEGs) in the different groups. Cluster analysis of DEGs among samples and groups. The color of the heat map indicates the relative gene expression. The deeper orange color indicates higher gene expression, whereas the deeper blue color indicates lower gene expression (Left). Differential expression volcano map reflecting the differently expressed genes. Gray indicates genes with no significant difference, red indicates genes that are significantly upregulated, and blue indicates genes that are significantly downregulated (Right). (A) IMQ + NS group compared with the control group. (B) BZLF group compared with the IMQ + NS group. FIGURE 6: Experimental verification and enrichment analysis of differentially expressed genes (DEGs) after treatment with Banzhilian formula (BZLF). (A) Venn diagram of up- and downregulated differentially expressed genes in the BZLF vs. the IMQ + NS group and the control group vs. the IMQ + NS group. (B) RT-PCR showed the mRNA expressions of the top four DEGs. (C) Enriched KEGG analysis of up- (left) and downregulated (right) DEGs. (D) Enriched Gene Ontology analysis of up- (left) and downregulated(right) DEGs. # p < 0.05, ## p < 0.01, ### p < 0.001, compared with the control group. p < 0.05, p < 0.01, p < 0.001, compared with the IMQ + NS group. ## 3.3.2 Biological characteristics of the potential pathways associated with BZLF treatment To further investigate the potential pathways associated with external BZLF treatment for psoriasis, we evaluated the screened DEGs by KEGG and GO analyses. KEGG analysis indicated that externally applied BZLF upregulated pathways that included the calcium signaling pathway, the cAMP signaling pathway, the cGMP−PKG signaling pathway, aldosterone synthesis and secretion, and salivary secretion (Figure 6C). GO analysis showed that enzyme regulator activity, calcium ion binding, heparin binding, and the neurotransmitter catabolic process were the most significantly upregulated gene categories (Figure 6D). In addition, the IL-17 signaling pathway, cytokine−cytokine receptor interaction, the chemokine signaling pathway, the NOD-like receptor signaling pathway, the TNF signaling pathway, the NF-kappa B signaling pathway, and cell adhesion molecules were downregulated following the external application of BZLF, as shown by KEGG analysis (Figure 6C). The cysteine-type endopeptidase inhibitor, cytokine activity, chemokine activity, cornified envelope, inflammatory response, neutrophil chemotaxis, and immune system process were the most significantly downregulated gene categories revealed by GO analysis (Figure 6D). ## 3.4 Core targets of BZLF in IMQ-induced psoriasis-like lesions Next, we constructed the core targets based on screened DEGs by RNA-seq. After excluding targets with confidences less than 0.4 in the STRING database, we imported the interaction information file containing 167 target proteins into Cytoscape to construct the protein‒protein interaction (PPI) network. By calculating the topological value, 26 core targets were screened out according to the following criteria: DC ≥ 8.31, BC ≥ 0.017, and CC ≥ 0.278 (Supplementary Table S2). These core targets are represented by red and blue nodes in the PPI network, respectively; the red nodes represent the top 10 core targets (Il1b, Itgam, Il17a, CCL20, PTGS2, CXCL2, SELL, MPO, TREM1, and LCN2) that were screened based on DC values (Supplementary Figure S3). ## 3.5 BZLF downregulates lipocalin-2 (LCN2) expression in IMQ-induced psoriasis-like lesions LCN2 is one of the top 10 core targets in the PPI network and likely mediates biological processes involved in the external application of BZLF for treating psoriasis. LCN2, which is a member of the Lipocalin superfamily, is a 25-kDa secreted protein that is expressed in a variety of cells and is involved in the transport of lipophilic small molecules such as steroids, lipopolysaccharides, iron, and fatty acids. Recent studies have suggested that LCN2 inhibits NLRC4 signaling through SREBP2 to alleviate psoriatic dermatitis (Ma et al., 2022). Therefore, we measured the mRNA and protein expression of LCN2 in the different groups on the 10th day and found that the mRNA and protein expression were upregulated in IMQ-induced psoriasis-like skin lesions compared with those in the control group, whereas BZLF inhibited LCN2 mRNA and protein expression (Figures 7A, B). FIGURE 7: Banzhilian formula (BZLF) downregulates LCN2 and MMP-9 expression in lesions. Expression of LCN2 and MMP-9 in skin lesions assessed by Western blot (A) and RT-PCR (B) on day 10. The data are expressed as mean ± SD. Three skin lesions in each group were included for analysis. p < 0.05, p < 0.01, p < 0.001, compared with the IMQ + NS group. ## 3.6 BZLF alleviates IMQ-induced psoriasis-like lesions by inhibiting the LCN2/MMP-9 axis Although we confirmed that externally applied BZLF ameliorates psoriasis-like lesions by inhibiting LCN2, we found that its downstream genes Serbp2 and Nlrc4 were not in our DEG list, as determined by RNA-seq. To further investigate the mechanism by which BZLF regulates LCN2, we conducted a protein‒protein interaction analysis using the Search Tool for the Retrieval of Interaction Gene/Proteins (STRING) database based on our sequencing results (Supplementary Table S3). The results showed that among the DEGs, MMP-9, LRP2, TIMP1, IL17A, and SAA1 had higher combined scores for LCN2, and the MMP-9 combined scores were the highest. Therefore, MMP-9 was chosen to examine the possible mechanism by which BZLF downregulates LCN2 to improve psoriasis-like skin lesions. Western blot (WB) showed that BZLF significantly reduced the protein expression of MMP-9 (Figure 7A), while RT-PCR indicated the same change at the mRNA level (Figure 7B). ## 4 Discussion In this study, we administered BZLF orally and found that external application of BZLF could improve erythema and scaling and reduce epidermal thickness in IMQ-induced psoriasis-like mouse skin lesions. According to the two main pathological characteristics of inflammatory cell infiltration and abnormal keratinocyte proliferation in psoriasis, we confirmed that BZLF downregulated the expression of inflammatory factors (Il17a, Tnf-α, and Cxcl1) in psoriasis-like lesions and inhibited the expression of the keratinocyte proliferation marker Ki67. Therefore, we believe that the external application of BZLF has a potential therapeutic effect and research value in treating psoriasis. Given the complexity of the composition and mechanism of action of traditional Chinese medicinal materials, we tried to elucidate the pharmacological mechanism of the external application of BZLF from a holistic perspective. We used RNA-seq to analyze skin tissue samples in the different groups, took the intersection of DEGs in the different groups, and verified the accuracy of the sequencing results by RT-PCR. The therapeutic effects of BZLF were regulated by core genes, especially Lcn2, which we identified through bioinformatics analysis of the RNA-seq results. We further used Western blotting and RT-PCR to confirm that BZLF downregulated the expression of LCN2 in IMQ-induced psoriasis-like mouse skin lesions. LCN2 is highly expressed in the skin lesions and the serum of patients with psoriasis (El-Hadidi et al., 2014; Wang et al., 2019) and can inhibit the synthesis of keratin, involucrin, and loricrin in KCs, leading to epidermal parakeratosis via the Tcf7L1-lipocalin 2 signaling axis. LCN2 can also recruit inflammatory cells, such as T cells and neutrophils, to skin lesions through the IL-23/IL17, p38-MAPK, and ERK-$\frac{1}{2}$ signaling pathways (Shao et al., 2016). In addition, LCN2 and other cytokines, such as IL-17, have a synergistic effect on skin cells (Hau et al., 2016). According to these results, we hypothesized that BZLF alleviated inflammation in skin lesions and inhibited the proliferation of KCs, which was associated with the downregulation of LCN2 expression. To further examine the regulatory mechanism by which BZLF acts on LCN2 to improve psoriasis-like skin lesions, we used the STRING database to predict target genes that might interact with LCN2 based on the DEGs from the RNA-seq results. The analysis results showed that MMP-9, which is an inflammatory factor, had the highest combined scores. Next, we found that MMP-9 was significantly elevated in the IMQ-induced psoriasis mouse model, and BZLF downregulated its expression at both the protein and RNA levels. Neutrophil infiltration and tortuous telangiectasia are pathological features of psoriasis (Chau et al., 2017), and MMP-9 can decompose a 62-amino acid peptide from IL-8 (CXCL8/CL8) to increase the chemotactic activity of neutrophils (Mieke et al., 2018). MMP-9 also participates in angiogenesis by releasing vascular endothelial growth factor (VEGF) (Zeng et al., 2020). Moreover, studies have indicated that MMP-9 induces skin vasodilation and hyperpermeability by activating vascular endothelial cells in skin, thereby promoting the development of psoriatic lesions (Chen et al., 2020). At present, MMP-9 has been shown to play an important role in the pathogenesis of psoriasis (Kvist-Hansen et al., 2021; Lu et al., 2022). Therefore, we hypothesized that BZLF alleviated inflammation in psoriasis-like skin lesions and inhibited the proliferation of KCs, which was related to the downregulation of the LCN2/MMP-9 axis. ## 5 Conclusion We examined the efficacy of the external application of BZLF in the treatment of psoriasis and analyzed its mechanism of action by RNA-seq and experimental validation. The specific regulatory mechanism of BZLF mainly involves the upregulation of lipid metabolism-related signaling pathways, downregulation of inflammation-related signaling pathways, and inhibition of the LCN2/MMP-9 axis. This study conducted a preliminary exploration of the external application of BZLF and provided an important material basis for the subsequent research and development of external TCM treatments. ## Data availability statement The data generated from this article can be found in the Gene Expression Omnibus database (https://www.ncbi.nlm.nih.gov/geo/), using accession number GSE223468. ## Ethics statement The animal study was reviewed and approved by the Ethics Committee of Yueyang Hospital of Integrated Traditional Chinese and Western Medicine (No. YYLAC-2021-107). ## Author contributions MX and YL conceptualized and planned the experiments. KQ, JTG, and CD performed most of the experiments and completed the original draft. CD and ZL completed the RNA-seq analysis. JTG completed HPLC. JTG, JG, and WL completed the verification experiments. KQ and XY analyzed the data. MX and JTG raised the animals and completed the protein-protein interaction analysis. YL and MX guided the experiments. All authors were involved in the writing and criticizing review of the manuscript and approved its final 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/fphar.2023.1055363/full#supplementary-material ## References 1. Bailly C., Vergoten G.. Fraxinellone: From pesticidal control to cancer treatment. Pestic. Biochem. Physiol. (2020) 168 104624. DOI: 10.1016/j.pestbp.2020.104624 2. Blauvelt A., Shi N., Somani N., Burge R., Zhu B. J., Terri L.. Comparison of two-year treatment adherence, persistence, discontinuation, reinitiation, and switching between psoriasis patients treated with ixekizumab or secukinumab in real-world settings. J. Am. Acad. Dermatol (2022) 86 581-589. DOI: 10.1016/j.jaad.2021.06.878 3. Chau T., Parsi K. K., Ogawa T., Kiuru M., Konia T., Chin-Shang L. Psoriasis or not? Review of 51 clinically confirmed cases reveals an expanded histopathologic spectrum of psoriasis. J. Cutan. pathology (2017) 44 1018-1026. DOI: 10.1111/cup.13033 4. Chen J. L., Zhu Z. L., Li Q. Y., Lin Y. T., Dang E. L., Meng H.. Neutrophils enhance cutaneous vascular dilation and permeability to aggravate psoriasis by releasing matrix metallopeptidase 9. J. Investigative Dermatology (2020) 141 787-799. DOI: 10.1016/j.jid.2020.07.028 5. Guideline for the diagnosis and treatment of psoriasis in China (2018 Edition). Chin. J. Dermatology (2019) 052 667-710 6. Dimitris D., Ekaterina-Michaela T., Christina K., Ioannis S., Ioanna S. K., Aggeliki L.. Melissa officinalis ssp. altissima extracts: A therapeutic approach targeting psoriasis in mice. J. Ethnopharmacol. (2020) 246 112208. DOI: 10.1016/j.jep.2019.112208 7. El-Hadidi H., Samir N., Shaker O. G., Otb S.. Estimation of tissue and serum lipocalin-2 in psoriasis vulgaris and its relation to metabolic syndrome. Arch. Dermatol Res. (2014) 306 239-245. DOI: 10.1007/s00403-013-1414-x 8. Fernández-Armenteros J. M., Gómez-Arbonés X., Buti-Soler M., Betriu-Bars A., Sanmartin-Novell V., Ortega-Bravo M.. Psoriasis, metabolic syndrome and cardiovascular risk factors. A population-based study. J. Eur. Acad. Dermatol Venereol. (2018) 33 128-135. DOI: 10.1111/jdv.15159 9. Hau C. S., Kanda N., Tada Y., Shibata S., Uozaki H., Fukusato T.. Lipocalin-2 exacerbates psoriasiform skin inflammation by augmenting T-helper 17 response. J. Dermatol (2016) 43 785-794. DOI: 10.1111/1346-8138.13227 10. Huang D. W., Sherman B. T., Lempicki R. A.. Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat. Protoc. (2009) 4 44-57. DOI: 10.1038/nprot.2008.211 11. Kuai L., Luo Y., Qu K. S., Ru Y., Luo Y., Ding X. J.. Transcriptomic analysis of the mechanisms for alleviating psoriatic dermatitis using taodan granules in an imiquimod-induced psoriasis-like mouse model. Front. Pharmacol. (2021) 12 632414. DOI: 10.3389/fphar.2021.632414 12. Kuai L., Zhang J. T., Deng Y., Xu S., Xu X. Z., Wu M. F.. Sheng-ji Hua-yu formula promotes diabetic wound healing of re-epithelization via Activin/Follistatin regulation. BMC Complement. Altern. Med. (2018) 18 32. DOI: 10.1186/s12906-017-2074-8 13. Kvist-Hansen A., Kaiser H., Wang X., Krakauer M., Gørtz P. M., McCauley B. D.. Neutrophil pathways of inflammation characterize the blood transcriptomic signature of patients with psoriasis and cardiovascular disease. Int. J. Mol. Sci. (2021) 22 10818. DOI: 10.3390/ijms221910818 14. Lee J. H., Kim H. J., Han K. D., Kim H. N., Park Y. M., Lee J. Y.. Cancer risk in 892089 patients with psoriasis in korea: A nationwide population-based cohort study. J. Dermatology (2019) 46 95-102. DOI: 10.1111/1346-8138.14698 15. Li W. B., Yan X. N., Tian Q., Liu Y. T., Zhao Y. D., Li M.. Effects of Banzhilian prescription on serum TNF-α and VEGF in patients with psoriasis vulgaris blood heat syndrome. Shaanxi J. Traditional Chin. Med. (2014) 3. DOI: 10.3969/j.issn.1000-7369.2014.09.060 16. Lu J., Wang Y., Li Y., Zhong X., Gong Y., Ding Y.. Based on gene expression analysis: Low-density neutrophil expression is a characteristic of the fast responders treated with guselkumab for psoriasis. Front. Immunol. (2022) 13 865875. DOI: 10.3389/fimmu.2022.865875 17. Ma J., Chen J., Xue K., Yu C., Dang E., Qiao H.. LCN2 mediates skin inflammation in psoriasis through the SREBP2NLRC4 Axis. J. Invest Dermatol (2022) 142 2194-2204. DOI: 10.1016/j.jid.2022.01.012 18. Michalek I. M., Loring B., John S. M.. A systematic review of worldwide epidemiology of psoriasis. JEADV (2017) 31 205-212. DOI: 10.1111/jdv.13854 19. Mieke G., Mieke D. B., Sara A. S., Jennifer V., Sofie K., Noëmie P.. Matrix metalloproteinase-9-generated COOH-but not NH2-terminal fragments of serum amyloid A1 retain potentiating activity in neutrophil migration to CXCL8, with loss of direct chemotactic and cytokine-inducing capacity. Front. Immunol. (2018) 2018 1081-3224. DOI: 10.3389/fimmu.2018.01081 20. Peng K. Y., Chou T. C.. Osthole exerts inhibitory effects on hypoxic colon cancer cells via EIF2[formula: See text] phosphorylation-mediated apoptosis and regulation of HIF-1[Formula: See text]. Am. J. Chin. Med. (2022) 50 621-637. DOI: 10.1142/S0192415X22500240 21. Qu K. S., Luo Y., Yan X. N., Kuai L., Ru Y., Luo Y.. Qinzhuliangxue mixture alleviates psoriasis-like skin lesions via inhibiting the IL6/STAT3 axis. J. Ethnopharmacol. (2021) 274 114041. DOI: 10.1016/j.jep.2021.114041 22. Shao S., Cao T., Jin L., Li B., Fang H., Zhang J. Y.. Increased lipocalin-2 contributes to the pathogenesis of psoriasis by modulating neutrophil chemotaxis and cytokine secretion. J. Invest Dermatol (2016) 136 1418-1428. DOI: 10.1016/j.jid.2016.03.002 23. Thaçi D., Eyerich K., Pinter A., Sebastian M., Unnebrink K., Rubant S.. Direct comparison of risankizumab and fumaric acid esters in systemic therapy–naïve patients with moderate-to-severe plaque psoriasis: A randomized controlled trial. Br. J. Dermatology (2022) 186 30-39. DOI: 10.1111/bjd.20481 24. Thomsen S. F., Skov L., Dodge R., Hedegaard M. S., Kjellberg J.. Socioeconomic costs and health inequalities from psoriasis: A cohort study. Dermatology (2019) 235 372-379. DOI: 10.1159/000499924 25. Wang D., Fang L., Pan G.. Association of serum lipocalin-2 concentrations with psoriasis and psoriatic arthritis: An updated meta-analysis. Dis. markers (2019) 2019 7361826-7361827. DOI: 10.1155/2019/7361826 26. Wu L., Zhao K. Q., Wang W., Cui L. N., Hu L. L., Jiang X. X.. Nuclear receptor coactivator 6 promotes HTR-8/SVneo cell invasion and migration by activating NF-κB-mediated MMP9 transcription. Cell Prolif. (2020a) 53 e12876. DOI: 10.1111/cpr.12876 27. Wu X., Deng X., Wang J., Li Q.. Baicalin inhibits cell proliferation and inflammatory cytokines induced by tumor necrosis factor α (TNF-α) in human immortalized keratinocytes (HaCaT) human keratinocytes by inhibiting the STAT3/nuclear factor kappa B (NF-κB) signaling pathway. Med. Sci. Monit. (2020b) 26 e919392. DOI: 10.12659/MSM.919392 28. Yang L., Meng X., Kuang H.. Comparisons of the pharmacokinetic and tissue distribution profiles of withanolide B after intragastric administration of the effective part of Datura metel L. in normal and psoriasis Guinea pigs. J. Chromatogr. B Anal. Technol. Biomed. Life Sci. (2018) 1083 284-288. DOI: 10.1016/j.jchromb.2018.02.022 29. Zeng Y., Wei L. J., Lali S. M., Chen Y. T., Yu J., Feng L.. miR-150-5p mediates extravillous trophoblast cell migration and angiogenesis functions by regulating VEGF and MMP9. Placenta (2020) 93 94-100. DOI: 10.1016/j.placenta.2020.02.019
--- title: Association between sarcopenia and kidney stones in United States adult population between 2011 and 2018 authors: - Yifan Zhang - Changxiu Tian - Yidi Wang - Houliang Zhang - Jinliang Ni - Wei Song - Huajuan Shi - Tao Zhang - Changbao Xu - Keyi Wang - Bo Peng journal: Frontiers in Nutrition year: 2023 pmcid: PMC10025351 doi: 10.3389/fnut.2023.1123588 license: CC BY 4.0 --- # Association between sarcopenia and kidney stones in United States adult population between 2011 and 2018 ## Abstract ### Purpose To investigate the relationship between kidney stones and sarcopenia in United States adult population between 2011 and 2018. ### Materials and methods We conducted a cross-section study based on the National Health and Nutrition Examination Survey (NHANES) including 39,156 individuals. Sarcopenia was assessed by the sarcopenia index. Association between kidney stones and sarcopenia verified by multiple logistic regression analysis and dose–response curves analysis using restricted cubic spline (RCS) regression. Meanwhile, propensity score matching (PSM) was performed to exclude the effect of confounding variables. ### Results There were 9,472 participants in the study by our accurate enrollment screening process. The odds of kidney stones decreased significantly with the increase of sarcopenia index. Logistic regression analysis showed that sarcopenia expressed significant differences in the participants which suffered kidney stone before PSM ($p \leq 0.001$). In model 4, adjusting all relevant covariates shown that adjusted odds ratio (aOR) of the $95\%$ confidence intervals for kidney stones in all participants, age <39 years and age ≥40 years, were, respectively, 1.286 (1.006–1,643), 1.697 (1.065–2.702), and 0.965 (0.700–1.330) for sarcopenia, and p values were 0.044, 0.026, and 0.827. After performing PSM, the aOR of the $95\%$ in modal 4 for kidney stones in all participants and age <40 year were 2.365 (1.598–3.500) and 6.793 (2.619–17.6180), respectively ($p \leq 0.01$), and especially the aOR in participants (age ≥40) was 1.771(1.138–2.757) with p value being 0.011. ### Conclusion Sarcopenia was positively related to the potential risk of kidney stones in the United States adult population. ## Introduction Urolithiasis was the most known urology disease around the worlds, and it caused a certain problem of various groups of patients. Kidney stones as the most common type of urolithiasis had an increasing prevalence over the past decades, placing high costs and clinical burdens on the healthcare system [1]. In the United States, kidney stones spent more than $2.1 billion on healthcare in 2000 [2]. Kidney stones formation had been shown to be related to environmental and genetic factors such as climate, diet, fluid intake, smoking, caffeine, age, gender, body mass index (BMI), and type 2 diabetes (DM) [3]. Sarcopenia had been defined as a progressive and systemic skeletal muscle disease associated with accelerated loss of muscle mass and function. More recently, sarcopenia had been defined as a disease with many adverse effects such as falls, functional decline, weakness, and death [4]. In 2018, EWGSOP2 updated the definition and diagnostic guidelines for sarcopenia, stating that people with low muscle strength, muscle mass/mass, and physical performance would be diagnosed with sarcopenia [5]. Interestingly, we found a strong correlation between sarcopenia and stones, and it had not been reported in any study. The purpose of this study was to investigate the exposure-response relationship between sarcopenia and the incidence of kidney stones in the National Health and Nutrition Examination Survey (NHANES) from 2011 to 2018. ## Study design and participants NHANES was a program that assessed the health and nutritional status of the American people, which combined interviews and physical examinations to collect data focusing on diet and health-related incidents. The study protocol was endorsed by the NHANES Institutional Review Board, and all participants provided informed consent during the survey. In this study, there were 39,156 participants in eight NHANES cycles 2011–2012, 2012–2013, 2013–2014, 2014–2015, 2015–2016, 2016–2017, and 2017–2018. The study excluded participants if their conditions met any of the following: [1] Their age was under 20 years old ($$n = 22$$,617); [2] They had no kidney stones or sarcopenia information ($$n = 10$$,815); [3] Their marital status, BMI, education, hypertension, and diabetes were unknown ($$n = 10$$,285); [4] Their blood urea nitrogen, creatinine and uric acid were unknown ($$n = 9$$,472). Finally, we collected 9,472 participants, as shown in Figure 1. Figure 1: Schematic flow diagram of inclusion and exclusion criteria for our study cohort. ## Exposure variable and outcomes variable Sarcopenia was the major exposure variable in this study. Sarcopenia was assessed by the sum of the muscle mass of the four limbs (ALM, appendicular lean mass). Dual-energy X-ray absorptiometry (DEXA) was used to measure ALM by NHANES. Pregnant participants and those participants who weight more than 136.4 kg or height more than 192.5 cm were excluded from the study, because these individuals could not be measured by DEXA. We calculated the sarcopenia index as following: sarcopenia index = total appendicular skeletal muscle mass (in kg)/BMI (kg/m2). Sarcopenia was defined by sarcopenia index: it judged to exist sarcopenia if sarcopenia index of men and women was less than 0.789 and 0.512, respectively. The major outcome variable of the study was the kidney stones history. The kidney stones history was assessed by the answers of participants. The participants who had suffered kidney stones were divided into kidney stones groups, and the rest of participants were divided into non-kidney stones groups. ## Potential covariates Based on previous studies, relevant covariates were identified. Continuous variables included age (<40 years/≥40 years); body mass index (BMI); blood urea nitrogen; and creatinine and uric acid. Categorical variables included Gender (male/female); Race (non-Hispanic white/non-Hispanic black/Mexican American/other Hispanic/other); Education level (less than high school/high school or equivalent/college or above); Marital status (married/unmarried); Hypertension; Smoking status (never/former/current); Alcohol use; Vigorous recreational activities; Moderate recreational activities; Sarcopenia; blood urea nitrogen; creatinine; and uric acid. ## Statistical analysis Continuous variables and categorical variables were presented as mean ± standard deviation and number (percentage), respectively. Comparisons among different groups were performed by t-tests and one-way ANOVA tests for normally distributed continuous variables, then non-normal continuous variables was compared by independent-samples Kruskal–Wallis tests, and categorical variables among different groups were determined statistical differences by Chi-square tests. Logistic regression analyzed the relationship between sarcopenia and the presence of kidney stones, using the corrected odds ratio (OR) and corresponding $95\%$ confidence intervals (CI) to describe the associations. In the extended model, model 1 was univariate analysis; model 2 was modified gender, age, and race; model 3 was model 2 plus education level, marital status, and BMI; model 4 was model 3 adding hypertension, smoking status, alcohol use, physical activities, blood urea nitrogen, creatinine, and uric acid. The powerful tool of restricted cubic spline (RCS) function was applied to describing dose–response relationships between continuous variables and outcomes, and was also utilized in our research to characterize the dose–response relationship among sarcopenia index and kidney stone risk, adjusted for model variables. All statistical analyses were conducted using IBM SPSS 20.0 software (IBM, United States) and GraphPad Prism8 software (GraphPad Software Inc., La Jolla, CA, United States). p-Values less than 0.05 were considered to be statistically significant. ## Participants’ characteristics As shown in Figure 1, 39,156 participants were recruited from the NHANES (2011–2018). During screening, basic characteristics of participants are summarized in Table 1. Eight thousand seven hundred and thirteen ($92\%$) participants were divided into none-stone formers group and 759 ($8\%$) into stone formers. Gender and education level had no significance between the two groups ($p \leq 0.05$). The others had significant differences between both groups, including age, race, marital status, BMI, hypertension, smoking status, alcohol use, vigorous recreational activities, moderate recreational activities, sarcopenia, blood urea nitrogen, creatinine, and uric acid ($p \leq 0.05$). Supplementary Table S1 summarizes basic characteristics of participant with no-sarcopenia group and sarcopenia group. Eight thousand six hundred and sixty-one ($91.4\%$) participants were divided into no-sarcopenia and 811 ($8.6\%$) into sarcopenia. We found that people with sarcopenia were more likely to be elderly (age ≥ 40 years), married, Mexican American, college degree or above, higher BMI, less vigorous recreational activities, less moderate recreational activities, lower creatinine, and higher uric acid ($p \leq 0.05$). Table 1 \| Characteristic \| Total \| None-stone formers \| Stone formers \| p-Value \| \| --- \| --- \| --- \| --- \| --- \| \| Characteristic \| N (%) \| N (%) \| N (%) \| p-Value \| \| Total patients \| 9472 \| 8,713 (92.0) \| 759 (8.0) \| \| \| Gender \| \| \| \| 0.587 \| \| Male \| 4,657 (49.2) \| 4,291 (49.2) \| 366 (48.2) \| \| \| Female \| 4,815 (50.8) \| 4,422 (50.8) \| 393 (51.8) \| \| \| Age \| \| \| \| <0.001 \| \| <40 years \| 4,793 (50.6) \| 4,518 (51.9) \| 275 (36.2) \| \| \| ≥40 years \| 4,679 (49.4) \| 4,195 (48.1) \| 484 (63.8) \| \| \| Race \| \| \| \| <0.001 \| \| Non-Hispanic white \| 3,428 (36.2) \| 3,055 (35.1) \| 373 (49.1) \| \| \| Non-Hispanic black \| 1934 (20.4) \| 1837 (21.1) \| 97 (12.8) \| \| \| Mexican American \| 1,374 (14.5) \| 1,274 (14.6) \| 100 (13.2) \| \| \| Other Hispanic \| 948 (10.0) \| 856 (9.8) \| 92 (12.1) \| \| \| Other \| 1788 (18.9) \| 1,691 (19.4) \| 97 (12.8) \| \| \| Education level \| \| \| \| 0.695 \| \| Less than high school \| 1,610 (17.0) \| 1,479 (17.0) \| 131 (17.3) \| \| \| High school or equivalent \| 2063 (21.8) \| 1907 (21.9) \| 156 (20.6) \| \| \| College or above \| 5,799 (61.2) \| 5,327 (61.1) \| 472 (62.2) \| \| \| Marital status \| \| \| \| 0.011 \| \| Married \| 4,625 (48.8) \| 4,221 (48.4) \| 404 (53.2) \| \| \| Unmarried \| 4,847 (51.2) \| 4,492 (51.6) \| 355 (46.8) \| \| \| BMI (kg/m2) \| \| \| \| <0.001 \| \| <25.0 \| 3,001 (31.7) \| 2,835 (32.6) \| 166 (21.9) \| \| \| 25.0–29.9 \| 2,966 (31.3) \| 2,731 (31.4) \| 235 (31.0) \| \| \| ≥30.0 \| 3,497 (37.0) \| 3,139 (36.1) \| 358 (47.2) \| \| \| Hypertension \| \| \| \| <0.001 \| \| Yes \| 2,223 (23.5) \| 1945 (22.3) \| 278 (36.6) \| \| \| No \| 7,249 (76.5) \| 6,768 (77.7) \| 481 (63.4) \| \| \| Smoking status \| \| \| \| <0.001 \| \| Never \| 5,733 (60.5) \| 5,332 (61.2) \| 401 (52.8) \| \| \| Former \| 1,613 (17.0) \| 1,457 (16.7) \| 156 (20.6) \| \| \| Current \| 2,126 (22.4) \| 1924 (22.1) \| 202 (26.6) \| \| \| Alcohol use \| \| \| \| 0.010 \| \| Yes \| 7,040 (74.3) \| 6,446 (74.0) \| 594 (78.3) \| \| \| No/Unknown \| 2,432 (25.7) \| 2,267 (26.0) \| 165 (21.7) \| \| \| Vigorous recreational activities \| \| \| \| <0.001 \| \| Yes \| 2,967 (31.3) \| 2,779 (31.9) \| 188 (24.8) \| \| \| No \| 6,505 (68/7) \| 5,934 (68.1) \| 571 (75.2) \| \| \| Moderate recreational activities \| \| \| \| 0.042 \| \| Yes \| 4,277 (45.2) \| 3,961 (45.5) \| 316 (41.6) \| \| \| No \| 5,195 (54.8) \| 4,752 (54.5) \| 443 (58.4) \| \| \| Sarcopenia \| \| \| \| <0.001 \| \| Yes \| 811 (8.6) \| 715 (8.2) \| 96 (12.6) \| \| \| No \| 8,661 (91.4) \| 7,998 (91.8) \| 663 (87.4) \| \| \| Blood urea nitrogen (mg/dl) \| 12.53 ± 4.50 \| 12.48 ± 4.40 \| 13.16 ± 5.42 \| <0.001 \| \| Creatinine (mg/dl) \| 0.85 ± 0.37 \| 0.85 ± 0.34 \| 0.89 ± 0.58 \| <0.001 \| \| Uric acid (mg/dl) \| 5.32 ± 1.39 \| 5.32 ± 1.39 \| 5.35 ± 1.39 \| <0.001 \| ## Sarcopenia and kidney stones Figure 2 shows the dose–response relationships between sarcopenia index and kidney stones. There was a no-linear association between LMA (lumbar muscle area) and the prevalence of kidney stones. It showed that the prevalence of kidney stones decreased with increase of sarcopenia index. Then, logistic regression analysis further confirmed that sarcopenia was positively related with prevalence of kidney stones (Table 2). The adjusted odds ratio (aOR) of all participants was 1.620 ($95\%$ CI, 1.290–2.033), 1.458 ($95\%$ CI, 1.152–1.846), 1.287 ($95\%$ CI, 1.010–1.639), and 1.286 ($95\%$ CI, 1.006–1.643), respectively ($p \leq 0.05$). The adjusted odds ratio (aOR) of participants (<40 years) was 1.955 (1.304–2.930), 2.089 (1.377–3.169), 1.926 (1.254–2.959), and 1.697 (1.065–2.702), respectively ($p \leq 0.05$). However, there were no significant differences about aOR on participants (≥40 years). There were adjusted covariate in four models: model 1: univariate analysis; model 2: gender, age and race; model 3: model 2 plus education level, marital status, and BMI; and model 4: model 3 plus hypertension, smoking status, alcohol use, physical activities, blood urea nitrogen, creatinine, and uric acid. Figure 2: Relative risk for kidney stones based on sarcopenia index before PSM. The shaded areas represent upper and lower $95\%$ CIs. Adjustment factors are as same as which presented in extended model 4. Restricted cubic spline (RCS) plot of the association between sarcopenia index and kidney stones. The solid and dashed lines represent the odds ratios and $95\%$ confidence intervals. TABLE_PLACEHOLDER:Table 2 ## Association after propensity score matching Participants in the study were performed propensity score matching, because of differences between none-stone formers group and formers group. The results after propensity score matching are shown in Figures 3, 4. After propensity score matching, logistic regression analysis clearly shown that sarcopenia was positively related with prevalence of kidney stones. Table 3 displays that p-values were less than 0.05 among participants (all, <40 years and ≥40 years) in four models. The adjusted odds ratio (aOR) of all participants was 2.325 (1.602–3.376), 2.300 (1.572–3.366), 2.342 (1.588–3.455), and 2.365 (1.598–3.500), respectively ($p \leq 0.01$). Then, the adjusted odds ratio (aOR) of participants (<40 years) was 5.600 (2.287–13.710), 5.945 (2.362–14.964), 6.334 (2.479–16.180), and 6.793 (2.619–17.618), respectively ($p \leq 0.01$). Finally, the adjusted odds ratio (aOR) of participants (≥40 years) was 1.787 (1.173–2.723), 1.761 (1.148–2.700), 1.756 (1.132–2.723), and 1.771 (1.138–2.757), respectively ($p \leq 0.05$). There were adjusted covariates in four models: model 1: univariate analysis; model 2: gender, age, and race; model 3: model 1 plus education level, marital status, and BMI; and model 4: model 3 plus hypertension, smoking status, alcohol use, physical activities, blood urea nitrogen, creatinine, and uric acid. Figure 3: Distribution of propensity score before and after matching. Figure 4: Relative risk for kidney stones based on sarcopenia index after PSM. The shaded areas represent upper and lower $95\%$ CIs. Adjustment factors are as same as which presented in extended model 4. Restricted cubic spline (RCS) plot of the association between sarcopenia index and kidney stones. The solid and dashed lines represent the odds ratios and $95\%$ confidence intervals. TABLE_PLACEHOLDER:Table 3 Figure 4 shows the dose–response relationships between sarcopenia index and kidney stones after propensity score matching. There was a no-linear association between sarcopenia index and the prevalence of kidney stones. Respectively comparing Figures 4, 2, Table 2, 3, it was proved that sarcopenia had more dramatic impact on the prevalence of kidney stones especially after PSM. With increasing of sarcopenia index, participants had lower prevalence of kidney stones. ## Discussion Kidney stones caused a series of damages to patient health and brought great social, economic, and healthy burden, due to its high incidence and recurrence rate. Therefore, it was essential for us to discover the risk factors of kidney stones, and the risk factors might help us to decrease the recurrence of stone disease and reduce burden of society and medicine. Our study discussed the relationship between sarcopenia and kidney stones. Firstly, we explored the association between sarcopenia and kidney stones. Then, we concluded that the odds of kidney stones decreased significantly with the increase of sarcopenia index which was negatively related to sarcopenia, and it was proved by the dose–response curve. However, we found that sarcopenia was not independent risk factor of kidney stones in participants (≥40 years). Therefore, PSM was performed to exclude the effects of other variables. After performing PSM, all participants showed that sarcopenia had the positively impact on prevalence of kidney stones, and the aOR in participants (age ≥40) in model 4 was 1.771 (1.138–2.757). In conclusion, sarcopenia was the independent hazard factor of kidney stones. Risk factors of kidney stones had increased around the world, and it might influence the more diagnosis of kidney stone. The formation of kidney stones was closely related to risk factors such as age, gender, dietary structure, environmental factors, genetics, abnormal urinary anatomy, and infection [3]. Especially, higher rates of obesity; diabetes; more intake of salt and animal protein; higher consumption of sugary beverages; global warming; less exercise; and sedentary behavior might contribute to higher incidence and prevalence of kidney stones (6–8). In recent years, studies had found a positive correlation between BMI, waist circumference, and the risk of kidney stones [9]. In general, as BMI increases, so does visceral obesity and hepatic steatosis, both of which were associated with low levels of urinary PH, and low urinary PH could lead to the development of uric acid stones [10]. Adipose tissue, as an endocrine organ, was a source of adipokines and inflammatory cytokines that can lead to insulin resistance, inflammatory, and enhanced oxidative stress states, and these conditions would lead to stone crystal formation [11, 12]. Sarcopenia was defined as a progressive and systemic skeletal muscle disease involving an accelerated loss of muscle mass and function, which was associated with increased adverse outcomes, including falls, functional decline, weakness, and death [13]. Age-related mechanisms that promoted sarcopenic episodes including inflammation, immune aging, anabolic resistance, and increased oxidative stress [14]. We found that the pathophysiological mechanisms of sarcopenia and kidney stones shared some of the same risk factors, such as obesity, insulin resistance, lack of physical activity, and chronic inflammation etc. In obese patients, ectopic deposition of fat in the liver and skeletal muscle caused an inflammatory response and insulin resistance, and fat tissue secreted adipokines and cytokines induced a decrease in skeletal muscle mass and function, leading to the development of sarcopenia (14–16). When insulin resistance occurred, the gluconeogenesis process in myocytes was promoted, resulting in decreased protein synthesis and increased catabolism, which lead to decreased muscle mass [17]. Then, for lack of physical activity, it lead to abdominal fat deposition, systemic inflammatory response, and insulin resistance, further reducing the body’s lipid and glucose oxidation [18]. Finally, some studies had found that patients with sarcopenia also have higher levels of inflammatory cytokines and inflammatory indicators that were involved in the activation of apoptosis, leading to a decrease in myofilament protein synthesis, which lead to sarcopenia [19]. It shown in our study that preliminary confirmation was published about the relationship between sarcopenia and the prevalence of kidney stones. Sarcopenia might contribute to the development and progression of kidney stones through increasing obesity, insulin resistance, inflammation, and decreasing recreational activities. In conclusion, we proved the associations between sarcopenia and the odds of kidney stones while controlling for potential variables, though several limitations still existed. First of all, we could not be completely convinced of the relationship between sarcopenia and kidney stones, because the study was a cross-sectional study. It needed more exploration to verify. Then, there was no consideration about relationship between position, type, and size of kidney stone and sarcopenia, and the basic characteristics of kidney stones was useful for prevention of kidney stones. Therefore, we should perform more studies to explore the potential mechanisms and to identify causality. ## Data availability statement Publicly available datasets were analyzed in this study. These data can be found here: National Health and Nutrition Examination Survey (NHANES). ## Author contributions BP, CX, and YW: conception and design. CT, HZ, and JN: administrative support. YW, YZ, and HS: provision of study materials or patients. TZ, YZ, and HZ: collection and assembly of data. CX, YZ, and HZ: data analysis and interpretation. All authors contributed to the article and approved the submitted version. ## Funding This study was financially supported by the Shanghai Association for Science and Technology Commission (Grant No. 21142203400), National Natural Science Foundation of China (Grant No. 81870517 and 32070646), and the National Key Research and Development Program of China (2021YFC2009300 and 2021YFC200930X). ## 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.2023.1123588/full#supplementary-material ## References 1. Pearle MS, Calhoun EA, Curhan GC. Urologic diseases in America project: urolithiasis. J Urol (2005) 173 848-57. DOI: 10.1097/01.ju.0000152082.14384.d7 2. Thongprayoon C, Krambeck AE, Rule AD. Determining the true burden of kidney stone disease. Nat Rev Nephrol (2020) 16 736-46. DOI: 10.1038/s41581-020-0320-7 3. Kittanamongkolchai W, Vaughan LE, Enders FT, Dhondup T, Mehta RA, Krambeck AE. The changing incidence and presentation of urinary stones over 3 decades. Mayo Clin Proc (2018) 93 291-9. DOI: 10.1016/j.mayocp.2017.11.018 4. Liguori I, Russo G, Aran L, Bulli G, Curcio F, Della-Morte D. Sarcopenia: assessment of disease burden and strategies to improve outcomes. Clin Interv Aging (2018) 13 913-27. DOI: 10.2147/CIA.S149232 5. Cruz-Jentoft AJ, Bahat G, Bauer J, Boirie Y, Bruyère O, Cederholm T. Sarcopenia: revised European consensus on definition and diagnosis. Age Ageing (2019) 48 601. DOI: 10.1093/ageing/afz046 6. Geiss LS, Wang J, Cheng YJ, Thompson TJ, Barker L, Li Y. Prevalence and incidence trends for diagnosed diabetes among adults aged 20 to 79 years, United States, 1980-2012. JAMA (2014) 312 1218-26. DOI: 10.1001/jama.2014.11494 7. Chooi YC, Ding C, Magkos F. The epidemiology of obesity. Metabolism (2019) 92 6-10. DOI: 10.1016/j.metabol.2018.09.005 8. Fakheri RJ, Goldfarb DS. Ambient temperature as a contributor to kidney stone formation: implications of global warming. Kidney Int (2011) 79 1178-85. DOI: 10.1038/ki.2011.76 9. Wong YV, Cook P, Somani BK. The association of metabolic syndrome and urolithiasis. Int J Endocrinol (2015) 2015 570674. DOI: 10.1155/2015/570674 10. Patel ND, Ward RD, Calle J, Remer EM, Monga M. Computerized tomography based diagnosis of visceral obesity and hepatic steatosis is associated with low urine pH. J Urol (2017) 198 1085-90. DOI: 10.1016/j.juro.2017.05.079 11. Khan SR. Reactive oxygen species, inflammation and calcium oxalate nephrolithiasis. Transl Androl Urol (2014) 3 256-76. DOI: 10.3978/j.issn.2223-4683.2014.06.04 12. Després J-P, Lemieux I, Bergeron J, Pibarot P, Mathieu P, Larose E. Abdominal obesity and the metabolic syndrome: contribution to global cardiometabolic risk. Arterioscler Thromb Vasc Biol (2008) 28 1039-49. DOI: 10.1161/ATVBAHA.107.159228 13. Cruz-Jentoft AJ, Sayer AA. Sarcopenia. Lancet (2019) 393 2636-46. DOI: 10.1016/S0140-6736(19)31138-9 14. Huang Y, Zeng M, Zhang L, Shi J, Yang Y, Liu F. Dietary inflammatory potential is associated with sarcopenia among chronic kidney disease population. Front Nutr (2022) 9 856726. DOI: 10.3389/fnut.2022.856726 15. De Fré CH, De Fré MA, Kwanten WJ, Op de Beeck BJ, Van Gaal LF, Francque SM. Sarcopenia in patients with non-alcoholic fatty liver disease: is it a clinically significant entity?. Obes Rev (2019) 20 353-63. DOI: 10.1111/obr.12776 16. Brunt EM, Wong VWS, Nobili V, Day CP, Sookoian S, Maher JJ. Nonalcoholic fatty liver disease. Nat Rev Dis Primers (2015) 1 15080. DOI: 10.1038/nrdp.2015.80 17. Lee Y-H, Kim SU, Song K, Park JY, Kim DY, Ahn SH. Sarcopenia is associated with significant liver fibrosis independently of obesity and insulin resistance in nonalcoholic fatty liver disease: Nationwide surveys (KNHANES 2008-2011). Hepatology (2016) 63 776-86. DOI: 10.1002/hep.28376 18. Biolo G, Cederholm T, Muscaritoli M. Muscle contractile and metabolic dysfunction is a common feature of sarcopenia of aging and chronic diseases: from sarcopenic obesity to cachexia. Clin Nutr (2014) 33 737-48. DOI: 10.1016/j.clnu.2014.03.007 19. Koo BK, Kim D, Joo SK, Kim JH, Chang MS, Kim BG. Sarcopenia is an independent risk factor for non-alcoholic steatohepatitis and significant fibrosis. J Hepatol (2017) 66 123-31. DOI: 10.1016/j.jhep.2016.08.019
--- title: 'Bi-directional causal effect between vitamin B12 and non-alcoholic fatty liver disease: Inferring from large population data' authors: - Liwan Fu - Yuquan Wang - Yue-Qing Hu journal: Frontiers in Nutrition year: 2023 pmcid: PMC10025356 doi: 10.3389/fnut.2023.1015046 license: CC BY 4.0 --- # Bi-directional causal effect between vitamin B12 and non-alcoholic fatty liver disease: Inferring from large population data ## Abstract ### Objectives Many observational studies evaluate the association between vitamin B12 and non-alcoholic fatty liver disease (NAFLD). However, the causality of this association remains uncertain, especially in European populations. We conducted a bidirectional Mendelian randomization study to explore the association between vitamin B12 and NAFLD. ### Methods Two-sample Mendelian randomization study was conducted. Summary statistics for vitamin B12 were acquired from a genome-wide association studies (GWAS) meta-analysis including 45,576 subjects. Summary-level data for NAFLD was obtained from a GWAS meta-analysis of 8,434 cases and 770,180 non-cases and another GWAS meta-analysis of 1,483 cases and 17,781 non-cases. Summary-level data for 4 enzymes including alkaline phosphatase (ALP), alanine aminotransferase (ALT), aspartate aminotransferase (AST), and gamma glutamyltransferase (GGT), was available from the UK Biobank. Inverse variance weighting (as main analysis), weighted median estimate, robust adjusted profile score, MR-Egger, and MR-PRESSO (sensitivity analyses) were performed to calculate causal estimates. ### Results Genetically predicted higher vitamin B12 concentrations were consistently associated with an increased NAFLD in two sources. The combined odds ratio (OR) of NAFLD was 1.30 ($95\%$ confidence interval (CI), 1.13 to 1.48; $p \leq 0.001$) per SD-increase in vitamin B12 concentrations. Genetic liability to NAFLD was also positively associated with vitamin B12 concentrations (Beta 0.08, $95\%$CI, 0.01 to 0.16; $$p \leq 0.034$$). Sensitivity analyses also revealed consistent results. Genetically predicted vitamin B12 concentrations showed no significant association with liver enzymes. ### Conclusion The present study indicates that increased serum vitamin B12 concentrations may play a role in NAFLD risk. NAFLD also has a causal impact on elevated vitamin B12 concentrations in the circulation. Notably, vitamin B12 concentrations imply the levels of vitamin B12 in the circulation, and higher intake of vitamin B12 may not directly lead to higher levels of serum vitamin B12, instead the higher levels of vitamin B12 in the circulation may be caused by the dysregulation of the metabolism of this vitamin in this study. There exist bidirectional causal effects between serum vitamin B12 concentrations and risk of NAFLD in European individuals. ## Introduction Non-alcoholic fatty liver disease (NAFLD), as the leading liver disease worldwide, influences approximately $25\%$ of the world population [1]. Due to the sedentary lifestyle, western diet, and obesity epidemic, the prevalence of NAFLD keeps increasing [2]. NAFLD, mostly encompassing non-alcoholic steatohepatitis (NASH), isolated hepatic steatosis, and cirrhosis [3], is considered as a feature of metabolic syndrome in the liver [4]. As NAFLD is always symptomless and difficult to be observed, it is hard to predict the progression of NAFLD [5]. Thus, identifying potential biomarker is needed to predict the emergence and development of NAFLD. Vitamin B12, mainly presenting two forms in humans: 5′-deoxyadenosylcobalamine and methyl cobalamin, was reported to be correlated with hepatitis and cirrhosis [6]. In addition, vitamin B12 was served as a cofactor for methyl malonyl CoA mutase, which managed the rate of long-chain fatty Acyl-CoA enter into mitochondria and influences lipid metabolic pathways [7]. The liver was served as a storage site for vitamin B12. An increased serum content of vitamin B12 found in acute and chronic liver diseases has been attributed to the release of the vitamin from the liver owing to hepatic necrosis and/or to an increased capacity of the serum to bind vitamin B12 as a result of abnormalities in the serum proteins. Moreover, several studies, in animal models, investigated the impact of vitamin B12 change in lipid metabolism [8]. The liver is an important organ for lipid metabolism. Thus, it is reasonable to suppose that serum vitamin B12 concentrations may change in the occurrence of liver injury. Importantly, vitamin B12 is a water soluble nutrient which plays an important role in human health. Because of its water solubility, its intake is generally not excessively restricted. Dietary intake of vitamin B12 is indispensable to the maintenance of human health and deficiencies can result in severe health consequences. Notably, it does not necessarily imply that higher levels of serum vitamin B12 is due to a higher intake, instead the dysregulation of the metabolism of this vitamin may be the driver of higher levels of vitamin B12 in the circulation. Vitamin B12 is cofactor for enzymes in one-carbon metabolism, which plays a central role in the generation of methyl donors in the form of S-adenosylmethionine (SAM), the sole methyl donor used by DNA, RNA, histone, and protein methyltransferases [9]. Therefore, the intake of vitamin B12 should be given great attention, especially its impact on NAFLD. Moreover, some study designs, including cross-sectional and case–control studies, have evaluated the association between vitamin B12 and NAFLD (10–22). Several studies implicated positive association between vitamin B12 and NAFLD [12, 16, 20, 21], whereas others indicated inverse association [10, 13, 19] or no association [11, 14, 15, 17, 18, 22]. Notably, a latest systematic review and meta-analysis comprising 361 NAFLD subjects and 510 controls demonstrated no association of vitamin B12 concentrations with risk of NAFLD [23]. Nevertheless, a variety of limitations, including a great heterogeneity in different studies, selection bias, and confounding factors, were not fully considered and described in its interpretation [23]. Importantly, potential reverse causality and residual confounding were unable to be explained by these observational studies, so aforementioned controversial findings deriving from observational studies could not provide the causal inference in the association between vitamin B12 concentrations and NAFLD. Leveraging genetic variants as instrumental variables for an exposure (e.g., vitamin B12), Mendelian randomization (MR) was able to strengthen causal inference of an exposure-outcome association via diminishing reverse causality and residual confounding [24]. Additionally, the sample size of the population, and the relationship between vitamin B12 concentrations and NAFLD differing between ethnicities need to be further considered [20, 22, 23]. Therefore, we performed a bidirectional 2-sample MR study on the basis of large populations to investigate this relationship in the European population. Concurrently, we also carried out a 2-sample MR study to estimate the association of vitamin B12 concentrations with NAFLD-related liver enzymes. ## Study design We performed this MR study based on summary statistics of genome-wide association analyses on serum vitamin B12 levels, NAFLD, and liver enzymes from large-scale genome-wide association studies (GWAS) (25–28), which included different studies and consortia encompassing the UK Biobank study, the Estonian Biobank, the FinnGen consortium, and so on (Supplementary Table S1). We first calculated genetic correlations of vitamin B12 concentrations with NAFLD and liver enzymes. Second, we conducted a forward MR analysis to evaluate the causal effect of genetic prediction of higher vitamin B12 concentrations on NAFLD risk and concentrations of liver enzymes. Considering the possibility that NAFLD might impact vitamin B12 concentrations as the liver is regarded as a storage site for vitamin B12 [6], we performed a reverse MR analysis to investigate the association between genetic liability to NAFLD and vitamin B12 concentrations. The detailed flow chart of vitamin B12 as exposure in this MR study is displayed in Figure 1. All the studies comprised in cited GWAS were approved by a relevant review board. No necessary for ethical permit in this MR analysis because of summary-level data. Figure 1: Flow chart of Vitamin B12 as exposure in this Mendelian randomization. ALP, alkaline phosphatase; ALT, alanine aminotransferase; AST, aspartate aminotransferase; GGT, gamma glutamyltransferase; Vitamin B12, serum Vitamin B12; IVW, inverse variance weighted; MR-PRESSO, Mendelian randomization pleiotropy residual sum and outlier; NAFLD, non-alcoholic fatty liver disease. ## Data sources of vitamin B12 We identified 18 single nucleotide polymorphism (SNPs) associated with serum vitamin B12 concentrations from a GWASs meta-analysis including 45,576 individuals of European descent [25] at the genome-wide significant level (p value less than 5E-08). Based on 1,000 Genomes European reference panel, we used the PLINK clumping technique to estimate linkage disequilibrium among these 18 SNPs. Consequently, 14 independent SNPs without linkage disequilibrium (r2 < 0.01) were chosen as alternative instruments for vitamin B12. Covariates, such as sex, age, and principal components, were adjusted in the corresponding association estimates (Supplementary Table S1). These instruments explain about $6.0\%$ of phenotypic variance for vitamin B12 and have been employed in previous MR studies [29, 30]. We used the internet resource (PhenoScanner V2) [31] to test whether these 14 instruments are correlated with other phenotypes, and noticed that 3 SNPs (rs1131603, rs2336573, and rs602662) were associated with other phenotypes (Supplementary Table S2, the search was conducted in April 2022). These 3 SNPs probably exerted pleiotropic effects and were removed, plus 3 other SNPs that do not match with NAFLD (Supplementary Table S3), and a total of 8 independent SNPs were selected as instruments for vitamin B12 eventually. As the exposure, vitamin B12 concentrations were transformed to one standard deviation (SD) unit (equivalent to standardization). For testing the strength of genetic instruments, we calculated F-statistics for vitamin B12 ($F = 242.4$), which was greater than 10, suggesting the support for the strength of genetic instruments. For the reverse MR study, summary-level data for vitamin B12 concentrations was obtained from a study from Vanderbilt University Medical Center [27] because of scanty SNPs associated with NAFLD were selected from this GWAS meta-analysis [25] when NAFLD was acted as an exposure. ## Data sources of non-alcoholic fatty liver disease Summary data for the associations of vitamin B12 associated SNPs with NAFLD was obtained from two large meta-analyses of GWAS. One GWAS was from Ghodsian et al. [ 28], which encompassed 8,434 cases and 770,180 controls of European descent. This GWAS meta-analysis included four cohorts: the UK Biobank, FinnGen, the Estonian Biobank and the Electronic Medical Records and Genomics (eMERGE) [32]. The eMERGE defined NAFLD cases by Electronic health record (EHR) codes (ICD9: 571.5, ICD9: 571.8, ICD9: 571.9, ICD10: K75.81, ICD10: K76.0 and ICD10: K76.9). NAFLD in the UK Biobank, FinnGen and the Estonian Biobank were defined by International Classification of Disease code K76.0. In these four cohorts, the associations were adjusted for age, gender, 10-main ancestry-based principal components and genotyping batch. The other GWAS from Anstee et al. [ 26] included 1,483 NAFLD cases and 17,781 controls of European descent. NAFLD cases in this GWAS were defined on the basis of abnormal biochemical estimates and/or an ultrasonographically tested bright liver, along with diagnostics of the metabolic syndrome; or observing abnormal biochemical tests and macroscopic appearances of a steatotic liver at the time of bariatric surgery [26]. Specific stages of NAFLD and histology are available in GWAS from Anstee et al., and the first 5 principal components were adjusted for the associations in Anstee et al., GWAS. For the reverse MR analysis, 6 SNPs associated with NAFLD at the genome-wide significant level ($p \leq 5$E-08) were selected as instruments from the Ghodsian et al., GWAS [28]. As rs58542926 was strongly correlated with rs10401969, we removed s58542926 and 5 independent SNPs were left as instruments in the reverse MR analysis eventually (Supplementary Table S4). Additionally, we performed a sensitivity analysis of the reverse MR by utilizing SNPs as instruments from the Anstee et al., GWAS at the genome-wide significant level. As a result, 6 independent SNPs (r2 < 0.01) were chosen as instruments in the sensitivity analysis of the reverse MR (Supplementary Table S5). ## Data sources of liver enzymes We acquired four liver enzymes, including alkaline phosphatase (ALP), alanine aminotransferase (ALT), aspartate aminotransferase (AST), and gamma glutamyltransferase (GGT), which were possibly correlated with NAFLD [28]. Summary data for the associations between vitamin B12 associated SNPs and these liver enzymes was obtained from the outcomes of the second wave in the UK Biobank through the Neale lab (Supplementary Table S6). ## Statistical analysis Overlapping samples between two datasets could result in bias for the estimated causal effects. Thus, we utilized linkage disequilibrium score regression (LDSC) to estimate sample overlap with LD hub1 [33]. Inverse variance weighting (IVW) with multiplicative random was used as the main analysis in this study [34]. Then, we combined estimates from Ghodsian et al., GWAS and Anstee et al., GWAS via the fixed-effects meta-analysis method. Four MR approaches encompassing weighted median estimate [35], robust adjusted profile score [36], MR-Egger [37], and MR-PRESSO [38], were conducted as sensitivity methods for evaluating the consistency of results or correcting for pleiotropy. The weighted median estimate could provide an unbiased test when $50\%$ SNPs are invalid instruments [35]. Robust adjusted profile score correctly considered the measurement error in the selected instruments to acquire precise estimates with smaller bias [36]. MR-Egger offered an assessment for horizontal pleiotropy via the p value of its intercept, and a test was obtained after the pleiotropic effects were adjusted. However, wider confidence intervals (CIs) were got due to a loss of statistical power [37]. MR-PRESSO is another method evaluating biases caused by pleiotropy (through the global test). It gives a corrected estimate by removing the outliers, and also provides a distortion test, which estimates whether the results with or without outliers were different [38]. We performed Cochran’s Q statistic to test the magnitude of heterogeneity [39] for the included SNPs. All analyses were conducted by R Version 4.1.0 utilizing R packages (“TwoSampleMR”) [40] and (“MRPRESSO”) [38]. ## Evaluation for sample overlap LDSC was conducted to evaluate sample overlap between exposure GWAS and outcome GWAS. Results showed approximately zero intercept of genetic covariance in pairs of exposure-outcome GWAS ($p \leq 0.1$ through z-test in all pairs, data not displayed), indicating no sample overlap in pairs of two GWAS datasets in this study. ## Forward Mendelian randomization analysis Firstly, we used Ghodsian et al., GWAS data to perform the forward MR analysis. As a result, genetic prediction of higher vitamin B12 concentrations was associated with an elevated risk of NAFLD (odds ratio (OR) per 1-SD increase, 1.27; $95\%$ confidence interval (CI), 1.09 to 1.45; $$p \leq 0.001$$) by the IVW-multiplicative random effect model. Then, the significance of the association was replicated in Anstee et al., GWAS (OR = 1.58; $95\%$CI, 1.04 to 2.19; $$p \leq 0.022$$) (Figure 2). Secondly, we combined tests from these two GWAS data sources with the fixed-effect model in the meta-analysis, and the combined OR of NAFLD was 1.30 ($95\%$CI, 1.13 to 1.48; $p \leq 0.001$) for a 1-SD elevation in genetic prediction of vitamin B12 concentrations (Figure 2). All the sensitivity analyses also showed significant results (Table 1). We found no evidence of heterogeneity for the included SNPs (Both Cochrane’s Q < 5 from Ghodsian et al., GWAS and Anstee et al., GWAS data source), and MR-Egger showed no horizontal pleiotropy (Table 1). Additionally, no outlier was identified during the MR-PRESSO analysis. Figure 2: Association between genetically predicted Vitamin B12 and NAFLD. Vitamin B12, serum Vitamin B12; GWAS, genome-wide association studies; OR, odds ratio; NAFLD, non-alcoholic fatty liver disease. TABLE_PLACEHOLDER:Table 1 Based on the UK Biobank data, genetic prediction of vitamin B12 concentrations showed no association with liver enzymes in the main analysis (IVW-multiplicative random effects). Other MR approaches revealed similar estimates in the sensitivity analyses (Supplementary Figure S1), except for negative association with GGT through robust adjusted profile score method (Beta, −0.69; $95\%$CI, −1.26 to-0.13; $$p \leq 0.016$$). Moderate heterogeneity was observed in the main analysis of AST, while no pleiotropy was detected in the MR-Egger analysis. ## Reverse Mendelian randomization analysis In the main analysis, liability to NAFLD revealed positive association with vitamin B12, and the effect size of vitamin B12 concentrations in 1-SD change was 0.08 ($95\%$CI, 0.01 to 0.16; $$p \leq 0.034$$) for 1-unit increase in the log-transformed OR of NAFLD (Figure 3). Other MR approaches persistently showed the significant association (Figure 3). No horizontal pleiotropy was detected in the MR-Egger analysis (P for the intercept was 0.28), and no outlier was identified during the MR-PRESSO analysis. We further employed 6 SNPs detected from Anstee et al., GWAS data for the sensitivity analysis and observed that liability to NAFLD also had positive association with vitamin B12 concentrations (Beta, 0.04; $95\%$CI, 0.002 to 0.07; $$p \leq 0.041$$) by the main analysis and other MR approaches (Figure 3). Figure 3: Association of genetic liability to NAFLD with Vitamin B12. Vitamin B12, serum Vitamin B12; IVW, inverse variance weighted; MR-PRESSO, Mendelian randomization pleiotropy residual sum and outlier. ## Discussion The present study elucidated positive genetic associations between serum vitamin B12 and NAFLD risk. Forward MR analysis showed that genetic prediction of increased serum vitamin B12 concentrations were robustly associated with an elevated risk of NAFLD. Moreover, the reverse MR analysis was able to give support for an effect of NAFLD on serum vitamin B12 concentrations. Our findings indicated that higher serum vitamin B12 was a strong casual factor of increased risk of NAFLD, which in turn would further increase serum vitamin B12. Previous studies on the relationship between vitamin B12 concentrations and NAFLD were controversial. A meta-analysis encompassing 8 cross-sectional and case–control studies published recently found that NAFLD cases had same vitamin B12 concentrations compared with those without NAFLD [23]. No association was implicated in earlier studies (11, 14–18). In contrast, a study involved in 614 Brazilian cases reported that higher serum concentrations of vitamin B12 were positively associated with the severity of fibrosis and steatosis [20]. Vitamin B12 acts as a coenzyme for a crucial methyl transfer reaction, and the recommended dietary allowance for adult reaches 2.4 μg of vitamin B12 per day [41]. Dietary intakes of vitamin B12 were also found increase in patients with NAFLD compared with controls in 101 Canadians [42] and 120 adult Jordanians [43], which were consistent with our main result that elevated vitamin B12 was causally associated with the increased risk of NAFLD. Nevertheless, both dietary intakes of vitamin B12 and serum concentrations of vitamin B12 were unable to observe significant associations between NAFLD patients and controls in 317 Iranians [44] and 54 participants in Greece [15], respectively. Otherwise, serum concentrations of vitamin B12 were found in a lower level in NAFLD cases compared with those of the controls in 75 Turks [13], and decreased vitamin B12 concentrations were significantly associated with an increased fibrosis grade, as well as non-alcoholic steatohepatitis in 83 cases in Israel [45]. However, previous publications on vitamin B12 and NAFLD usually contained a small number of subjects and did not consider covariates in univariate analysis. Previously, we have established genetic statistics to detect genetic variation in complex diseases and used MR approaches to evaluate the causal relationships between complex diseases, including the casual association of Hcy with NAFLD (46–54). Recently, a national population-based survey from NHANES showed positive associations of vitamin B12 with liver steatosis and fibrosis linearly [21], which are consistent with our findings to some degree. Additionally, different races, and distinct diagnostic criteria may be reasons for conflicting results in different studies. In contrast, our study on the basis of several large populations of European descent displayed a significant positive association of genetically predicted vitamin B12 concentrations with NAFLD risk. Inadequate power or unobserved confounding effects would contribute to the lack of association. Compared to aforementioned cross-sectional and case–control studies, we performed several large populations encompassing over approximately 10,000 NAFLD patients and over 780,000 controls, conducted state-of-art approaches of causal analysis (various MR methods) to minish reverse causality and unobserved confounding impact, and eventually, offered sufficient power to evaluate the causal effect of vitamin B12 concentrations on NAFLD risk. Interestingly, our study was able to establish a significant impact of genetic liability to NAFLD on increased serum concentrations of vitamin B12. Similarly, the association of genetic liability to NAFLD with serum vitamin B12 concentrations was consistent in our sensitivity analyses. Therefore, bi-directional causal effects were observed, and instances of true bi-directional pathway might exist, and in other words, a proposed positive effect of vitamin B12 on NAFLD risk, whereas NAFLD also exerts a positive effect on vitamin B12, possibly as part of a positive feedback loop. For patients at high risk of NAFLD, this study might indicate that the intake of vitamin B12 may be restricted because it might elevate the serum vitamin B12 concentrations, and then increase the risk of NAFLD. However, the dysregulation in the pathways that regulate B12 metabolism may influence the serum level of vitamin B12, and the dysregulation of the metabolism of this vitamin may be the driver of higher levels of vitamin B12 in the circulation. Therefore, it does not necessarily imply that higher levels of serum vitamin B12 is due to a higher intake. A recent randomized controlled trial indicated that vitamin B12 supplementation significantly decreased serum levels of homocysteine compared to placebo, while no significant difference was revealed in between-group comparisons for fasting blood glucose, malondialdehyde, and liver steatosis [55]. Because this study was unable to support our findings, further studies with different doses will reveal additional evidence. In addition, new trials with dose-responsive effect of serum vitamin B12 on NAFLD still need to be explored for this finding. This MR analysis showed no association between genetic prediction of vitamin B12 concentrations and liver enzymes, which indicated that vitamin B12 was possibly unable to affect NAFLD risk through the pathways involved in liver enzymes. Extensive research has manifested the importance of vitamin B12 status in adjusting one-carbon metabolism [56, 57]. As is known to all, the transsulfuration pathway is a part of one-carbon metabolism, and it acts as an important role in many progress of chronic diseases and has been correlated with inflammation, oxidative stress, ER stress, insulin resistance, portal hypertension, and steatosis [57]. The relationship between the transsulfuration pathway and oxidative stress has been believed to be regulated through the adjustment of the antioxidant glutathione production. Experimental and animal studies demonstrated that the downstream products of vitamin B12 metabolism are increased and oxidative stress, steatosis, and fibrosis develop in the liver in cystathionine β-synthase-deficient mice [57]. Furthermore, the lack of cystathionine β-synthase seemingly upregulates the expression of genes with regard to ER stress, hepatic lipid homeostasis, as well as genes linked to hepatic steatosis [57], whereas knockout of cystathionine γ-lyase results in decreased hepatic lipolysis [58]. The aforementioned evidence may, to some extent, explain the possible mechanism by which vitamin B12 causes the risk of NAFLD. Some strengths appear in the present study. The mentionable merit is MR design, which enhanced causal inference through decreasing reverse causation and residual confounding. Employing two large population-based GWAS meta for the association between genetic prediction of vitamin B12 concentrations and NAFLD, significantly improved the statistical power and solidified our results. Additionally, sensitivity analyses showed robust association and no unbalanced pleiotropy. Our findings were restricted to European populations and principal component analysis was adopted in the performed GWAS analyses. Thus, the population structure was possibly reduced. However, the population confinement blocked the generalizability of the present findings to other populations. Limitations should also be mentioned when interpreting our results. As the lack of dose-responsive effect of vitamin B12 on NAFLD in this study, it is hard to apply the results of the present study to clinical practice without recommendation of a beneficial range. PhenoScanner V2 [31] was utilized to screen out the genetic instruments employed for vitamin B12 concentrations and associated with some phenotypes at the genome-wide significant level (Supplementary Table S2) for the horizontal pleiotropy concern. Even so, there probably still exist potential unobserved pleiotropic SNPs. It is worth noting that the definition of NAFLD patients was distinct between the Ghodsian et al., GWAS and Anstee et al., GWAS, which perhaps lead to heterogeneity in meta-analysis of associations in despite of a small degree. As previously mentioned, the associations of vitamin B12 concentrations with NAFLD may differ in populations of different ancestries. More studies focused on individuals of non-European descent are needed in the future. Last but not least, due to the lack of sex-stratified data in this study, whether the association of genetically predicted vitamin B12 concentrations with NAFLD risk differs between men and women needs further studies. ## Conclusion In conclusion, the present study unraveled positive relationship between vitamin B12 concentrations and NAFLD, and elucidated bidirectional causal effects between vitamin B12 concentrations and risk of NAFLD in European individuals. Significantly, vitamin B12 concentrations imply the levels of vitamin B12 in the circulation, and higher levels of serum vitamin B12 may not be directly caused by a higher intake, instead the dysregulation of the metabolism of this vitamin may contribute to higher levels of vitamin B12 in the circulation. Our findings supply clinical implications, as they implicate that serum vitamin B12 may play a part in NAFLD risk. Similarly, NAFLD also has a causal impact on elevated serum vitamin B12 concentrations in the circulation. ## 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 LF and Y-QH: study concept and design and drafting of the manuscript. LF, Y-QH, and YW: acquisition of data and critical revision of the manuscript for important intellectual content. LF: analysis and interpretation of data. All authors have read and approved the final version of manuscript. ## Funding This study was supported by grants to LF and Y-QH from the National Natural Science Foundation of China (grants nos. 82204063, 11971117, and 11571082). ## 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.2023.1015046/full#supplementary-material ## References 1. Younossi Z, Tacke F, Arrese M, Chander Sharma B, Mostafa I, Bugianesi E. Global perspectives on non-alcoholic fatty liver disease and non-alcoholic steatohepatitis. Hepatology (2019) 69 2672-82. DOI: 10.1002/hep.30251 2. Younossi ZM, Koenig AB, Abdelatif D, Fazel Y, Henry L, Wymer M. Global epidemiology of non-alcoholic fatty liver disease-meta-analytic assessment of prevalence, incidence, and outcomes. Hepatology (2016) 64 73-84. DOI: 10.1002/hep.28431 3. Powell EE, Wong VW, Rinella M. Non-alcoholic fatty liver disease. Lancet (2021) 397 2212-24. DOI: 10.1016/s0140-6736(20)32511-3 4. Yki-Järvinen H. Non-alcoholic fatty liver disease as a cause and a consequence of metabolic syndrome. Lancet Diabetes Endocrinol (2014) 2 901-10. DOI: 10.1016/s2213-8587(14)70032-4 5. Reid BM, Sanyal AJ. Evaluation and management of non-alcoholic steatohepatitis. Eur J Gastroenterol Hepatol (2004) 16 1117-22. DOI: 10.1097/00042737-200411000-00005 6. Raza S, Tewari A, Rajak S, Sinha RA. Vitamins and non-alcoholic fatty liver disease: a molecular insight. Liver Res (2021) 5 62-71. DOI: 10.1016/j.livres.2021.03.004 7. O'Leary F, Samman S. Vitamin B12 in health and disease. Nutrients (2010) 2 299-316. DOI: 10.3390/nu2030299 8. Deshmukh U, Katre P, Yajnik CS. Influence of maternal vitamin B12 and folate on growth and insulin resistance in the offspring. Nestle Nutr Inst Workshop Ser (2013) 74 145-54. DOI: 10.1159/000348463 9. Lyon P, Strippoli V, Fang B, Cimmino L. B vitamins and one-carbon metabolism: implications in human health and disease. Nutrients (2020) 12 2867. DOI: 10.3390/nu12092867 10. Gulsen M, Yesilova Z, Bagci S, Uygun A, Ozcan A, Ercin CN. Elevated plasma homocysteine concentrations as a predictor of steatohepatitis in patients with non-alcoholic fatty liver disease. J Gastroenterol Hepatol (2005) 20 1448-55. DOI: 10.1111/j.1440-1746.2005.03891.x 11. Hirsch S, Poniachick J, Avendaño M, Csendes A, Burdiles P, Smok G. Serum folate and homocysteine levels in obese females with non-alcoholic fatty liver. Nutrition (2005) 21 137-41. DOI: 10.1016/j.nut.2004.03.022 12. Sazci A, Akpinar G, Aygun C, Ergul E, Senturk O, Hulagu S. Association of apolipoprotein E polymorphisms in patients with non-alcoholic steatohepatitis. Dig Dis Sci (2008) 53 3218-24. DOI: 10.1007/s10620-008-0271-5 13. Koplay M, Gulcan E, Ozkan F. Association between serum vitamin B12 levels and the degree of steatosis in patients with non-alcoholic fatty liver disease. J Investig Med (2011) 59 1137-40. DOI: 10.2310/JIM.0b013e31822a29f5 14. Polyzos SA, Kountouras J, Patsiaoura K, Katsiki E, Zafeiriadou E, Deretzi G. Serum homocysteine levels in patients with non-alcoholic fatty liver disease. Ann Hepatol (2012) 11 68-76. PMID: 22166563 15. Polyzos SA, Kountouras J, Patsiaoura K, Katsiki E, Zafeiriadou E, Zavos C. Serum vitamin B12 and folate levels in patients with non-alcoholic fatty liver disease. Int J Food Sci Nutr (2012) 63 659-66. DOI: 10.3109/09637486.2011.649249 16. de Carvalho SC, Muniz MT, Siqueira MD, Siqueira ER, Gomes AV, Silva KA. Plasmatic higher levels of homocysteine in non-alcoholic fatty liver disease (NAFLD). Nutr J (2013) 12 37. DOI: 10.1186/1475-2891-12-37 17. Leach NV, Dronca E, Vesa SC, Sampelean DP, Craciun EC, Lupsor M. Serum homocysteine levels, oxidative stress and cardiovascular risk in non-alcoholic steatohepatitis. Eur J Intern Med (2014) 25 762-7. DOI: 10.1016/j.ejim.2014.09.007 18. Kasapoglu B, Turkay C, Yalcin KS, Kosar A, Bozkurt A. MTHFR 677C/T and 1298A/C mutations and non-alcoholic fatty liver disease. Clin Med (2015) 15 248-51. DOI: 10.7861/clinmedicine.15-3-248 19. Moretti R, Caruso P, Gazzin S. Non-alcoholic fatty liver disease and neurological defects. Ann Hepatol (2019) 18 563-70. DOI: 10.1016/j.aohep.2019.04.007 20. Bertol FS, Araujo B, Jorge BB, Rinaldi N, De Carli LA, Tovo CV. Role of micronutrients in staging of non-alcoholic fatty liver disease: a retrospective cross-sectional study. World J Gastrointest Surg (2020) 12 269-76. DOI: 10.4240/wjgs.v12.i6.269 21. Liu X, Shen H, Chen M, Shao J. Clinical relevance of vitamins and carotenoids with liver steatosis and fibrosis detected by transient Elastography in adults. Front Nutr (2021) 8 760985. DOI: 10.3389/fnut.2021.760985 22. Li L, Huang Q, Yang L, Zhang R, Gao L, Han X. The association between non-alcoholic fatty liver disease (NAFLD) and advanced fibrosis with serological vitamin B12 markers: results from the NHANES 1999& ndash; 2004. Nutrients (2022) 14 1224. DOI: 10.3390/nu14061224 23. Costa D, Guahnon M, Seganfredo F, Pinto L, Tovo C, Fernandes S. Vitamin B12 and homocysteine levels in patients with NAFLD: a systematic review and metanalysis. Arq Gastroenterol (2021) 58 234-9. DOI: 10.1590/s0004-2803.202100000-42 24. Burgess S, Thompson SG. Mendelian randomization: Methods for using genetic variants in causal estimation (2015) 25. Grarup N, Sulem P, Sandholt CH, Thorleifsson G, Ahluwalia TS, Steinthorsdottir V. Genetic architecture of vitamin B12 and folate levels uncovered applying deeply sequenced large datasets. PLoS Genet (2013) 9 e1003530. DOI: 10.1371/journal.pgen.1003530 26. Anstee QM, Darlay R, Cockell S, Meroni M, Govaere O, Tiniakos D. Genome-wide association study of non-alcoholic fatty liver and steatohepatitis in a histologically characterised cohort. J Hepatol (2020) 73 505-15. DOI: 10.1016/j.jhep.2020.04.003 27. Dennis JK, Sealock JM, Straub P, Lee YH, Hucks D, Actkins K. Clinical laboratory test-wide association scan of polygenic scores identifies biomarkers of complex disease. Genome Med (2021) 13 6. DOI: 10.1186/s13073-020-00820-8 28. Ghodsian N, Abner E, Emdin CA, Gobeil É, Taba N, Haas ME. Electronic health record-based genome-wide meta-analysis provides insights on the genetic architecture of non-alcoholic fatty liver disease. Cell Rep Med (2021) 2 100437. DOI: 10.1016/j.xcrm.2021.100437 29. Yuan S, Carter P, Vithayathil M, Kar S, Mason AM, Burgess S. Genetically predicted circulating B vitamins in relation to digestive system cancers. Br J Cancer (2021) 124 1997-2003. DOI: 10.1038/s41416-021-01383-0 30. Yuan S, Mason AM, Carter P, Burgess S, Larsson SC. Homocysteine, B vitamins, and cardiovascular disease: a Mendelian randomization study. BMC Med (2021) 19 97. DOI: 10.1186/s12916-021-01977-8 31. Kamat MA, Blackshaw JA, Young R, Surendran P, Burgess S, Danesh J. Pheno scanner V2: an expanded tool for searching human genotype-phenotype associations. Bioinformatics (2019) 35 4851-3. DOI: 10.1093/bioinformatics/btz469 32. Jongstra-Bilen J, Haidari M, Zhu SN, Chen M, Guha D, Cybulsky MI. Low-grade chronic inflammation in regions of the normal mouse arterial intima predisposed to atherosclerosis. J Exp Med (2006) 203 2073-83. DOI: 10.1084/jem.20060245 33. Zheng J, Erzurumluoglu AM, Elsworth BL, Kemp JP, Howe L, Haycock PC. LD hub: a centralized database and web interface to perform LD score regression that maximizes the potential of summary level GWAS data for SNP heritability and genetic correlation analysis. Bioinformatics (2017) 33 272-9. DOI: 10.1093/bioinformatics/btw613 34. Bowden J, Del Greco MF, Minelli C, Davey Smith G, Sheehan N, Thompson J. A framework for the investigation of pleiotropy in two-sample summary data Mendelian randomization. Stat Med (2017) 36 1783-802. DOI: 10.1002/sim.7221 35. Bowden J, Davey Smith G, Haycock PC, Burgess S. Consistent estimation in Mendelian randomization with some invalid instruments using a weighted median estimator. Genet Epidemiol (2016) 40 304-14. DOI: 10.1002/gepi.21965 36. Zhao Q, Wang J, Hemani G, Bowden J, Small DS. Statistical inference in two-sample summary-data Mendelian randomization using robust adjusted profile score. Ann Stat (2020) 48 28. DOI: 10.1214/19-AOS1866 37. Bowden J, Davey Smith G, Burgess S. Mendelian randomization with invalid instruments: effect estimation and bias detection through egger regression. Int J Epidemiol (2015) 44 512-25. DOI: 10.1093/ije/dyv080 38. Verbanck M, Chen C-Y, Neale B, Do R. Detection of widespread horizontal pleiotropy in causal relationships inferred from Mendelian randomization between complex traits and diseases. Nat Genet (2018) 50 693-8. DOI: 10.1038/s41588-018-0099-7 39. Higgins JP, Thompson SG. Quantifying heterogeneity in a meta-analysis. Stat Med (2002) 21 1539-58. DOI: 10.1002/sim.1186 40. Hemani G, Zheng J, Elsworth B, Wade K, Haberland V, Baird D. The MR-base platform supports systematic causal inference across the human phenome. elife (2018) 7 e34408. DOI: 10.7554/elife.34408 41. Thiamin R. Dietary reference intakes for Thiamin, riboflavin, niacin, vitamin B6, folate, vitamin B12, pantothenic acid, biotin, and choline (1998) 42. Da Silva HE, Arendt BM, Noureldin SA, Therapondos G, Guindi M, Allard JP. A cross-sectional study assessing dietary intake and physical activity in Canadian patients with non-alcoholic fatty liver disease vs healthy controls. J Acad Nutr Diet (2014) 114 1181-94. DOI: 10.1016/j.jand.2014.01.009 43. Tayyem RF, Al-Dayyat HM, Rayyan YM. Relationship between lifestyle factors and nutritional status and non-alcoholic fatty liver disease among a group of adult Jordanians. Arab J Gastroenterol (2019) 20 44-9. DOI: 10.1016/j.ajg.2019.01.008 44. Zolfaghari H, Askari G, Siassi F, Feizi A, Sotoudeh G. Intake of nutrients, fiber, and sugar in patients with non-alcoholic fatty liver disease in comparison to healthy individuals. Int J Prev Med (2016) 7 98. DOI: 10.4103/2008-7802.188083 45. Mahamid M, Mahroum N, Bragazzi NL, Shalaata K, Yavne Y, Adawi M. Folate and B12 levels correlate with histological severity in NASH patients. Nutrients (2018) 10 440. DOI: 10.3390/nu10040440 46. Fu L, Cheng H, Zhao X, Hou D, Xie X, Mi J. Distinct causal effects of body fat distribution on cardiometabolic traits among children: findings from the BCAMS study. Nutr Metab Cardiovasc Dis (2022) 32 1753-65. DOI: 10.1016/j.numecd.2022.03.030 47. Fu L, Li YN, Luo D, Deng S, Hu YQ. Plausible relationship between homocysteine and obesity risk via MTHFR gene: a meta-analysis of 38,317 individuals implementing Mendelian randomization. Diabetes Metab Syndr Obes (2019) 12 1201-12. DOI: 10.2147/dmso.s205379 48. Fu L, Li YN, Luo D, Deng S, Wu B, Hu YQ. Evidence on the causal link between homocysteine and hypertension from a meta-analysis of 40, 173 individuals implementing Mendelian randomization. J Clin Hypertens (2019) 21 1879-94. DOI: 10.1111/jch.13737 49. Fu L, Wang Y, Hu YQ. Association between homocysteine and non-alcoholic fatty liver disease: Mendelian randomisation study. Eur J Clin Investig (2022) 53 e13895. DOI: 10.1111/eci.13895 50. Fu L, Wang Y, Hu YQ. Causal effects of B vitamins and homocysteine on obesity and musculoskeletal diseases: a Mendelian randomization study. Front Nutr (2022) 9 1048122. DOI: 10.3389/fnut.2022.1048122 51. Fu L, Wang Y, Hu YQ. Inferring causal effects of homocysteine and B-vitamin concentrations on bone mineral density and fractures: Mendelian randomization analyses. Front Endocrinol (2022) 13 1037546. DOI: 10.3389/fendo.2022.1037546 52. Fu L, Wang Y, Li T, Hu YQ. A novel approach integrating hierarchical clustering and weighted combination for association study of multiple phenotypes and a genetic variant. Front Genet (2021) 12 654804. DOI: 10.3389/fgene.2021.654804 53. Fu L, Wang Y, Li T, Yang S, Hu YQ. A novel hierarchical clustering approach for joint analysis of multiple phenotypes uncovers obesity variants based on ARIC. Front Genet (2022) 13 791920. DOI: 10.3389/fgene.2022.791920 54. Fu L, Zhang M, Hu YQ, Zhao X, Cheng H, Hou D. Gene-gene interactions and associations of six hypertension related single nucleotide polymorphisms with obesity risk in a Chinese children population. Gene (2018) 679 320-7. DOI: 10.1016/j.gene.2018.09.019 55. Talari HR, Molaqanbari MR, Mokfi M, Taghizadeh M, Bahmani F, Tabatabaei SMH. The effects of vitamin B12 supplementation on metabolic profile of patients with non-alcoholic fatty liver disease: a randomized controlled trial. Sci Rep (2022) 12 14047. DOI: 10.1038/s41598-022-18195-8 56. Rush EC, Katre P, Yajnik CS. Vitamin B12: one carbon metabolism, fetal growth and programming for chronic disease. Eur J Clin Nutr (2014) 68 2-7. DOI: 10.1038/ejcn.2013.232 57. Werge MP, McCann A, Galsgaard ED, Holst D, Bugge A, Albrechtsen NJW. The role of the transsulfuration pathway in non-alcoholic fatty liver disease. J Clin Med (2021) 10 1081. DOI: 10.3390/jcm10051081 58. Mani S, Li H, Yang G, Wu L, Wang R. Deficiency of cystathionine gamma-lyase and hepatic cholesterol accumulation during mouse fatty liver development. Sci Bull (2015) 60 336-47. DOI: 10.1007/s11434-014-0722-7
--- title: Alpha1-antitrypsin protects the immature mouse brain following hypoxic-ischemic injury authors: - Shan Zhang - Wendong Li - Yiran Xu - Tao Li - Joakim Ek - Xiaoli Zhang - Yafeng Wang - Juan Song - Changlian Zhu - Xiaoyang Wang journal: Frontiers in Cellular Neuroscience year: 2023 pmcid: PMC10025360 doi: 10.3389/fncel.2023.1137497 license: CC BY 4.0 --- # Alpha1-antitrypsin protects the immature mouse brain following hypoxic-ischemic injury ## Abstract Introduction: Preterm brain injury often leads to lifelong disabilities affecting both cognitive and motor functions, and effective therapies are limited. Alpha1-antitrypsin (AAT), an endogenous inhibitor of serine proteinases with anti-inflammatory, anti-apoptotic, and cytoprotective properties, might be beneficial in treating preterm brain injury. The aim of this study was to investigate whether AAT has neuroprotective effects in a mouse preterm brain injury model. Methods: Preterm brain injury was induced on postnatal day 5, and mouse pups’ right common carotid arteries were cut between two ligations followed by hypoxia induction. Brain injury was evaluated through immunohistochemistry staining and magnetic resonance imaging. Fluoro-Jade B and immunohistochemistry staining were performed to investigate the neuronal cell death and blood-brain barrier (BBB) permeability. The motor function and anxiety-like behaviors were revealed by CatWalk gait analysis and the open field test. Results: After hypoxia-ischemia (HI) insult, brain injury was alleviated by AAT treatment, and this was accompanied by reduced BBB permeability, reduced neuronal cell death and caspase-3 activation, and inhibition of microglia activation. In addition, AAT administration significantly improved HI-induced motor function deficiencies in mice. The neuroprotective effect of AAT was more pronounced in male mice. Conclusion: AAT treatment is neuroprotective against preterm brain injury in neonatal mice, and the effect is more pronounced in males. ## Introduction Preterm birth is a major global health problem in terms of neonatal morbidity and mortality (Perin et al., 2022). Improvements in perinatal medicine and neonatal intensive care have increased preterm infant survival, but survivors are at high risk for neonatal morbidities such as brain injury (Song et al., 2016; Juul et al., 2020; Liu et al., 2020). Preterm brain injury is mainly manifested as white-matter injury or intraventricular hemorrhage and is a leading cause of neurodevelopmental disabilities such as cerebral palsy (CP), intellectual disability, autism spectrum disorders, deafness, and blindness (Hafström et al., 2018; Ballabh and de Vries, 2021; Song et al., 2021). The pathogenesis of preterm brain injury is multifactorial, and both infection/inflammation and hypoxia-ischemia (HI) are thought to play crucial roles by inducing oxidative stress and neuroinflammation and subsequent neural cell death and by inhibiting pre-oligodendrocyte maturation (Hagberg et al., 2015; van Tilborg et al., 2018). Growing evidence shows that there is a tertiary phase of ongoing inflammation and neural cell death for months after the initial brain injury, which interrupts brain repair and functional development and contributes to neurological sequelae (Fleiss and Gressens, 2012; Zhang et al., 2017). There is currently no widely accepted therapeutic strategy to prevent or treat preterm brain injury, although several preclinical and clinical studies have shown the neuroprotective effects or reduced incidence of neurological disability in preterm infants with the use of recombinant human erythropoietin and stem cell therapy (Song et al., 2016; Vaes et al., 2021; Yates et al., 2021; Wu et al., 2022). Supportive care to maintain the stability of vital signs is still the main treatment for preterm brain injury. There is thus a pressing need for improving our understanding of the mechanisms of perinatal brain injury and for conducting comparative and translational studies on how to reduce cell death, increase cell survival, and promote brain regeneration and repair after preterm brain injury. Serine proteases appear to play important roles in cellular physiology, and endogenous serine protease inhibitors regulate the activity of the serine proteases in the cell. Both serine proteases and serine protease inhibitors have been implicated in the development, plasticity, and pathology of the nervous system (Yoshida and Shiosaka, 1999). The neuroprotective effects of certain serine protease inhibitors have been reported, raising questions about their potential role in treating brain injury (Zhang et al., 2022). Alpha1-antitrypsin (AAT) is an abundant serine protease inhibitor belonging to the serine protease superfamily (Hazari et al., 2017). AAT has been shown to have an anti-inflammatory function that modulates the inflammatory response by reducing caspase-1 activity (Toldo et al., 2011), inducing the production of the anti-inflammatory cytokine IL-10 (Janciauskiene et al., 2007), and suppressing microglia-mediated neuroinflammation (Gold et al., 2014; Zhou et al., 2018). In addition to its anti-inflammatory effect, AAT also has anti-apoptotic properties and has been shown to inhibit caspase-3 activity (Petrache et al., 2006) and to regulate apoptotic cell death in neurons and neutrophils (Bergin et al., 2014; Cabezas-Llobet et al., 2018). Furthermore, AAT plays a cytoprotective role in vascular endothelial cells to reduce ischemia-reperfusion-induced vascular injury (Feng et al., 2015). Although many studies on AAT therapy have focused on α1-antitrypsin deficiency-related lung disease (Miravitlles et al., 2017; McEnery et al., 2022), other studies have demonstrated that AAT is also effective in treating type 1 diabetes and ischemic stroke damage (Guttman et al., 2014; Moldthan et al., 2014). However, the role of AAT in preterm brain injury is unknown. An important neuroanatomical feature of the brain injury observed in preterm infants is a failure in myelination, referred to as white matter injury in the preterm brain (Ophelders et al., 2020). In the central nervous system, myelin is formed by oligodendrocytes, which mature from premyelinating oligodendrocyte (preOLs; Khwaja and Volpe, 2008). Due to the large quantity of preOLs in the periventricular white matter in infants between gestational weeks 24 and 32, the neonatal human brain is particularly susceptible to HI during this period of development (Back et al., 2001). In rodents, preOLs are abundant from postnatal day 2 (P2) to P5 (Craig et al., 2003). Thus, in this study, we used the well-established neonatal HI mouse model adapted to P5 mice (Albertsson et al., 2014) to investigate whether AAT has neuroprotective effects against preterm brain injury. ## Animals and the hypoxic-ischemic brain injury model C57BL/6J male and female mice aged 8–10 weeks were purchased from Janvier Labs (Paris, France) and housed in a temperature-controlled and pathogen-free environment with a 12:12-h light-dark cycle. The pups were generated by crossing female and male mice. P5 mouse pups were anesthetized with isoflurane ($5\%$ for induction, 1.5–$2.0\%$ for maintenance) in a mixture of air and oxygen (1:1), and the right common carotid artery was cut between double ligatures. After the surgical operation, pups were returned to their dams for 1 h. Later, pups were placed in a chamber perfused with a humidified gas mixture ($10\%$ oxygen in nitrogen) at 36°C for 45 min. After hypoxic exposure, the pups were returned to their dams until sacrifice or until weaning at P21. Pups that died or bled profusely during the surgical operation were excluded from study. A total of 12 pups were excluded, and a total of 170 pups were used in all experiments. The pups were arbitrarily assigned to each group stratified by sex. Pups for behavior tests were assigned to normal, vehicle-treated HI, and AAT-treated HI groups, and pups for other experiments were assigned to vehicle-treated HI and AAT-treated HI groups. The sample size was calculated based on experiences in our previous studies (Li et al., 2019; Rodriguez et al., 2020). Mice were maintained in the Laboratory for Experimental Biomedicine of Gothenburg University, and all of the experiments were performed in accordance with Swedish national guidelines established by the Swedish Board of Agriculture (SJVFS 2019: 9) and were approved by the Gothenburg Animal Ethics Committee ($\frac{2200}{2019}$). ## AAT administration The powdered AAT (Sigma, Cat# 9041-92-3) was dissolved in saline at 1 mg to 20 μl for the stock solution and then diluted five times in saline for the working solution. We performed a comparison between intranasal and intraperitoneal administration of AAT treatment and found that the concentration of AAT in the brain tissue was higher after intraperitoneal administration compared to intranasal administration (Supplementary Figure 1). Thus, AAT was administered intraperitoneally in our study. The treatment dose of AAT was chosen based on previous research found in the literature (Janciauskiene et al., 2011; Toldo et al., 2011, 2016; Lewis, 2012; Moldthan et al., 2014; Mauro et al., 2017) and our preliminary experiments. AAT treatment (50 mg/kg) was given twice, with the first dose being given immediately after HI at P5 and the second dose given 72 h later at P8. Vehicle control pups received the same volume of saline intraperitoneally. The whole experimental procedure is illustrated in Figure 1. Figure 1: Experimental timeline. ## Immunohistochemistry staining Pups were deeply anesthetized with an overdose of sodium pentobarbital and perfused intracardially with PBS. The mouse brains were fixed in $5\%$ buffered formaldehyde (Histofix; Histolab, Gothenburg, Sweden) at 4°C overnight. After dehydration with graded ethanol and xylene, brain samples were paraffin-embedded and cut into 5 μm coronal sections. Brain sections were deparaffinized in xylene and rehydrated in ethanol. Antigen retrieval was performed by boiling sections in sodium citrate buffer, and nonspecific binding was blocked with $4\%$ donkey serum in PBS. The primary antibodies were mouse anti-microtubule-associated protein 2 (MAP2; 1:1,000 dilution, clone HM-2, Sigma, M4403), mouse anti-myelin basic protein (MBP; 1:500 dilution, clone SMI94, BioLegend, 836504), rabbit anti-cleaved caspase-3 (1:200 dilution, ASP175, Cell Signaling Technology, Beverly, Cat# 9661), goat anti-albumin (1:6,000 dilution, Abcam, Cambridge, UK, Cat# ab19194), and rat anti-galectin-3 (1:200 dilution, Invitrogen, Carlsbad, CA, USA, Cat# 14-5301-82). After primary antibody incubation, the appropriate biotinylated secondary antibodies (1:200 dilutions; all from Vector Laboratories, Burlingame, CA, USA) were added for 60 min at room temperature. After blocking endogenous peroxidase activity with $3\%$ H2O2, sections were visualized with Vectastain ABC Elite (Vector Laboratories) and 0.5 mg/ml 3,3’-diaminobenzidine enhanced with ammonium nickel sulfate, β-D glucose, ammonium chloride, and β-glucose oxidase. ## Fluoro-jade B staining After deparaffinization, sections were incubated with freshly prepared $0.06\%$ potassium permanganate (KMnO4) for 15 min and rinsed in distilled water. Slides were then incubated with $0.0004\%$ Fluoro-Jade B (Merck Millipore, Burlington, USA, Cat# AG310-30MG) in $0.09\%$ acetic acid for 30 min in a dark container. After rinsing with distilled water, slides were dehydrated and covered with mounting medium. ## Brain injury evaluation Paraffin-embedded brain samples were cut into 5 μm thick coronal sections, and eight consecutive coronal sections with an interval of 500 μm between each section were measured for each brain. The gray-matter area was determined by measuring the MAP2 immunoreactive area, and the subcortical white matter area was determined by measuring the MBP immunoreactive area. The MAP2-positive and MBP-positive areas in each section were measured in both hemispheres, and the volume was calculated using the following formula: V = ΣA × P × T, where V = the total volume, ΣA = the sum of area measurements, P = the inverse of the sampling fraction, and T = the section thickness. The total tissue loss volume ratio of gray matter and subcortical white matter was calculated using the following formula: [(contralateral hemisphere MAP-2 or MBP-positive volume − ipsilateral hemisphere MAP-2 or MBP-positive volume) / contralateral hemisphere MAP-2 or MBP-positive volume]. All of the evaluations were performed using Image J software by investigators blinded to group assignment. ## Cell counting Area contours with fixed locations were drawn and measured in every 50th section. The gelactin-3-positive and caspase-3-positive cells were counted within a defined area (one visual field) of the cortex (100×), striatum (100×), CA1 (100×), and habenular nuclei (200×) for each section. The Fluoro-Jade B-positive cells were counted within the same area of the cortex (100×), striatum (100×), CA1 (200×), and habenular nuclei (200×) for each section. All the evaluations were performed using Image J software by investigators blinded to group assignment. ## Protein extraction and immunoblotting Brain tissue from the parietal cortex in both hemispheres was dissected out and homogenized immediately on ice with a Dounce tissue homogenizer (Sigma, D8938) in tissue lysis buffer [15 mM Tris-HCl, pH 7.6, 320 mM sucrose, 1 mM dithiothreitol, 1 mM MgCl2, 3 mM EDTA-K, and $0.5\%$ protease inhibitor cocktail (Sigma, P8340)]. The homogenate was centrifuged at 4°C and 9,200× g for 15 min, and the protein concentration of the supernatant was measured using the bicinchoninic acid method. Individual samples of 20 μg protein were loaded and run on $4\%$–$12\%$ NuPAGE Bis-Tris gels (Invitrogen, Cat# NP0336BOX) then transferred to reinforced nitrocellulose membranes (Bio-Rad, Cat# 162-0112). Membranes were incubated with mouse anti-fodrin (1:1,000 dilution, Enzo Life Sciences, Cat# BML-FG6090-0500) and rabbit anti-actin (1:200 dilution, Sigma, Cat# A2066) overnight at 4°C. After washing, the membranes were incubated with peroxidase-labeled goat anti-rabbit IgG antibody (1:2,000 dilution, Vector, Cat# PI-1000) or peroxidase-labeled horse anti-mouse IgG antibody (1:4,000 dilution, Vector, Cat# PI-2000). Immunoreactive species were visualized using the Super Signal West Pico PLUS Chemiluminescent Substrate (ThermoFisher Scientific, Cat# 34580) and a LAS 3000 cooled CCD camera (Fujifilm, Tokyo, Japan). ## Caspase-3 activity assay A total of 25 μl homogenate sample was mixed with 75 μl extraction buffer containing 50 mM Tris-HCl (pH 7.3), 100 mM NaCl, 5 mM EDTA, 1 mM EGTA, 1 mM PMSF, and $1\%$ protease inhibitor cocktail on a microtiter plate. After incubation for 15 min at room temperature, 25 μM caspase-3-substrate (Ac-DEVD-AMC, Peptide Institute, 670613) in 100 μl assay buffer was added. Caspase-3 activity was measured using a Spectramax Gemini microplate fluorometer (excitation/emission wavelength $\frac{380}{460}$ nm every 2 min for 1 h at 37°C) and expressed as pmol AMC/mg protein per minute. ## T2-weighted imaging T2-weighted imaging was performed on a preclinical MR scanner 4.7 T (MR Solution, United Kingdom) at P60. Fast spin echo sequence was used for T2-weighted imaging, and the scanning parameters were set as follows: repetition time = 5,000 ms, echo time = 51 ms, reversal angle = 180°, field of view = 22 mm × 22 mm, matrix = 256 × 256, number of slices = 18, slice thickness = 1.0 mm. Preclinical Scan 1.2 software was used for image acquisition, and the volume and statistics functions in the ITK-SNAP 3.8.0 software was used for tissue volume quantification. The total tissue loss volume was calculated as the contralateral hemisphere volume minus the ipsilateral hemisphere volume. ## CatWalk XT gait analysis The CatWalk XT is a gait analysis system with a 1.3 m horizontal glass gate covered by a removable tunnel creating a dimmed light on the walkway. The mice were trained to walk at the beginning of the walkway and to traverse the plate towards their home cage voluntarily. Data were collected by a high-speed color camera located beneath the glass gate and sent to a connected computer. Data analysis was performed automatically by the CatWalk XT software. After successive trainings for 5 days, mice performed a minimum of three nonstop runs for quantification in the CatWalk XT analysis software. A maximum speed variation of $60\%$, a camera gain of 28.7 dB, and a detection threshold of 0.1 were set for the detection of all parameters used in the experiments. ## Open field test The open field test was conducted in a black acrylic glass box (100 cm × 100 cm × 40 cm) with an overhead lamp pointed at the center of the field. Each mouse was placed in the corner of the apparatus, and locomotion parameters were recorded for 5 min. All mice underwent a 30-min acclimation period prior to the start of open field test. ## Statistical analysis The normality of all data was tested by the Shapiro–Wilk test, and the homogeneity of variance of all data was tested by Levene’s test. For comparisons between two groups, unpaired t-tests were used for data with normal distribution and homogeneity of variance, and the Mann–Whitney U-test was applied for data with a non-normal distribution. For comparisons between three groups, one-way ANOVA with a Bonferroni post-hoc test was used for data with normal distribution and homogeneity of variance, and the Kruskal–Wallis test was used for data with a non-normal distribution. For two-dimensional data, two-way ANOVA with Bonferroni post-hoc test was used for data with normal distribution and homogeneity of variance, and the Scheirer–Ray–Hare test was used for data with a non-normal distribution. The results are presented as means ± standard deviations, and $p \leq 0.05$ was considered statistically significant. IBM SPSS 21.0 (NY, USA) was used for data analysis. ## AAT treatment alleviated brain injury after HI We first wanted to examine whether AAT treatment affected HI-induced brain damage. Brain injury was evaluated by measuring the total tissue loss volume ratio at 7 days after HI based on MAP2 immunochemistry staining (Figure 2A). The average value in AAT-treated mice was $35.2\%$ lower compared to vehicle-treated littermates (Figure 2B). However, a significant difference between AAT-treated and vehicle-treated mice was seen in males but not in females (Figure 2C). The average value was reduced by $42.5\%$ in AAT-treated male mice compared to vehicle-treated male littermates, while the average value was only $22.8\%$ lower in AAT-treated female mice compared to vehicle-treated female littermates (Figure 2C). Figure 2: AAT treatment reduced HI-induced neonatal mouse brain injury. (A) Representative MAP2 staining of coronal brain sections at the hippocampus level (upper panels) and striatum level (lower panels) at P12 after HI in the vehicle-treated and AAT-treated groups. (B) Quantification of total brain tissue volume loss ratio at P12 (n = 26/group, p = 0.003). (C) Quantification of total brain tissue volume loss ratio at P12 in male and female mice (n = 15/group for males, p = 0.004; n = 11/group for females, p = 0.277). (D) Representative MBP staining at the hippocampus level showing the myelin structure in the subcortical white matter at P12 after HI in the vehicle-treated and AAT-treated groups. The right panels show higher magnifications of the MBP-stained subcortical white matter in the ipsilateral hemisphere. (E) Quantification of the volume loss of MBP+ subcortical white matter at P12 (n = 26/group, p = 0.002). (F) Quantification of the volume loss of MBP+ subcortical white matter at P12 in male and female mice (n = 15/group for males, p = 0.001; n = 11/group for females, p = 0.505). (G) Representative cerebral coronal T2-weighted images at P60 after HI in vehicle-treated and AAT-treated groups. (H) Quantification of total brain tissue volume loss at P60 (n = 10/group, p = 0.011). (I) Quantification of total brain tissue volume loss at P60 in male and female mice (n = 5/group for males, p = 0.222; n = 5/group for females, p = 0.079). p < 0.05, *p < 0.01. For evaluating the effect of AAT treatment on white matter, MBP immunochemistry staining was performed at 7 days after HI (Figure 2D). The white matter injury was calculated as the MBP-positive tissue loss volume ratio in the subcortical white-matter area. The average value in AAT-treated mice was decreased by $31.2\%$ compared to vehicle-treated littermates (Figure 2E). Analyzing the data according to sex, we found again that AAT treatment significantly reduced the degree of white-matter injury in males but not in females, and the average value was $42.9\%$ lower in AAT-treated male mice compared to vehicle-treated male littermates, but there was no significant difference in females between the AAT and vehicle groups (Figure 2F). We also evaluated the long-term effect of AAT treatment on brain injury in neonatal mice by magnetic resonance imaging (Figure 2G). According to the results of T2-weighted imaging analysis at P60 (55 days after HI), there was a significant $47.9\%$ reduction in the total tissue loss volume in the AAT-treated group compared to the vehicle-treated group (Figure 2H), but no sex difference was observed (Figure 2I). ## AAT treatment improved motor function deficiency but not anxiety-like behavior after HI To determine whether AAT treatment affected HI-induced motor function deficiency and anxiety-like behavior in the mice, a battery of neurobehavior tests were performed at P60 (55 days after HI). Gait analysis by CatWalk XT showed that the paw parameter values of print area, maximum contact area, and mean intensity of all four paws were significantly reduced in the vehicle-treated group compared to normal mice, which indicated that motor development in neonatal mice was markedly impaired after HI injury. However, no significant differences in these parameters between vehicle-treated and AAT-treated groups were observed (Supplementary Figure 2). Stride length is defined as the distance between the placement of a paw and the subsequent placement of the same paw (Figure 3A), and the stride length of all four paws was significantly lower in vehicle-treated mice compared to normal mice, and this deficiency was rescued by AAT treatment (Figure 3B). We also analyzed changes in the stride length of all four paws between the AAT-treated and vehicle-treated groups by sex, and the protective effect of AAT administration was evident in both males and females (Figures 3C,D). Figure 3: Motor function deficiency and anxiety-like behaviors in neonatal HI mice after AAT treatment. (A) Graphical representation of stride length in the normal, vehicle-treated, and AAT-treated groups. RF: right front paw; RH: right hind paw; LF: left front paw; LH: left hind paw. (B) Analysis of stride length of four paws in three groups [n = 22/group, normal vs. veh (), RF (p = 0.007), RH (p = 0.006), LF (p = 0.005), LH (p = 0.006); veh vs. AAT (#), RF (p = 0.000), RH (p = 0.000), LF (p = 0.000), LH (p = 0.001)]. (C) Analysis of stride length of four paws in male mice [n = 14/group, normal vs. veh (), RF (p = 0.538), RH (p = 0.592), LF (p = 0.25), LH (p = 0.353); veh vs. AAT (#), RF (p = 0.007), RH (p = 0.01), LF (p = 0.007), LH (p = 0.015)]. (D) Analysis of stride length of four paws in female mice [n = 8/group, normal vs. veh (), RF (p = 0.001), RH (p = 0.000), LF (p = 0.004), LH (p = 0.003); veh vs. AAT (#), RF (p = 0.009), RH (p = 0.006), LF (p = 0.017), LH (p = 0.018)]. (E) Representative heat map of locomotion in normal, vehicle-treated, and AAT-treated mice. (F) Analysis of the total distance in the open field in the three groups (n = 22/group, normal vs. veh (), p = 0.018; veh vs. AAT (#), p = 0.589). (G) Analysis of distance traveled in the center area in the three groups (n = 22/group, normal vs. veh (), p = 0.000; veh vs. AAT (#), p = 0.054). (H) Analysis of the total distance in the open field in male and female mice (n = 14/group for males, normal vs. veh (), p = 0.397; veh vs. AAT (#), p = 1.000; n = 8/group for females, normal vs. veh (), p = 0.676; veh vs. AAT (#), p = 0.22). (I) Analysis of distance traveled in the center area in male and female mice (n = 14/group, normal vs. veh (), p = 0.005; veh vs. AAT (#), p = 1.000; n = 8/group for females, normal vs. veh (), p = 0.001; veh vs. AAT (#), p = 0.003). p < 0.05, p < 0.01, *p < 0.001; #p < 0.05, ##p < 0.01, ###p < 0.001. Anxiety-like behavior and locomotor activity were examined by the open field test (Figure 3E). The open field test of locomotor activity requires normal motor skills and is suitable for the evaluation of anxiety level and the response to a novel environment. Some outcomes, such as time spent in the center and distance traveled in the center, likely gauge some aspects of emotionality, including anxiety. We found that the total distance traveled in the open field of vehicle-treated HI mice was reduced compared to normal mice (Figure 3F). In addition, normal mice would walk along the periphery in a new environment, but this phenomenon was not observed in vehicle-treated HI mice, and the distance traveled in the center area by vehicle-treated HI mice was increased compared to normal mice (Figure 3G). AAT treatment did not have an effect on either the total distance traveled in the open field (Figure 3F) or the distance traveled in the center area (Figure 3G). Analyzing the data according to sex, AAT administration had no apparent effect on the total distance traveled in the open field in either male or female HI mice (Figure 3H), but an effect of AAT treatment on the distance traveled in the center area was found in female HI mice (Figure 3I). ## AAT treatment decreased the permeability of the blood-brain barrier (BBB) and inhibited microglia activation after HI Extravagated albumin can be detected with immunohistochemical methods and is a straightforward way to demonstrate the presence of BBB leakage. We observed albumin staining in the parenchyma of the cortex, hippocampus, and part of the thalamic region in the ipsilateral hemispheres at 7 days after HI (Figure 4A). The albumin-positive area in the injured hemisphere was significantly reduced in AAT-treated mice compared to vehicle-treated mice (Figure 4B), but no sex difference was observed between the two groups (Figure 4C). Figure 4: Albumin extravasation and microglia activation in neonatal HI mice after AAT treatment. (A) Representative pictures at the hippocampal level showing the albumin-positive areas in the neonatal mouse brain at 24 h after HI in vehicle-treated and AAT-treated groups. (B) Measurement of the albumin-positive area at 24 h after HI (n = 14/group, p = 0.022). (C) Measurement of the albumin-positive area at 24 h after HI in male and female mice (n = 7/group, p = 0.107 for males and p = 0.111 for females). (D) Representative images showing the immunochemistry staining of activated macroglia in the cortex (Cx), hippocampus (CA1), habenula nucleus (HN), and striatum (Str) at 24 h after HI in the vehicle-treated and AAT-treated groups. (E) Quantification of galectin-3-positive cells at 24 h after HI in the Cx, CA1, HN, and Str [n = 14/group, Cx (p = 0.508), CA1 (p = 0.464), HN (p = 0.172), Str (p = 0.03)]. (F) Quantification of galectin-3-positive cells in the striatum area at 24 h after HI in male and female mice (n = 7/group, p = 0.057 for males and p = 0.262 for females). p < 0.05.* Galectin-3 immunochemistry staining was performed to detect the activation of microglia in brain tissue at 24 h after HI (Figure 4D). We measured the density of activated microglia cells in four brain regions (the cortex, CA1, habenular nuclei, and striatum), but only the density of activated microglia cells in the striatum was statistically reduced in AAT-treated vs. vehicle-treated mice (Figure 4E). Furthermore, the effects of AAT on microglial activation in the striatum did not exhibit any sex differences (Figure 4F). ## AAT treatment reduced neuronal cell death after HI The quantification of neuronal cell death based on the number of Fluoro Jade B-positive cells demonstrated substantial overall neuronal cell loss at 24 h after HI (Figure 5A). Regional analysis showed that the numbers of dead or dying neuronal cells in the habenular nuclei and striatum were significantly reduced in AAT-treated compared to vehicle-treated mice (Figure 5B), but no sex difference was observed (Figures 5C,D). Figure 5: Neuronal cell death in neonatal HI mice after AAT treatment. (A) Representative Fluoro-Jade B staining in the cortex (Cx), hippocampus (CA1), habenula nucleus (HN), and striatum (Str) at 24 h after HI in vehicle-treated and AAT-treated groups. (B) Quantification of Fluoro-Jade B-labeled cells at 24 h after HI in the Cx, CA1, HN, and Str [n = 12/group, cortex (p = 0.192), CA1 (p = 0.719), HN (p = 0.005), Str (p = 0.012)]. (C) Quantification of Fluoro-Jade B-labeled cells in the habenula nucleus area at 24 h after HI in male and female mice (n = 6/group, p = 0.05 for males and p = 0.054 for females). (D) Quantification of Fluoro-Jade B-labeled cells in the striatum area at 24 h after HI in male and female mice (n = 6/group, p = 0.07 for males and p = 0.087 for females). p < 0.05, *p < 0.01. ## AAT treatment inhibited apoptosis after HI Caspase-dependent apoptotic cell death was investigated by measuring the active form of caspase-3 in different brain regions at 24 h after HI (Figure 6A). In the analyzed brain regions, caspase-3-positive cells were clearly increased in the cortex, habenular nuclei, and striatum in vehicle-treated mice compared to AAT-treated mice, but no significant changes were seen in the CA1 area between the two groups (Figure 6B). Caspase-3-positive cells were significantly decreased in AAT-treated male mice compared to vehicle-treated male mice in both the cortex and habenular nuclei (Figures 6C,D). However, in the striatum caspase-3 activation was significantly inhibited in AAT-treated female mice compared to vehicle-treated female mice (Figure 6E). Additionally, the caspase-3 activity assay clearly showed that AAT treatment significantly reduced caspase-3 activation after HI (Figure 6F), which was more pronounced in females (Figure 6G). Figure 6: Activation of caspase-3 and calpain in neonatal HI mice after AAT treatment. (A) The photomicrographs show immunochemistry staining of activated caspase-3 in the cortex (Cx), hippocampus (CA1), habenula nucleus (HN), and striatum (Str) at 24 h after HI in the vehicle-treated and AAT-treated groups. (B) Quantification of cleaved caspase-3-labeled cells at 24 h after HI in the Cx, CA1, HN, and Str (n = 14/group, p = 0.005 in Cx, p = 0.437 in CA1, p = 0.006 in HN, p = 0.001 in Str). (C) Quantification of cleaved caspase-3-labeled cells in the cortex at 24 h after HI in male and female mice (n = 7/group, p = 0.008 for males and p = 0.193 for females). (D) Quantification of cleaved caspase-3-labeled cells in the habenula nucleus at 24 h after HI in male and female mice (n = 7/group, p = 0.028 for males and p = 0.067 for females). (E) Quantification of cleaved caspase-3-labeled cells in the striatum at 24 h after HI in male and female mice (n = 7/group, p = 0.059 for males and p = 0.006 for females). (F) The caspase-3 activity in cortical tissue homogenate was measured at 24 h after HI in the AAT-treated and vehicle-treated groups (Veh = 12, AAT = 11, p = 0.011). (G) Measurement of caspase-3 activity at 24 h after HI in male and female mice (Veh = 6, AAT = 5, p = 0.273 in male; n = 6/group, p = 0.015 in female). (H) Representative immunoblotting of fodrin and actin in the cortical tissue from the injured hemisphere of the vehicle and AAT treatment groups at 24 h after HI. (I) Quantification of 120 kDa fragment expression at 24 h after HI in vehicle-treated and AAT-treated groups (n = 12/group, p = 0.024). (J) Quantification of 145/150 kDa fragment expression at 24 h after HI in vehicle-treated and AAT-treated groups (n = 12/group, p = 0.002). (K) Quantification of 120 kDa fragment expression at 24 h after HI in male and female mice (n = 6/group, p = 0.051 for males and p = 0.097 for females). (L) Quantification of 145/150 kDa fragment expression at 24 h after HI in male and female mice (n = 6/group, p = 0.007 for males and p = 0.086 for females). p < 0.05, *p < 0.01. The 280 kDa non-erythroid fodrin protein is a widely studied substrate for calpain. The identification of fodrin cleavage fragments is a method to detect the activation of calpain and caspase-3 (Wang et al., 2001), and calpain-mediated proteolysis of fodrin results in 145 kDa and 150 kDa fragments while caspase-3-mediated cleavage results in a 120 kDa fragment (Figure 6H). From the Western blot quantification results, it was observed that the expression of both the $\frac{145}{150}$ kDa fragments and 120 kDa fragment were significantly decreased in AAT-treated mice compared to vehicle-treated mice, which indicated that AAT treatment reduced the activation of both calpain and caspase-3 in the neonatal brain after HI injury (Figures 6I,J). The expression of proteolytically generated fodrin fragments between the two groups was also analyzed by sex, and the results showed that the inhibition of calpain activation by AAT administration was more pronounced in males (Figures 6K,L). ## Discussion In this work we conducted a preclinical study to test the therapeutic efficacy of AAT in a mouse model of neonatal brain injury. We show for the first time that AAT treatment attenuates HI-induced preterm brain injury and improves motor function deficits and that the neuroprotective effect of AAT is more robust in males. We found that AAT treatment attenuated both gray and white matter injury in an HI-induced preterm brain injury mouse model. Even though AAT is a 52 kDa protein, our results indicate that systemic administration AAT may cross BBB in the HI-injured immature brain, which could be related to the increased BBB permeability after cerebral HI insult (Ek et al., 2015). Improving neurofunctional outcome is one of the focuses of developing new intervention strategies for preterm brain injury. The stride length in HI mice after AAT treatment was improved significantly at P60. As a well-established parameter in the CatWalk XT analysis, an increase in the stride length after injury suggests greater trunk stability. To note, no other differences among the analysis parameters were observed between vehicle and AAT-treated mice, and one potential reason for this could be that we administered AAT only twice in a short period of time after HI, which may not have been long enough for AAT to give full protection and therefore may not reflect the effects of AAT on long-term neurodevelopment. In addition, the developing brain is highly neuroplastic, which may allow for rapid functional compensation, and thus the measurement of rodent functional defects should be performed in the early stage after brain injury (Ismail et al., 2017). The open field test is used to analyze the general locomotor activity and anxiety-like behavior in rodents (Seibenhener and Wooten, 2015). The total distance traveled in the open field can reflect the baseline locomotion of mice in a more stressful condition. We observed that the baseline locomotion of HI mice in the open field was reduced compared to normal mice, and AAT administration did not show any apparent effect on the baseline locomotion of HI mice. Mice that prefer staying close to the walls and to travel more in the periphery can be described as showing thigmotaxis, which is more pronounced in mice showing signs of anxiety-like behavior (Lamprea et al., 2008). In the present study, it was found that the distance traveled in the center area by HI mice was increased, which is the exact opposite of the natural reflex of normal mice, indicating that mice had lower thigmotaxis and anxiety after HI. Although there was no significant difference in the distance traveled in the center area between vehicle-treated HI mice and AAT -treated HI mice, the distance traveled in the center area was significantly reduced in female mice of the AAT treated group, indicating that the thigmotaxis and anxiety level of AAT-treated HI female mice tended to be more consistent with those of normal mice. However, the mechanisms of these findings remain unclear and require further investigation. Microglia play a pivotal role in perinatal brain injury, and they initially respond to stimuli such as HI or infection with the production of pro-inflammatory cytokines that eventually exacerbate brain injury (Hagberg et al., 2015). Microglia are associated with axonal damage and myelinating oligodendrocytes, which are major pathological components of white-matter injury (Shao et al., 2021), and it has been reported that suppressing the activation of microglia is one of the mechanisms for reducing HI-induced brain injury in neonatal mice (Arvin et al., 2002; Mallard et al., 2019). Our results showed that activated microglia in the striatum area were significantly reduced after AAT administration in HI mice. The BBB plays an important role in maintaining homeostasis and protecting neurons. The permeability of the BBB increases after HI insult in the neonatal mouse brain and persists for more than 24 h (Ek et al., 2015), and it is widely accepted that BBB dysfunction and microglia are closely related (Haruwaka et al., 2019). We found that BBB permeability was decreased after AAT treatment, and thus we speculate that AAT might attenuate brain injury by inhibiting microglial activation and decreasing BBB permeability during the initial pro-inflammatory phase of preterm brain injury following HI. However, the causal relationship between decreased BBB permeability, suppressed microglial activity, and AAT’s neuroprotective effect needs to be further explored. The premature brain is susceptible to injury from infective, ischemic, and inflammatory insults (Vannucci and Hagberg, 2004; Eklind et al., 2005; Gussenhoven et al., 2018), and preOLs in the premature brain are particularly vulnerable to these insults due to limited antioxidant defense mechanisms and high levels of mitochondrial oxygen consumption (Buser et al., 2010; Spaas et al., 2021). In preterm brain injury, impaired preOLs negatively impact the maturation of oligodendrocytes and cause myelination failure (Buser et al., 2012; Motavaf and Piao, 2021). In addition to myelination disturbance, neuronal cell death is also involved in preterm brain injury. Current evidence suggests that apoptotic cell death plays a prominent role in preterm brain injury, especially the caspase-3-dependent apoptotic pathway (Truttmann et al., 2020). Apoptosis is a form of programmed cell death, and apoptosis inhibition reduces brain injury and improves neurological function in rodent models of cerebral ischemia (Han et al., 2002; Blomgren et al., 2007). Several caspase family members induce apoptosis and participate in the final execution phase of apoptosis, but caspase-3 appears to be an especially important effector enzyme in neuronal apoptosis (Porter and Jänicke, 1999; Broughton et al., 2009). We found that AAT treatment reduced caspase-3 activation after HI, and the neuroprotective effect of AAT treatment might be related to the inhibition of caspase-dependent apoptotic cell death. Calpain is a calcium-dependent cysteine protease that has been proposed to participate in the turnover of cytoskeletal proteins and in the regulation of kinases, transcription factors, and receptors. Although calpain activation has historically been assumed to result in necrotic cell death, a contribution to apoptosis has also been suggested (Altznauer et al., 2004; Harwood et al., 2005). The expression of $\frac{145}{150}$ kDa fodrin fragments in AAT-treated mice was significantly reduced, suggesting that the activation of calpain was decreased after AAT treatment, which is in line with previous studies. It has been shown that AAT modulates microglial-mediated neuroinflammation by inhibiting calpain activation in vitro (Gold et al., 2014), and calpain inhibitors have also been shown to have neuroprotective effects in perinatal brain injury (Blomgren et al., 1999; Kawamura et al., 2005). Altogether our results suggest that AAT exerts its neuroprotective effect mainly by reducing neuronal apoptotic cell death by inhibiting the activation of caspase-3 and calpain after HI injury in the immature brain. We previously showed that there is cross talk between caspase-3 and calpain activation (Blomgren et al., 2001), but whether AAT selectively inhibits caspase-3 or calpain alone or if it non-selectively inhibits both needs to be investigated further. Sex-specific differences in the efficacy of certain neuroprotective drugs have been demonstrated in animal models of neonatal HI (Hagberg et al., 2004; Nijboer et al., 2007; Daher et al., 2018; Li et al., 2019; Rodriguez et al., 2020). Our previous study found that adaptaquin treatment reduced neonatal HI brain injury in a sex-dependent manner, and the neuroprotective effect was more obvious in males (Li et al., 2019), while another study reported the sex-dependent efficacy of magnesium sulfate in protecting against neonatal HI brain injury (Daher et al., 2018). In the present study, we found that the reduction in brain damage after AAT administration was more pronounced in male mice. It has been previously suggested that the sex differences in neuroprotection might be related to differences in the activation of neuronal cell death pathways after brain injury (Lang and McCullough, 2008), differences in sex hormones (Siddiqui et al., 2016), and differences in mitochondrial dysfunction (Demarest and McCarthy, 2015), but the underlying mechanisms behind the current observations need to be explored further. There are some limitations in our study. First, the HI-induced preterm brain injury model does not fully represent preterm brain injury in infants. Currently, there is a lack of a perfect rodent model to best mimic brain injury in preterm infants. Although multiple animal models have been used to study preterm brain injury, such as IL-1β or LPS systemic injection-induced diffuse white matter injury in neonatal rodents (Favrais et al., 2011; Fan et al., 2013), ibotenic acid injection-induced white and gray matter injury in P5 mice (Sárközy et al., 2007), collagenase injection-induced intracranial hemorrhage in P5 rats (Jinnai et al., 2020), and HI-induced brain injury in P5 mice (Albertsson et al., 2014), none of them fully represent the pathology of brain injury seen in preterm infants. Second, the neuroprotective effect of AAT is more obvious in males according to the overall findings in our study, but the specific mechanisms underlying the sex differences in AAT neuroprotection are still unclear. ## Conclusion AAT has neuroprotective effects in the immature brain following HI and thus may serve as a potential therapeutic strategy for preterm brain injury. The mechanisms underlying AAT’s neuroprotective effects in the immature brain warrant further investigation. ## 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/s. ## Ethics statement The animal study was reviewed and approved by Gothenburg Animal Ethics Committee. ## Author contributions CZ and XW conceived and designed the experiments. SZ, WL, TL, and YW performed the experiments. SZ, WL, YX, TL, XZ, and JE analyzed the data. SZ, YX, WL, XZ, CZ, and XW interpreted the results and prepared the figures. SZ, YX, CZ, and XW drafted the manuscript. JS, JE, CZ, and XW edited and revised the manuscript. The final manuscript was approved by all authors. 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/fncel.2023.1137497/full#supplementary-material. ## References 1. Albertsson A. M., Bi D., Duan L., Zhang X., Leavenworth J. W., Qiao L.. The immune response after hypoxia-ischemia in a mouse model of preterm brain injury. J. Neuroinflammation (2014) 11 153. DOI: 10.1186/s12974-014-0153-z 2. Altznauer F., Conus S., Cavalli A., Folkers G., Simon H. U.. Calpain-1 regulates Bax and subsequent Smac-dependent caspase-3 activation in neutrophil apoptosis. J. Biol. Chem. (2004) 279 5947-5957. DOI: 10.1074/jbc.M308576200 3. Arvin K. L., Han B. H., Du Y., Lin S. Z., Paul S. M., Holtzman D. M.. Minocycline markedly protects the neonatal brain against hypoxic-ischemic injury. Ann. Neurol. (2002) 52 54-61. DOI: 10.1002/ana.10242 4. Back S. A., Luo N. L., Borenstein N. S., Levine J. M., Volpe J. J., Kinney H. C.. Late oligodendrocyte progenitors coincide with the developmental window of vulnerability for human perinatal white matter injury. J. Neurosci. (2001) 21 1302-1312. DOI: 10.1523/JNEUROSCI.21-04-01302.2001 5. Ballabh P., de Vries L. S.. White matter injury in infants with intraventricular haemorrhage: mechanisms and therapies. Nat. Rev. Neurol. (2021) 17 199-214. DOI: 10.1038/s41582-020-00447-8 6. Bergin D. A., Reeves E. P., Hurley K., Wolfe R., Jameel R., Fitzgerald S.. The circulating proteinase inhibitor α-1 antitrypsin regulates neutrophil degranulation and autoimmunity. Sci. Transl. Med. (2014) 6 217ra1. DOI: 10.1126/scitranslmed.3007116 7. Blomgren K., Hallin U., Andersson A. L., Puka-Sundvall M., Bahr B. A., McRae A.. Calpastatin is up-regulated in response to hypoxia and is a suicide substrate to calpain after neonatal cerebral hypoxia-ischemia. J. Biol. Chem. (1999) 274 14046-14052. DOI: 10.1074/jbc.274.20.14046 8. Blomgren K., Leist M., Groc L.. Pathological apoptosis in the developing brain. Apoptosis (2007) 12 993-1010. DOI: 10.1007/s10495-007-0754-4 9. Blomgren K., Zhu C., Wang X., Karlsson J. O., Leverin A. L., Bahr B. A.. Synergistic activation of caspase-3 by m-calpain after neonatal hypoxia-ischemia: a mechanism of “pathological apoptosis”?. J. Biol. Chem. (2001) 276 10191-10198. DOI: 10.1074/jbc.M007807200 10. Broughton B. R., Reutens D. C., Sobey C. G.. Apoptotic mechanisms after cerebral ischemia. Stroke (2009) 40 e331-e339. DOI: 10.1161/STROKEAHA.108.531632 11. Buser J. R., Maire J., Riddle A., Gong X., Nguyen T., Nelson K.. Arrested preoligodendrocyte maturation contributes to myelination failure in premature infants. Ann. Neurol. (2012) 71 93-109. DOI: 10.1002/ana.22627 12. Buser J. R., Segovia K. N., Dean J. M., Nelson K., Beardsley D., Gong X.. Timing of appearance of late oligodendrocyte progenitors coincides with enhanced susceptibility of preterm rabbit cerebral white matter to hypoxia-ischemia. J. Cereb. Blood Flow Metab. (2010) 30 1053-1065. DOI: 10.1038/jcbfm.2009.286 13. Cabezas-Llobet N., Camprubí S., García B., Alberch J., Xifró X.. Human alpha 1-antitrypsin protects neurons and glial cells against oxygen and glucose deprivation through inhibition of interleukins expression. Biochim. Biophys Acta Gen. Subj. (2018) 1862 1852-1861. DOI: 10.1016/j.bbagen.2018.05.017 14. Craig A., Luo L. N., Beardsley D. J., Wingate-Pearse N., Walker D. W., Hohimer A. R.. Quantitative analysis of perinatal rodent oligodendrocyte lineage progression and its correlation with human. Exp. Neurol. (2003) 181 231-240. DOI: 10.1016/s0014-4886(03)00032-3 15. Daher I., Le Dieu-Lugon B., Lecointre M., Dupré N., Voisin C., Leroux P.. Time- and sex-dependent efficacy of magnesium sulfate to prevent behavioral impairments and cerebral damage in a mouse model of cerebral palsy. Neurobiol. Dis. (2018) 120 151-164. DOI: 10.1016/j.nbd.2018.08.020 16. Demarest T. G., McCarthy M. M.. Sex differences in mitochondrial (dys)function: implications for neuroprotection. J. Bioenerg. Biomembr. (2015) 47 173-188. DOI: 10.1007/s10863-014-9583-7 17. Ek C. J., D’Angelo B., Baburamani A. A., Lehner C., Leverin A. L., Leverin A. L.. Brain barrier properties and cerebral blood flow in neonatal mice exposed to cerebral hypoxia-ischemia. J. Cereb. Blood Flow Metab. (2015) 35 818-827. DOI: 10.1038/jcbfm.2014.255 18. Eklind S., Mallard C., Arvidsson P., Hagberg H.. Lipopolysaccharide induces both a primary and a secondary phase of sensitization in the developing rat brain. Pediatr. Res. (2005) 58 112-116. DOI: 10.1203/01.PDR.0000163513.03619.8D 19. Fan L.-W., Kaizaki A., Tien L.-T., Pang Y., Tanaka S., Numazawa S.. Celecoxib attenuates systemic lipopolysaccharide-induced brain inflammation and white matter injury in the neonatal rats. Neuroscience (2013) 240 27-38. DOI: 10.1016/j.neuroscience.2013.02.041 20. Favrais G., van de Looij Y., Fleiss B., Ramanantsoa N., Bonnin P., Stoltenburg-Didinger G.. Systemic inflammation disrupts the developmental program of white matter. Ann. Neurol. (2011) 70 550-565. DOI: 10.1002/ana.22489 21. Feng Y., Hu L., Xu Q., Yuan H., Ba L., He Y.. Cytoprotective role of alpha-1 antitrypsin in vascular endothelial cell under hypoxia/reoxygenation condition. J. Cardiovasc. Pharmacol. (2015) 66 96-107. DOI: 10.1097/FJC.0000000000000250 22. Fleiss B., Gressens P.. Tertiary mechanisms of brain damage: a new hope for treatment of cerebral palsy?. Lancet Neurol. (2012) 11 556-566. DOI: 10.1016/S1474-4422(12)70058-3 23. Gold M., Dolga A. M., Koepke J., Mengel D., Culmsee C., Dodel C.. α1-antitrypsin modulates microglial-mediated neuroinflammation and protects microglial cells from amyloid-β-induced toxicity. J. Neuroinflammation (2014) 11 165. DOI: 10.1186/s12974-014-0165-8 24. Gussenhoven R., Westerlaken R. J. J., Ophelders D., Jobe A. H., Kemp M. W., Kallapur S. G.. Chorioamnionitis, neuroinflammation and injury: timing is key in the preterm ovine fetus. J. Neuroinflammation (2018) 15 113. DOI: 10.1186/s12974-018-1149-x 25. Guttman O., Yossef R., Freixo-Lima G., Rider P., Porgador A., Lewis E. C.. α1-Antitrypsin modifies general NK cell interactions with dendritic cells and specific interactions with islet β-cells in favor of protection from autoimmune diabetes. Immunology (2014) 144 530-539. DOI: 10.1111/imm.12403 26. Hafström M., Källén K., Serenius F., Maršál K., Rehn E., Drake H.. Cerebral palsy in extremely preterm infants. Pediatrics (2018) 141 e20171433. DOI: 10.1542/peds.2017-1433 27. Hagberg H., Mallard C., Ferriero D. M., Vannucci S. J., Levison S. W., Vexler Z. S.. The role of inflammation in perinatal brain injury. Nat. Rev. Neurol. (2015) 11 192-208. DOI: 10.1038/nrneurol.2015.13 28. Hagberg H., Wilson M. A., Matsushita H., Zhu C., Lange M., Gustavsson M.. PARP-1 gene disruption in mice preferentially protects males from perinatal brain injury. J. Neurochem. (2004) 90 1068-1075. DOI: 10.1111/j.1471-4159.2004.02547.x 29. Han B. H., Xu D., Choi J., Han Y., Xanthoudakis S., Roy S.. Selective, reversible caspase-3 inhibitor is neuroprotective and reveals distinct pathways of cell death after neonatal hypoxic-ischemic brain injury. J. Biol. Chem. (2002) 277 30128-30136. DOI: 10.1074/jbc.M202931200 30. Haruwaka K., Ikegami A., Tachibana Y., Ohno N., Konishi H., Hashimoto A.. Dual microglia effects on blood brain barrier permeability induced by systemic inflammation. Nat. Commun. (2019) 10 5816. DOI: 10.1038/s41467-019-13812-z 31. Harwood S. M., Yaqoob M. M., Allen D. A.. Caspase and calpain function in cell death: bridging the gap between apoptosis and necrosis. Ann. Clin. Biochem. (2005) 42 415-431. DOI: 10.1258/000456305774538238 32. Hazari Y. M., Bashir A., Habib M., Bashir S., Habib H., Qasim M. A.. Alpha-1-antitrypsin deficiency: genetic variations, clinical manifestations and therapeutic interventions. Mutat. Res. Rev. Mutat. Res. (2017) 773 14-25. DOI: 10.1016/j.mrrev.2017.03.001 33. Ismail F. Y., Fatemi A., Johnston M. V.. Cerebral plasticity: windows of opportunity in the developing brain. Eur. J. Paediatr. Neurol. (2017) 21 23-48. DOI: 10.1016/j.ejpn.2016.07.007 34. Janciauskiene S. M., Bals R., Koczulla R., Vogelmeier C., Köhnlein T., Welte T.. The discovery of α1-antitrypsin and its role in health and disease. Respir. Med. (2011) 105 1129-1139. DOI: 10.1016/j.rmed.2011.02.002 35. Janciauskiene S. M., Nita I. M., Stevens T.. Alpha1-antitrypsin, old dog, new tricks. Alpha1-antitrypsin exerts. J. Biol. Chem. (2007) 282 8573-8582. DOI: 10.1074/jbc.M607976200 36. Jinnai M., Koning G., Singh-Mallah G., Jonsdotter A., Leverin A. L., Svedin P.. A model of germinal matrix hemorrhage in preterm rat pups. Front. Cell. Neurosci. (2020) 14 535320. DOI: 10.3389/fncel.2020.535320 37. Juul S. E., Comstock B. A., Wadhawan R., Mayock D. E., Courtney S. E., Robinson T.. A randomized trial of erythropoietin for neuroprotection in preterm infants. N Engl. J. Med. (2020) 382 233-243. DOI: 10.1056/NEJMoa1907423 38. Kawamura M., Nakajima W., Ishida A., Ohmura A., Miura S., Takada G.. Calpain inhibitor MDL 28170 protects hypoxic-ischemic brain injury in neonatal rats by inhibition of both apoptosis and necrosis. Brain Res. (2005) 1037 59-69. DOI: 10.1016/j.brainres.2004.12.050 39. Khwaja O., Volpe J. J.. Pathogenesis of cerebral white matter injury of prematurity. Arch. Dis. Child. Fetal Neonatal Ed. (2008) 93 F153-F161. DOI: 10.1136/adc.2006.108837 40. Lamprea M., Cardenas F. P., Setem J., Morato S.. Thigmotactic responses in an open-field. Braz. J. Med. Biol. Res. (2008) 41 135-140. DOI: 10.1590/s0100-879x2008000200010 41. Lang J. T., McCullough L. D.. Pathways to ischemic neuronal cell death: are sex differences relevant?. J. Transl. Med. (2008) 6 33. DOI: 10.1186/1479-5876-6-33 42. Lewis E. C.. Expanding the clinical indications for (α)1-antitrypsin therapy. Mol. Med. (2012) 18 957-970. DOI: 10.2119/molmed.2011.00196 43. Li K., Li T., Wang Y., Xu Y., Zhang S., Culmsee C.. Sex differences in neonatal mouse brain injury after hypoxia-ischemia and adaptaquin treatment. J. Neurochem. (2019) 150 759-775. DOI: 10.1111/jnc.14790 44. Liu S., Zhang X., Liu Y., Yuan X., Yang X., Zhang R.. Early application of caffeine improves white matter development in very preterm infants. Respir. Physiol. Neurobiol. (2020) 281 103495. DOI: 10.1016/j.resp.2020.103495 45. Mallard C., Tremblay M.-E., Vexler Z. S.. Microglia and neonatal brain injury. Neuroscience (2019) 405 68-76. DOI: 10.1016/j.neuroscience.2018.01.023 46. Mauro A. G., Mezzaroma E., Marchetti C., Narayan P., Del Buono M. G., Capuano M.. A preclinical translational study of the cardioprotective effects of plasma-derived alpha-1 anti-trypsin in acute myocardial infarction. J. Cardiovasc. Pharmacol. (2017) 69 273-278. DOI: 10.1097/FJC.0000000000000474 47. McEnery T., White M. M., Gogoi D., Coleman O., Bergin D., Jundi B.. Alpha-1 antitrypsin therapy modifies neutrophil adhesion in patients with obstructive lung disease. Am. J. Respir. Cell Mol. Biol. (2022) 67 76-88. DOI: 10.1165/rcmb.2021-0433OC 48. Miravitlles M., Dirksen A., Ferrarotti I., Koblizek V., Lange P., Mahadeva R.. European respiratory society statement: diagnosis and treatment of pulmonary disease in α(1)-antitrypsin deficiency. Eur. Respir. J. (2017) 50 1700610. DOI: 10.1183/13993003.00610-2017 49. Moldthan H. L., Hirko A. C., Thinschmidt J. S., Grant M. B., Li Z., Peris J.. Alpha 1-antitrypsin therapy mitigated ischemic stroke damage in rats. J. Stroke Cerebrovasc. Dis. (2014) 23 e355-e363. DOI: 10.1016/j.jstrokecerebrovasdis.2013.12.029 50. Motavaf M., Piao X.. Oligodendrocyte development and implication in perinatal white matter injury. Front. Cell. Neurosci. (2021) 15 764486. DOI: 10.3389/fncel.2021.764486 51. Nijboer C. H., Groenendaal F., Kavelaars A., Hagberg H. H., van Bel F., Heijnen C. J.. Gender-specific neuroprotection by 2-iminobiotin after hypoxia-ischemia in the neonatal rat via a nitric oxide independent pathway. J. Cereb. Blood Flow Metabol. (2007) 27 282-292. DOI: 10.1038/sj.jcbfm.9600342 52. Ophelders D., Gussenhoven R., Klein L., Jellema R. K., Westerlaken R. J. J., Hütten M. C.. Preterm brain injury, antenatal triggers and therapeutics: timing is key. Cells (2020) 9 1871. DOI: 10.3390/cells9081871 53. Perin J., Mulick A., Yeung D., Villavicencio F., Lopez G., Strong K. L.. Global, regional and national causes of under-5 mortality in 2000–19: an updated systematic analysis with implications for the sustainable development goals. Lancet Child Adolesc. Health (2022) 6 106-115. DOI: 10.1016/S2352-4642(21)00311-4 54. Petrache I., Fijalkowska I., Medler T. R., Skirball J., Cruz P., Zhen L.. alpha-1 antitrypsin inhibits caspase-3 activity, preventing lung endothelial cell apoptosis. Am. J. Pathol. (2006) 169 1155-1166. DOI: 10.2353/ajpath.2006.060058 55. Porter A. G., Jänicke R. U.. Emerging roles of caspase-3 in apoptosis. Cell Death Differ. (1999) 6 99-104. DOI: 10.1038/sj.cdd.4400476 56. Rodriguez J., Xie C., Li T., Sun Y., Wang Y., Xu Y.. Inhibiting the interaction between apoptosis-inducing factor and cyclophilin A prevents brain injury in neonatal mice after hypoxia-ischemia. Neuropharmacology (2020) 171 108088. DOI: 10.1016/j.neuropharm.2020.108088 57. Sárközy G., Griesmaier E., He X., Kapelari K., Urbanek M., Simbruner G.. T3 replacement does not prevent excitotoxic cell death but reduces developmental neuronal apoptosis in newborn mice. Eur. J. Paediatr. Neurol. (2007) 11 129-135. DOI: 10.1016/j.ejpn.2006.11.009 58. Seibenhener M. L., Wooten M. C.. Use of the open field maze to measure locomotor and anxiety-like behavior in mice. J. Vis. Exp. (2015) 96: e52434. DOI: 10.3791/52434 59. Shao R., Sun D., Hu Y., Cui D.. White matter injury in the neonatal hypoxic-ischemic brain and potential therapies targeting microglia. J. Neurosci. Res. (2021) 99 991-1008. DOI: 10.1002/jnr.24761 60. Siddiqui A. N., Siddiqui N., Khan R. A., Kalam A., Jabir N. R., Kamal M. A.. Neuroprotective role of steroidal sex hormones: an overview. CNS Neurosci. Ther. (2016) 22 342-350. DOI: 10.1111/cns.12538 61. Song J., Sun H., Xu F., Kang W., Gao L., Guo J.. Recombinant human erythropoietin improves neurological outcomes in very preterm infants. Ann. Neurol. (2016) 80 24-34. DOI: 10.1002/ana.24677 62. Song J., Wang Y., Xu F., Sun H., Zhang X., Xia L.. Erythropoietin improves poor outcomes in preterm infants with intraventricular hemorrhage. CNS Drugs (2021) 35 681-690. DOI: 10.1007/s40263-021-00817-w 63. Spaas J., van Veggel L., Schepers M., Tiane A., van Horssen J., Wilson D. M.. Oxidative stress and impaired oligodendrocyte precursor cell differentiation in neurological disorders. Cell. Mol. Life Sci. (2021) 78 4615-4637. DOI: 10.1007/s00018-021-03802-0 64. Toldo S., Mauro A. G., Marchetti C., Rose S. W., Mezzaroma E., Van Tassell B. W.. Recombinant human alpha-1 antitrypsin-Fc fusion protein reduces mouse myocardial inflammatory injury after ischemia-reperfusion independent of elastase inhibition. J. Cardiovasc. Pharmacol. (2016) 68 27-32. DOI: 10.1097/FJC.0000000000000383 65. Toldo S., Seropian I. M., Mezzaroma E., Van Tassell B. W., Salloum F. N., Lewis E. C.. Alpha-1 antitrypsin inhibits caspase-1 and protects from acute myocardial ischemia-reperfusion injury. J. Mol. Cell. Cardiol. (2011) 51 244-251. DOI: 10.1016/j.yjmcc.2011.05.003 66. Truttmann A. C., Ginet V., Puyal J.. Current evidence on cell death in preterm brain injury in human and preclinical models. Front. Cell Devel. Biol. (2020) 8 27. DOI: 10.3389/fcell.2020.00027 67. Vaes J. E. G., van Kammen C. M., Trayford C., van der Toorn A., Ruhwedel T., Benders M. J. N. L.. Intranasal mesenchymal stem cell therapy to boost myelination after encephalopathy of prematurity. Glia (2021) 69 655-680. DOI: 10.1002/glia.23919 68. van Tilborg E., Achterberg E. J. M., van Kammen C. M., van der Toorn A., Groenendaal F., Dijkhuizen R. M.. Combined fetal inflammation and postnatal hypoxia causes myelin deficits and autism-like behavior in a rat model of diffuse white matter injury. Glia (2018) 66 78-93. DOI: 10.1002/glia.23216 69. Vannucci S. J., Hagberg H.. Hypoxia-ischemia in the immature brain. J. Exp. Biol. (2004) 207 3149-3154. DOI: 10.1242/jeb.01064 70. Wang X., Karlsson J. O., Zhu C., Bahr B. A., Hagberg H., Blomgren K.. Caspase-3 activation after neonatal rat cerebral hypoxia-ischemia. Biol. Neonate (2001) 79 172-179. DOI: 10.1159/000047087 71. Wu Y., Sun Y., Wang X., Zhu C.. The regulated cell death and potential interventions in preterm infants after intracerebral hemorrhage. Curr. Neuropharmacol. (2022). DOI: 10.2174/1570159X21666221117155209 72. Yates N., Gunn A. J., Bennet L., Dhillon S. K., Davidson J. O.. Preventing brain injury in the preterm infant-current controversies and potential therapies. Int. J. Mol. Sci. (2021) 22 1671. DOI: 10.3390/ijms22041671 73. Yoshida S., Shiosaka S.. Plasticity-related serine proteases in the brain (review). Int. J. Mol. Med. (1999) 3 405-409. DOI: 10.3892/ijmm.3.4.405 74. Zhang Y., Chen Q., Chen D., Zhao W., Wang H., Yang M.. SerpinA3N attenuates ischemic stroke injury by reducing apoptosis and neuroinflammation. CNS Neurosci. Ther. (2022) 28 566-579. DOI: 10.1111/cns.13776 75. Zhang X., Rocha-Ferreira E., Li T., Vontell R., Jabin D., Hua S.. γδT cells but not αβT cells contribute to sepsis-induced white matter injury and motor abnormalities in mice. J. Neuroinflammation (2017) 14 255. DOI: 10.1186/s12974-017-1029-9 76. Zhou T., Huang Z., Zhu X., Sun X., Liu Y., Cheng B.. Alpha-1 antitrypsin attenuates M1 microglia-mediated neuroinflammation in retinal degeneration. Front. Immunol. (2018) 9 1202. DOI: 10.3389/fimmu.2018.01202
--- title: Is there any association between early trimester Triglyceride–glucose index and incidence of hypertensive disorder of pregnancy and adverse pregnancy outcomes? authors: - Yali Pan - Su Zou - Yingjia Xu - Ruomin Di - Huafen Gu - Zhangsheng Wang - Xiang Wei - Chenxi Yang - Gaofeng Zhang journal: Frontiers in Endocrinology year: 2023 pmcid: PMC10025371 doi: 10.3389/fendo.2023.1093991 license: CC BY 4.0 --- # Is there any association between early trimester Triglyceride–glucose index and incidence of hypertensive disorder of pregnancy and adverse pregnancy outcomes? ## Abstract ### Background Insulin resistance (IR) is a normal feature of pregnancy and plays a crucial role in the pathophysiology of hypertensive disorder of pregnancy (HDP). The triglyceride-glucose index (TyG index) has been shown as a simple and reliable alternative IR marker. This work aimed to investigate the association between the TyG index and the incidence of HDP and adverse pregnancy outcomes. ### Methods From January 2016 to December 2018, 289 women with HDP and 861 women without HDP were recruited at Shanghai Fifth People’s Hospital, Fudan University to determine the relationship between the TyG index and the incidence of HDP and adverse pregnancy outcomes. ### Results In the case-control study, the incidence of HDP was found to be significantly associated with the TyG index. Moreover, logistic regression indicated that the TyG index is an independent risk factor for HDP development and incidence of low birth weight (LBW) and fetal distress. In the cohort study, the results showed that the TyG index increased, there was a stepwise increase in HDP incidence, SBP, and DBP levels one week before delivery as well as in LBW and fetal distress incidence. The early trimester TyG index was positively associated with pre-pregnancy BMI, systolic blood pressure (SBP), and diastolic blood pressure (DBP) one week before delivery. Spline regression showed that there was a significant linear association between HDP incidence and early trimester TyG index when it was >8.5. ### Conclusions This work suggested that the early trimester TyG index was closely associated with the development of HDP and adverse pregnancy outcomes. ## Background Hypertensive disorder of pregnancy (HDP) is defined as a group of diseases coexisting with pregnancy and hypertension, including gestational hypertension, preeclampsia, eclampsia, chronic hypertension with preeclampsia, and chronic hypertension with pregnancy [1]. Except for chronic hypertension, HDP occurred mostly after 20 weeks of gestation and returned to normal within 12 weeks after delivery. Over the past few decades, HDP affect $5\%$–$10\%$ of pregnancies worldwide [2]. In China, HDP prevalence has been reported to be 5.55–$5.57\%$ in 2011 [3], and with the improvement of living standards and support for multiple births in recent years, the prevalence of HDP has increased to $7.2\%$ in 2021 [4]. As a group of common pregnancy complications, HDP can lead to many adverse pregnancy outcomes including preterm delivery, placental abruption, cesarean section, postpartum hemorrhage, and a higher incidence of low birth weight (LBW) and fetal distress [3]. Insulin resistance (IR) is a normal feature of pregnancy and becomes more severe as the pregnancy progresses. Previous studies have shown that IR plays a crucial role in the pathophysiology of HDP (5–9). The triglyceride-glucose index (TyG index), based on fasting glucose and triglycerides, has been shown as a simple and reliable surrogate measure to reflect IR compared with the euglycemic-hyperinsulinemic clamp (10–12), which is the ‘gold standard’ for evaluating IR. Also, compared with the Homeostatic Model Assessment of Insulin Resistance (HOMA-IR), the TyG index is more accessible in clinical practice. Moreover, recent studies have demonstrated that the TyG index presented a better performance than the HOMA-IR index in identifying patients with IR and was more strongly associated with arterial stiffness in patients with type 2 diabetes mellitus (T2DM) in comparison with the HOMA-IR index [13, 14]. In addition, Zheng et al. reported that the TyG index is an independent risk factor for hypertension and has a dose-dependent relationship, which can independently predict hypertension events [15]. To the best of our knowledge, the relationship between the TyG index and the incidence of HDP remains unclear. The purpose of this study was to investigate the association between the TyG index and the incidence of HDP and adverse pregnancy outcomes. ## Study population In this study, we recruited all women of Obstetrics at Shanghai Fifth People’s Hospital, Fudan University from January 2016 to December 2018, and there were 1400 pregnant women were retrospectively screened at digital medical record systems, the women had no other medical diagnosis at the beginning of pregnancy and were truly representative. Also, all the women were followed up until 12 weeks after delivery. The retrospective analysis process followed the procedure described in Figure 1. Women were excluded from the study for any of the following: [1] had a diagnosis of chronic hypertension before pregnancy or before 20 weeks’ gestation; [2] gestational diabetes mellitus; [3] multiple gestations or pregnancy by assisted reproductive technology; [4] serious liver dysfunction (alanine transaminase above 2.5 times upper limit) and renal dysfunction (estimated Glomerular Filtration Rate below 90 ml/min/1.73m²); [5] autoimmune diseases or malignant tumors; [6] participants with missing TyG index measurements or other data. In total, 1150 women (861 without HDP and 289 with HDP) were finally included in the analysis. The study protocol was approved by the Institutional Review Board of Shanghai Fifth People’s Hospital, Fudan University. Figure 1: Flowchart of the study. HDP, hypertensive disorder of pregnancy; TyG, Triglyceride–glucose. ## Data collection The health cards of all pregnant women were obtained from 9 to 12 weeks’ gestation and included information about age, occupation, last menstruation, method of conception, parity, obstetric history, family history of hypertension, and pre-pregnancy weight. Subsequently, at the first visit, blood pressure (BP), weight, height, blood count (Sysmex XN9000, Japan), biochemistry results (Cobas 8000, Roche, Switzerland), and fasting glucose were recorded. Fasting venous blood samples were performed after at least 8 hours of fast. The Triglyceride–glucose (TyG) index was calculated as ln[triglycerides (mg/dL) fasting glucose (mg/dL)/2]. BP was measured with an automated sphygmomanometer (HEM-7124) in the seated position on two occasions 4h apart after resting for at least 5 minutes. The elbow of the arm used to measure BP was supported at heart level. In the presence of raised BP, routine obstetric examination was performed every 2 to 4 weeks in the outpatient clinic until 34 weeks’ gestation and thereafter every week. Each women’s body weight and height were measured in light clothing without shoes. The weighing scale, height meter, and automated sphygmomanometers were calibrated every 6 months and all measurements were taken by the same outpatient nurse. Body mass index (BMI) was calculated as weight (kg)/height (m)². After delivery, details including gestational age at delivery, mode of delivery, newborn weight, sex of the neonate and adverse pregnancy outcomes including LBW, macrosomia, preterm birth, placental abruption, postpartum hemorrhage, cesarean section, rupture of membranes, and fetal distress were recorded by medical staff. ## Definitions HDP is a group of diseases coexisting with elevated blood pressure (systolic blood pressure (SBP) ≥ 140 mmHg and/or a diastolic blood pressure (DBP) ≥ 90 mmHg on two occasions at least 4h apart) during pregnancy [16]. Preterm birth is defined as that occurring after 28 weeks and before 37 completed weeks of gestation [17]. Low birth weight (LBW) is defined as that the neonates who were born with birth weights lower than 2500g [18]. Macrosomia was defined as birth weight more than 4000g [18]. Fetal distress is defined as a non-reassuring fetal status that the baby does not have the adequate amount of oxygen supply before labor, during the labor process or after the period of labor [19]. Postpartum hemorrhage is defined as losing more than 500 milliliters of blood 24 hours after vaginal birth, and more than 1000 milliliters of blood after a cesarean birth [20]. ## Outcomes The outcome was the incidence of HDP and adverse pregnancy outcome including LBW, macrosomia, preterm birth, placental abruption, postpartum hemorrhage, cesarean section, rupture of membranes, and fetal distress. ## Statistical analysis To avoid the confounding effects on BP between women with or without HDP, a propensity score matching (PSM) method was employed to match variables of age, pre-pregnancy BMI, family history of hypertension and parity. Matching tolerance was 0.02. We used Shapiro–Wilk test and the shape of the histogram to check the normality. Proportions (%) were used for categorical variables and mean and standard deviation (SD) or median and interquartile range (IQR) was used for continuous variables. χ2 test, the Student’s t-test and Mann-Whitney U test were used to identify the difference between groups. To determine whether TyG index was an independent risk factor, logistic regression analysis was performed with HDP classified in a binary manner (presence/absence) as the dependent variable. Receiver operating characteristic curves were performed to assess the predictability for HDP. To further validate the association of TyG index with HDP and pregnancy outcomes, a cohort study including all subjects was established in which patients were divided into three groups by tertiles of TyG index. χ2, Kruskal-Wallis and post hoc test were used to identify the difference in the mean between groups. Linear association between TyG index and pre-pregnancy BMI, SBP and DBP one week before delivery were assessed by simple linear regression analysis. Continuous association of TyG index with HDP incidence was determined by spline regression analysis. All analyses were performed using SPSS 26.0. A two-tailed P value < 0.05 was considered statistically significant. ## Characteristics of women with and without HDP in all subjects and subjects after PSM In this study, HDP developed in 289 women ($25.13\%$) among the 1150 subjects, and women with older age ($P \leq 0.001$), higher pre-pregnancy BMI ($$P \leq 0.002$$) and family history of hypertension ($P \leq 0.001$) were more likely to develop HDP (Table 1). At the first visit, compared with women without HDP, patients with HDP had a much higher level of TG (1.83 ± 0.83 vs. 1.62 ± 0.63 mmol/L, $P \leq 0.001$), FBG (4.20 ± 0.48 vs. 4.04 ± 0.40 mmol/L, $P \leq 0.001$), TyG index (8.63 ± 0.41 vs. 8.49 ± 0.36, $P \leq 0.001$) and WBC (8.96 ± 1.82 vs. 8.63 ± 1.98×10^9/L, $$P \leq 0.012$$), whereas the difference in TC or LDL was not significant. Also, there were no significant differences in SBP, DBP, ALT, AST, and creatin at baseline between the two groups. One week before delivery, women with HDP had a much higher SBP (123 ± 14 vs. 121 ± 13 mmHg, $$P \leq 0.022$$), DBP (77 ± 11 vs. 75 ± 11 mmHg, $$P \leq 0.026$$), and creatinine (50.12 ± 8.83 vs. 46.39 ± 7.40 μmol/L, $P \leq 0.001$). In addition, there were significant differences in weight gain (16.93 ± 4.72 vs. 15.79 ± 4.38 kg, $P \leq 0.001$) during the whole pregnancy between the two groups. For pregnancy outcome, obviously, mothers with HDP tended to deliver lower-weight newborns (3220.3 ± 619.8 vs. 3376.8 ± 417.5 g, $P \leq 0.001$), and had a higher rate of fetal distress ($10.0\%$ vs. $3.3\%$, $P \leq 0.001$), cesarean section ($59.5\%$ vs. $42.9\%$, $P \leq 0.001$), preterm ($10.7\%$ vs. $3.5\%$, $P \leq 0.001$), postpartum hemorrhage ($11.8\%$ vs. $3.6\%$, $P \leq 0.001$) and delivering low birth weight infants ($10.7\%$ vs. $2.1\%$, $P \leq 0.001$) than mothers without HDP (Table 1). Table 1* \| Unnamed: 0 \| All subjects \| All subjects.1 \| All subjects.2 \| All subjects.3 \| All subjects.4 \| After PSM \| After PSM.1 \| After PSM.2 \| After PSM.3 \| Unnamed: 10 \| Unnamed: 11 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| \| Women without HDP \| Women with HDP \| Women with HDP \| Women with HDP \| P \| Women without HDP \| Women with HDP \| Women with HDP \| P \| \| \| \| n \| 861 \| 861 \| 861 \| 289 \| 289 \| 289 \| \| 230 \| 230 \| 230 \| \| \| Age (years) \| 28.2 ± 5.6 \| 28.2 ± 5.6 \| 28.2 ± 5.6 \| 29.7 ± 5.3 \| 29.7 ± 5.3 \| 29.7 ± 5.3 \| <0.001 \| 28.7 ± 6.0 \| 28.7 ± 5.0 \| 28.7 ± 5.0 \| 0.933 \| \| Pre-pregnancy BMI (kg/m²) \| 22.0 ± 3.0 \| 22.0 ± 3.0 \| 22.0 ± 3.0 \| 22.7 ± 3.5 \| 22.7 ± 3.5 \| 22.7 ± 3.5 \| 0.002 \| 21.5 ± 2.8 \| 21.8 ± 3.0 \| 21.8 ± 3.0 \| 0.144 \| \| Family history of hypertension \| Family history of hypertension \| Family history of hypertension \| Family history of hypertension \| Family history of hypertension \| Family history of hypertension \| Family history of hypertension \| Family history of hypertension \| Family history of hypertension \| Family history of hypertension \| \| \| \| No \| 772(89.7) \| 772(89.7) \| 772(89.7) \| 230(79.6) \| 230(79.6) \| 230(79.6) \| <0.001 \| 199(86.5) \| 202(87.8) \| 202(87.8) \| 0.676 \| \| Yes \| 89(10.3) \| 89(10.3) \| 89(10.3) \| 59(20.4) \| 59(20.4) \| 59(20.4) \| \| 31(13.5) \| 28(12.2) \| 28(12.2) \| \| \| Parity \| Parity \| Parity \| Parity \| Parity \| Parity \| Parity \| Parity \| Parity \| Parity \| \| \| \| Nulliparous \| 352(40.9) \| 352(40.9) \| 352(40.9) \| 166(57.4) \| 166(57.4) \| 166(57.4) \| <0.001 \| 123(53.5) \| 113(49.1) \| 113(49.1) \| 0.351 \| \| Parous \| 509(59.1) \| 509(59.1) \| 509(59.1) \| 123(42.6) \| 123(42.6) \| 123(42.6) \| \| 107(46.5) \| 117(50.9) \| 117(50.9) \| \| \| At the first visit \| At the first visit \| At the first visit \| At the first visit \| At the first visit \| At the first visit \| At the first visit \| At the first visit \| At the first visit \| At the first visit \| \| \| \| SBP (mmHg) \| 116 ± 9 \| 116 ± 9 \| 116 ± 9 \| 116 ± 10 \| 116 ± 10 \| 116 ± 10 \| 0.977 \| 116 ± 8 \| 116 ± 6 \| 116 ± 6 \| 0.198 \| \| DBP (mmHg) \| 70 ± 8 \| 70 ± 8 \| 70 ± 8 \| 71 ± 8 \| 71 ± 8 \| 71 ± 8 \| 0.458 \| 71 ± 7 \| 71 ± 7 \| 71 ± 7 \| 0.840 \| \| ALT (units/L) \| 14.0(10.3-18.0) \| 14.0(10.3-18.0) \| 14.0(10.3-18.0) \| 15(10.0-22.6) \| 15(10.0-22.6) \| 15(10.0-22.6) \| 0.153 \| 14.0(11.0-18.0) \| 14.3(10.0-22.0) \| 14.3(10.0-22.0) \| 0.508 \| \| AST (units/L) \| 19.0(15.0-23.0) \| 19.0(15.0-23.0) \| 19.0(15.0-23.0) \| 19.4(15.0-25.0) \| 19.4(15.0-25.0) \| 19.4(15.0-25.0) \| 0.146 \| 19.0(15.2-24.0) \| 19.0(15.0-25.0) \| 19.0(15.0-25.0) \| 0.585 \| \| Creatinine (μmol/L) \| 44.15 ± 5.74 \| 44.15 ± 5.74 \| 44.15 ± 5.74 \| 44.47 ± 6.17 \| 44.47 ± 6.17 \| 44.47 ± 6.17 \| 0.421 \| 43.94 ± 5.93 \| 44.19 ± 6.41 \| 44.19 ± 6.41 \| 0.664 \| \| TC (mmol/L) \| 4.81 ± 0.87 \| 4.81 ± 0.87 \| 4.81 ± 0.87 \| 4.70 ± 0.84 \| 4.70 ± 0.84 \| 4.70 ± 0.84 \| 0.057 \| 4.80 ± 0.87 \| 4.69 ± 0.87 \| 4.69 ± 0.87 \| 0.159 \| \| TG (mmol/L) \| 1.62 ± 0.63 \| 1.62 ± 0.63 \| 1.62 ± 0.63 \| 1.83 ± 0.83 \| 1.83 ± 0.83 \| 1.83 ± 0.83 \| <0.001 \| 1.46 ± 0.54 \| 1.78 ± 0.72 \| 1.78 ± 0.72 \| <0.001 \| \| HDL (mmol/L) \| 1.94 ± 0.43 \| 1.94 ± 0.43 \| 1.94 ± 0.43 \| 1.80 ± 0.38 \| 1.80 ± 0.38 \| 1.80 ± 0.38 \| <0.001 \| 1.96 ± 0.42 \| 1.84 ± 0.37 \| 1.84 ± 0.37 \| 0.001 \| \| LDL (mmol/L) \| 2.73 ± 0.76 \| 2.73 ± 0.76 \| 2.73 ± 0.76 \| 2.64 ± 0.72 \| 2.64 ± 0.72 \| 2.64 ± 0.72 \| 0.072 \| 2.62 ± 0.78 \| 2.62 ± 0.77 \| 2.62 ± 0.77 \| 0.927 \| \| FBG (mmol/L) \| 4.04 ± 0.40 \| 4.04 ± 0.40 \| 4.04 ± 0.40 \| 4.20 ± 0.48 \| 4.20 ± 0.48 \| 4.20 ± 0.48 \| <0.001 \| 4.05 ± 0.39 \| 4.21 ± 0.49 \| 4.21 ± 0.49 \| <0.001 \| \| TyG index \| 8.49 ± 0.36 \| 8.49 ± 0.36 \| 8.49 ± 0.36 \| 8.63 ± 0.41 \| 8.63 ± 0.41 \| 8.63 ± 0.41 \| <0.001 \| 8.39 ± 0.35 \| 8.62 ± 0.39 \| 8.62 ± 0.39 \| <0.001 \| \| WBC (10^9/L) \| 8.63 ± 1.98 \| 8.63 ± 1.98 \| 8.63 ± 1.98 \| 8.96 ± 1.82 \| 8.96 ± 1.82 \| 8.96 ± 1.82 \| 0.012 \| 8.73 ± 1.83 \| 8.91 ± 1.83 \| 8.91 ± 1.83 \| 0.282 \| \| One week before delivery \| One week before delivery \| One week before delivery \| One week before delivery \| One week before delivery \| One week before delivery \| One week before delivery \| One week before delivery \| One week before delivery \| One week before delivery \| \| \| \| SBP (mmHg) \| 121 ± 13 \| 123 ± 14 \| 123 ± 14 \| 123 ± 14 \| 0.022 \| 120 ± 6 \| 135 ± 12 \| 135 ± 12 \| <0.001 \| \| \| \| DBP (mmHg) \| 75 ± 11 \| 77 ± 11 \| 77 ± 11 \| 77 ± 11 \| 0.026 \| 75 ± 6 \| 88 ± 8 \| 88 ± 8 \| <0.001 \| \| \| \| ALT (units/L) \| 8.7(7.0-11.0) \| 9.0(7.0-12.1) \| 9.0(7.0-12.1) \| 9.0(7.0-12.1) \| 0.06 \| 9.0(7.0-11.0) \| 9.0(7.0-12.0) \| 9.0(7.0-12.0) \| 0.237 \| \| \| \| Creatinine (μmol/L) \| 46.39 ± 7.40 \| 50.12 ± 8.83 \| 50.12 ± 8.83 \| 50.12 ± 8.83 \| <0.001 \| 46.82 ± 7.00 \| 50.43 ± 9.04 \| 50.43 ± 9.04 \| <0.001 \| \| \| \| TC (mmol/L) \| 6.12 ± 1.11 \| 6.16 ± 1.22 \| 6.16 ± 1.22 \| 6.16 ± 1.22 \| 0.633 \| 6.12 ± 1.10 \| 6.20 ± 1.25 \| 6.20 ± 1.25 \| 0.494 \| \| \| \| TG (mmol/L) \| 3.04 ± 0.97 \| 3.63 ± 1.26 \| 3.63 ± 1.26 \| 3.63 ± 1.26 \| <0.001 \| 3.02 ± 0.98 \| 3.57 ± 1.20 \| 3.57 ± 1.20 \| <0.001 \| \| \| \| HDL (mmol/L) \| 1.87 ± 0.38 \| 1.81 ± 0.43 \| 1.81 ± 0.43 \| 1.81 ± 0.43 \| 0.035 \| 1.85 ± 0.39 \| 1.83 ± 0.43 \| 1.83 ± 0.43 \| 0.562 \| \| \| \| LDL (mmol/L) \| 3.64 ± 1.04 \| 3.54 ± 1.03 \| 3.54 ± 1.03 \| 3.54 ± 1.03 \| 0.252 \| 3.64 ± 1.03 \| 3.58 ± 1.02 \| 3.58 ± 1.02 \| 0.542 \| \| \| \| FBG (mmol/L) \| 4.01 ± 0.47 \| 4.15 ± 0.59 \| 4.15 ± 0.59 \| 4.15 ± 0.59 \| 0.001 \| 4.00 ± 0.43 \| 4.13 ± 0.57 \| 4.13 ± 0.57 \| 0.003 \| \| \| \| TyG index \| 9.13 ± 0.34 \| 9.33 ± 0.35 \| 9.33 ± 0.35 \| 9.33 ± 0.35 \| <0.001 \| 9.12 ± 0.34 \| 9.31 ± 0.35 \| 9.31 ± 0.35 \| <0.001 \| \| \| \| WBC (10^9/L) \| 9.12 ± 2.34 \| 9.09 ± 2.05 \| 9.09 ± 2.05 \| 9.09 ± 2.05 \| 0.845 \| 8.92 ± 2.19 \| 9.04 ± 2.09 \| 9.04 ± 2.09 \| 0.544 \| \| \| \| Weight gain (kg) \| 15.79 ± 4.38 \| 16.93 ± 4.72 \| 16.93 ± 4.72 \| 16.93 ± 4.72 \| <0.001 \| 15.82 ± 4.59 \| 16.92 ± 4.59 \| 16.92 ± 4.59 \| 0.011 \| \| \| \| \| All subjects \| All subjects \| All subjects \| All subjects \| All subjects \| After PSM \| After PSM \| After PSM \| After PSM \| \| \| \| \| Women without HDP \| Women with HDP \| Women with HDP \| P \| P \| Women without HDP \| Women without HDP \| Women with HDP \| P \| \| \| \| n \| 861 \| 289 \| 289 \| 289 \| \| 230 \| 230 \| 230 \| \| \| \| \| Pregnancy outcome \| Pregnancy outcome \| Pregnancy outcome \| Pregnancy outcome \| Pregnancy outcome \| Pregnancy outcome \| Pregnancy outcome \| Pregnancy outcome \| Pregnancy outcome \| Pregnancy outcome \| \| \| \| Fetus sex (female) \| 422(49.9) \| 156(54.0) \| 156(54.0) \| 156(54.0) \| 0.229 \| 117(50.9) \| 127(55.2) \| 127(55.2) \| 0.350 \| \| \| \| Newborn weight (g) \| 3376.8 ± 417.5 \| 3220.3 ± 619.8 \| 3220.3 ± 619.8 \| 3220.3 ± 619.8 \| <0.001 \| 3384.8 ± 395.4 \| 3158.4 ± 623.2 \| 3158.4 ± 623.2 \| <0.001 \| \| \| \| LBW \| 18(2.1) \| 31(10.7) \| 31(10.7) \| 31(10.7) \| <0.001 \| 10(4.3) \| 25(10.9) \| 25(10.9) \| 0.008 \| \| \| \| Macrosomia \| 47(5.5) \| 18(6.2) \| 18(6.2) \| 18(6.2) \| 0.624 \| 9(3.9) \| 10(4.3) \| 10(4.3) \| 0.815 \| \| \| \| Fetal distress \| 28(3.3) \| 29(10.0) \| 29(10.0) \| 29(10.0) \| <0.001 \| 9(3.9) \| 23(10.0) \| 23(10.0) \| 0.010 \| \| \| \| Cesarean section \| 363(42.9) \| 172(59.5) \| 172(59.5) \| 172(59.5) \| <0.001 \| 93(40.4) \| 128(55.7) \| 128(55.7) \| <0.001 \| \| \| \| Preterm \| 30(3.5) \| 31(10.7) \| 31(10.7) \| 31(10.7) \| <0.001 \| 11(4.8) \| 23(10.0) \| 23(10.0) \| 0.032 \| \| \| \| Rupture of membranes \| 102(11.8) \| 22(7.6) \| 22(7.6) \| 22(7.6) \| 0.058 \| 23(10.0) \| 16(7.0) \| 16(7.0) \| 0.241 \| \| \| \| Placental abruption \| 3(0.4) \| 4(1.4) \| 4(1.4) \| 4(1.4) \| 0.071 \| 0.0(0.0) \| 3(1.3) \| 3(1.3) \| 0.247 \| \| \| \| Postpartum hemorrhage \| 31(3.6) \| 34(11.8) \| 34(11.8) \| 34(11.8) \| <0.001 \| 21(9.1) \| 26(11.3) \| 26(11.3) \| 0.441 \| \| \| To eliminate confounding effects on BP, a PSM method was employed to match variables of age, pre-pregnancy BMI, family history of hypertension, and parity. After matching, there were no significant differences in WBC at the first visit, HDL one week before delivery, and incidence of postpartum hemorrhage between women with and without HDP. Nonetheless, there remained a significantly higher TG ($P \leq 0.001$), FBG ($P \leq 0$,001), and TyG index ($P \leq 0.001$) at baseline, higher SBP ($P \leq 0.001$), DBP ($P \leq 0.001$), and creatinine ($P \leq 0.001$) one week before delivery and higher incidence of LBW ($$P \leq 0.008$$), fetal distress ($$P \leq 0.010$$), cesarean section ($P \leq 0.001$) and preterm ($$P \leq 0.032$$) in women with HDP compared with control subjects (Table 1). ## The early trimester TyG index was an independent risk factor and diagnostic predictive factor for HDP development, LBW and fetal distress In case-control study, we used multiple logistic regression to analyze the odds ratio (OR) of HDP incidence in each TyG group for all subjects. After adjusted by age, pre-pregnancy BMI, family history of hypertension, parity and weight gain, compared to the lowest tertile of the TyG index, the OR for subjects in tertile 2 and tertile 3 were 1.46 ($95\%$CI: 1.02-2.09; $$P \leq 0.040$$) and 2.44 ($95\%$CI: 1.71-3.48; $P \leq 0.001$) respectively (Table 2). The same results were showed in the logistic regression for the subjects after PSM (Supplementary Table 1). All of these showed that early trimester TyG index as an independent risk factor was closely correlated with the development of HDP. The ROC curves of the TyG index as a marker to predict the incidence of HDP are illustrated in Figure 2. Clinical model include age, pre-pregnancy BMI, family history of hypertension, and parity. The AUC of the TyG index for predicting the occurrence of HDP was 0.684 ($95\%$ CI: 0.647–0.721), but the clinical model without TyG index for predicting was 0.657 ($95\%$ CI: 0.620-0.695), adding TyG index to the model significantly increased its discriminatory capacity (AUC: 0.725; $95\%$ CI: 0.690-0.760). Besides, TyG index was also an independent risk factor for the incidence of LBW and fetal stress. After adjusted by age, pre-pregnancy BMI, family history of hypertension, parity and weight gain, compared to the lowest tertile of the TyG index, the OR for incident LBW and fetal distress in tertile 3 were 2.59 ($95\%$CI: 1.25-5.33; $$P \leq 0.010$$) and 2.92 ($95\%$CI: 1.40-6.10; $$P \leq 0.004$$) (Supplementary Table 2). ## Comparison of parameters during pregnancy among three groups categorized by tertiles of the early trimester TyG index in the cohort study All subjects were divided into three groups according to tertiles of the early trimester TyG index: lowest group (<8.35), middle group (8.35-8.70), and highest group (>8.70). There was a stepwise increase in the incidence of HDP ($18.9\%$, $23.6\%$, and $33.4\%$; $P \leq 0.001$), age (28.1 ± 4.8 vs. 28.6 ± 5.3 vs. 29.1 ± 5.3 years; $$P \leq 0.038$$), pre-pregnancy BMI (21.5 ± 2.9 vs. 22.1 ± 3.0 vs. 22.8 ± 3.3 kg/m²; $P \leq 0.001$), DBP (70 ± 8 vs. 70 ± 8 vs. 71 ± 7 mmHg; $$P \leq 0.003$$), TC (4.54 ± 0.78 vs. 4.77 ± 0.81 vs. 5.05 ± 0.93 mmol/L; $P \leq 0.001$), TG (1.10 ± 0.19 vs. 1.55 ± 0.20 vs. 2.41 ± 0.70 mmol/L; $P \leq 0.001$), HDL (2.00 ± 0.41 vs. 1.93 ± 0.41 vs. 1.77 ± 0.41 mmol/L; $P \leq 0.001$), LDL (2.49 ± 0.65 vs. 2.69 ± 0.70 vs. 2.95 ± 0.83 mmol/L; $P \leq 0.001$), FBG (3.98 ± 0.39 vs. 4.10 ± 0.40 vs. 4.18 ± 0.47 mmol/L; $P \leq 0.001$) and WBC (8.40 ± 1.87 vs. 8.65 ± 2.01 vs. 9.12 ± 1.90 10^9/L; $P \leq 0.001$) at the first visit, level of SBP (116 ± 12 vs. 118 ± 9 vs. 132 ± 13 mmHg; $P \leq 0.001$), DBP (70 ± 9 vs. 73 ± 8 vs. 85 ± 10 mmHg; $P \leq 0.001$), TG (2.85 ± 0.94 vs. 3.26 ± 0.96 vs. 3.89 ± 1.30 mmol/L; $P \leq 0.001$), HDL (1.93 ± 0.42 vs. 1.84 ± 0.40 vs. 1.75 ± 0.37 mmol/L; $P \leq 0.001$) and FBG (3.98 ± 0.44 vs. 4.11 ± 0.51 vs. 4.15 ± 0.63 mmol/L; $$P \leq 0.002$$) one week before delivery. Likewise, the incidence of LBW($3.1\%$, $3.1\%$, and $6.8\%$, $$P \leq 0.014$$), fetal distress ($2.8\%$, $5.1\%$, and $7.1\%$; $$P \leq 0.025$$), cesarean section ($29.4\%$, $46.5\%$, and $67.8\%$, $P \leq 0.001$), preterm ($4.8\%$, $3.6\%$ and $7.6\%$, $$P \leq 0.042$$) and placental abruption ($0.0\%$, $0.5\%$ and $1.4\%$, $$P \leq 0.028$$) increase as TyG index increased (Table 3). Table 3* \| Unnamed: 0 \| Tertile 1 \| Tertile 2 \| Tertile 3 \| P \| \| --- \| --- \| --- \| --- \| --- \| \| TyG index range \| <8.35 \| 8.35-8.70 \| >8.70 \| \| \| N \| 392 \| 390 \| 368 \| \| \| Women with HDP \| 74(18.9) \| 92(23.6) \| 123(33.4) \| <0.001 \| \| Age (years) \| 28.1 ± 4.8 \| 28.6 ± 5.3 \| 29.1 ± 5.3† \| 0.038 \| \| Pre-pregnancy BMI (kg/m²) \| 21.5 ± 2.9 \| 22.1 ± 3.0* \| 22.8 ± 3.3†# \| <0.001 \| \| Family history of hypertension \| 30(7.6) \| 51(13.1) \| 72(20.5) \| <0.001 \| \| Parity (Nulliparous) \| 211(53.8) \| 169(43.3) \| 138(37.5) \| <0.001 \| \| At the first visit \| At the first visit \| At the first visit \| At the first visit \| At the first visit \| \| SBP (mmHg) \| 116 ± 11 \| 115 ± 10 \| 116 ± 7 \| 0.115 \| \| DBP (mmHg) \| 70 ± 8 \| 70 ± 8 \| 71 ± 7† \| 0.003 \| \| ALT (units/L) \| 13.2(10.0-18.8) \| 14.0(10.0-20.9) \| 13.1(9.0-20.8) \| 0.507 \| \| AST (units/L) \| 18.0(15.0-23.0) \| 19.0(15.0-24.0) \| 19.0(15.0-24.0) \| 0.471 \| \| Creatinine (μmol/L) \| 43.99 ± 5.69 \| 44.39 ± 6.09 \| 44.32 ± 5.76 \| 0.754 \| \| TC (mmol/L) \| 4.54 ± 0.78 \| 4.77 ± 0.81* \| 5.05 ± 0.93†# \| <0.001 \| \| TG (mmol/L) \| 1.10 ± 0.19 \| 1.55 ± 0.20* \| 2.41 ± 0.70†# \| <0.001 \| \| HDL (mmol/L) \| 2.00 ± 0.41 \| 1.93 ± 0.41 \| 1.77 ± 0.41†# \| <0.001 \| \| LDL (mmol/L) \| 2.49 ± 0.65 \| 2.69 ± 0.70* \| 2.95 ± 0.83†# \| <0.001 \| \| FBG (mmol/L) \| 3.98 ± 0.39 \| 4.10 ± 0.40* \| 4.18 ± 0.47†# \| <0.001 \| \| WBC (10^9/L) \| 8.40 ± 1.87 \| 8.65 ± 2.01 \| 9.12 ± 1.90†# \| <0.001 \| \| One week before delivery \| One week before delivery \| One week before delivery \| One week before delivery \| One week before delivery \| \| SBP (mmHg) \| 116 ± 12 \| 118 ± 9 \| 132 ± 13†# \| <0.001 \| \| DBP (mmHg) \| 70 ± 9 \| 73 ± 8* \| 85 ± 10†# \| <0.001 \| \| ALT (units/L) \| 11.0(8.0-18.0) \| 11.0(7.7-19.3) \| 11.0(7.7-19.0) \| 0.928 \| \| Creatinine (μmol/L) \| 47.42 ± 8.01 \| 47.32 ± 7.84 \| 47.23 ± 8.03 \| 0.876 \| \| TC (mmol/L) \| 6.12 ± 1.17 \| 6.12 ± 1.08 \| 6.18 ± 1.16 \| 0.840 \| \| TG (mmol/L) \| 2.85 ± 0.94 \| 3.26 ± 0.96* \| 3.89 ± 1.30†# \| <0.001 \| \| HDL (mmol/L) \| 1.93 ± 0.42 \| 1.84 ± 0.40 \| 1.75 ± 0.37†# \| <0.001 \| \| LDL (mmol/L) \| 3.66 ± 0.99 \| 3.59 ± 1.04 \| 3.52 ± 1.07 \| 0.389 \| \| FBG (mmol/L) \| 3.98 ± 0.44 \| 4.11 ± 0.51* \| 4.15 ± 0.63† \| 0.002 \| \| WBC (10^9/L) \| 9.19 ± 2.25 \| 9.01 ± 2.39 \| 9.14 ± 2.16 \| 0.500 \| \| Weight gain \| 16.42 ± 4.36 \| 15.82 ± 4.58 \| 16.07 ± 4.50 \| 0.166 \| \| Pregnancy outcome \| Pregnancy outcome \| Pregnancy outcome \| Pregnancy outcome \| Pregnancy outcome \| \| Fetus sex (female) \| 197(49.9) \| 183(47.0) \| 185(52.7) \| 0.306 \| \| Newborn weight (g) \| 3314.7 ± 433.2 \| 3356.6 ± 440.3 \| 3337.6 ± 558.4 \| 0.327 \| \| LBW \| 12(3.1) \| 12(3.1) \| 25(6.8) \| 0.014 \| \| Macrosomia \| 15(3.8) \| 21(5.4) \| 29(7.9) \| 0.052 \| \| Fetal distress \| 11(2.8) \| 20(5.1) \| 26(7.1) \| 0.025 \| \| Cesarean section \| 116(29.4) \| 181(46.5) \| 238(67.8) \| <0.001 \| \| Preterm \| 19(4.8) \| 14(3.6) \| 28(7.6) \| 0.042 \| \| Rupture of membranes \| 47(12.0) \| 40(10.3) \| 37(10.1) \| 0.635 \| \| Placental abruption \| 0(0.0) \| 2(0.5) \| 5(1.4) \| 0.028 \| \| Postpartum hemorrhage \| 23(5.9) \| 15(3.8) \| 27(7.3) \| 0.112 \| ## The early trimester TyG index was closely associated with pre-pregnancy BMI, SBP and DBP one week before delivery To investigate the correlation between the early trimester TyG index and pre-pregnancy BMI, SBP or DBP one week before delivery, correlation analysis was performed. Simple linear regression analyses were performed to determine the association of the early trimester TyG index with pre-pregnancy BMI, SBP and DBP one week before delivery. There was a significant and moderate linear association for TyG index with pre-pregnancy BMI [β=0.18; $t = 6.35$; $P \leq 0.001$] (Figure 3A), SBP [β=0.47; $t = 17.78$; $P \leq 0.001$] (Figure 3B), and DBP [β=0.51; $t = 20.00$; $P \leq 0.001$] (Figure 3C). Figure 3:* Simple linear regression analysis between the TyG index at first visit and pre-pregnancy BMI, SBP and DBP one week before delivery. The TyG index showed a significant and moderate linear association with pre-pregnancy BMI (β=0.31; F=65.21; adjusted R²=0.09; P<0.001) (A), SBP (β=0.25; F=41.48; adjusted R²=0.06; P<0.001) (B), and DBP (β=0.38; F=106.11; adjusted R²=0.14; P<0.001) (C). SBP, systolic blood pressure; DBP, diastolic blood pressure. ## The early trimester TyG index was closely associated with the incidence of HDP After adjusting for age, pre-pregnancy BMI, family history of hypertension and parity, a spline model showed a significant relationship between continuous early trimester TyG index and incidence of HDP. We found an increasing trend of incidence of HDP with a higher TyG index despite the lack of a linear relationship between the TyG index and the incidence of HDP. The risk of developing HDP increased when the TyG index was greater than 8.5 (Figure 4). Figure 4: Continuous association of the TyG index at the first visit with the incidence of HDP. Adjusted for age, pre-pregnancy BMI, family history of hypertension and parity. The risk of developing HDP increased when the TyG index was greater than 8.5. ## Discussion This retrospective study is the one to confirm the significant positive association and a dose-response relation between the early trimester TyG index and the incidence of HDP in a large sample size. Furthermore, the results provide evidence that the early trimester TyG index is independently associated with the incidence of LBW and fetal distress. This study elucidated the substantial role of TyG index in predicting the development of HDP and adverse pregnancy outcome. Pregnancy is characterized by a number of metabolic adaptations. Insulin resistance (IR) is a normal characteristic of pregnancy and increases physiologically as the pregnancy progresses to support the normal fetal development and growth [21]. Some studies demonstrated the mechanism of IR in pregnancy, Daniela et al. showed that some steroid hormones which are elevated in pregnancy, such as progesterone and corticosteroids, contribute to impaired insulin sensitivity and glucose tolerance [22], Marilyn et al. showed that some cytokines and hormones secreted by the placenta including leptin and TNFα could be implicated in IR during pregnancy [21]. Some markers correlate with IR, such as triglycerides, free fatty acids, small dense LDL particles and PAI-1, increase as normal pregnancy and associated IR progress [23, 24]. Moreover, a growing number of studies have described the central role of IR in development of HDP. In a study using the euglycemic clamp technique, compared with controls, women with gestational hypertension exhibited approximately $40\%$ lower steady-state insulin sensitivity index and approximately $33\%$ higher mean plasma TG [25]. IR may increase gestational blood pressure through activation of the sympathetic nervous system [8, 26], sodium reabsorption by the distal nephron segments [6], enhanced vascular resistance and endothelial dysfunction [26, 27]. Furthermore, the recognition that features of insulin resistance persists many years after pregnancy among women which raises the risk for future cardiovascular disease (28–30). These observations suggest that interventions to reduce insulin resistance may reduce the risk of both hypertension in pregnancy and later life cardiovascular complications. The TyG index, the product of fasting glucose and triglycerides, which has high sensitivity ($96.5\%$) and specificity ($85.0\%$; AUC: 0.858) for diagnosis of IR [31], is a simple, reliable and early marker of IR, and can be widely used in clinical practice especially in primary hospital because all clinical laboratories can measure triglycerides and glucose and quantification of insulin levels is not required. Besides, Ana Carolina et al. showed that the TyG index had a slightly better performance compared with the HOMA-IR in identifying patients with IR [13]. Although there were few studies on the relationship between the TyG index and HDP, in the study of the TyG index and hypertension, Yi Wang et al. showed that compared with those with the lowest category of T yG index, subjects with the highest category of TyG index were associated with higher odds of hypertension [32], the reason for its lower risk ratio than in our study may be that the degree of IR is more pronounced during pregnancy. In addition, in a longitudinal population-based study, Zheng et al. revealed that the TyG index can predict the incidence of hypertension among the Chinese population [15], after 9 years follow up, compared with the lowest TyG group, the hazard ratios for subjects in quartile 2, quartile 3 and quartile 4 increased, and were statistically significant. And numerous studies have also found that the TyG index was closely associated with hypertension (33–35). A retrospective cohort study showed that High maternal triglyceride level had higher risks of HDP in all maternal FPG strata, and both the early-pregnancy FPG and mTG levels should be screened among overall population including the low-risk population to reduce the incidence of pregnancy complications [36]. The TyG index, which is the product of FBG and TG, may have a more obvious association with HDP. In our investigation, to avoid confounding, a PSM method was applied and revealed that women with HDP had a much higher level of the early trimester TyG index. We also confirmed that the TyG index was independently associated with the incidence of HDP and was a diagnostic predictive factor for HDP development in case-control study. Further analysis of the relationship between the early trimester TyG index and HDP was performed in the cohort study. We found the incidence of HDP increased progressively with the increase of the early trimester TyG index and the risk of developing HDP increased when the TyG index was greater than 8.5. All of these indicated a close relation between the early trimester TyG index and HDP development. Another finding in our study was that the TyG index was also associated with adverse pregnancy outcome. In the case-control study, we found that a higher TyG index was an independent risk factor for LBW and fetal distress. In the cohort study, women with the highest tertile of the TyG index had the highest risk for cesarean section, preterm and placental abruption. Maternal hyperglycemia and obesity predispose offspring to metabolic dysfunction [37], although the underlying mechanism is elusive, and the fetal hyperglycemia and IR disrupt normal surfactant synthesis and function, which may lead to adverse outcome in neonates [38]. In addition, the women with higher blood pressure have decreased uteroplacental blood flow, which obstructs fetal growth and exacerbates fetal hypoxia and ischemia [39], thereby increases the risk of fetal distress and the need for timely termination of pregnancy. There were some limitations to our study. First, all subjects were derived from one center, which may have led to biased results. Second, we did not measure dietary sodium intake and physical activity which may affect the blood pressure or lipids. Third, this is a retrospective study, due to the lack of information, fasting insulin was not obtained. Also, the TyG index is associated with gestational diabetes and the incidence of post-partum diabetes in women, and we are collecting data on women with gestational diabetes, we would like to explore the relationship between gestational diabetes and HDP once we have data. Despite these limitations, this study first demonstrated the association between the TyG index and HDP development. ## Conclusions In conclusion, our study suggested that the early trimester TyG index was closely associated with HDP development and adverse pregnancy outcomes. Besides, the TyG index could be a novel and clinically effective indicator for identifying the risk of HDP especially in primary hospital. ## 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 Medical Ethics Committee of Shanghai Fifth People’s Hospital Affiliated to Fudan University. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements. ## Author contributions YP and GZ designed the study. YP, SZ, ZW, CY and XW collected the data. YP, RD and HG analyzed the data. YP wrote the main manuscript text. GZ and YX contributed to the refinement of 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/fendo.2023.1093991/full#supplementary-material ## References 1. Magee LA, Smith GN, Bloch C, Côté AM, Jain V, Nerenberg K. Guideline no. 426: Hypertensive disorders of pregnancy: Diagnosis, prediction, prevention, and management. J obstet gynaecol Canada JOGC = J d'obstetrique gynecol du Canada JOGC. (2022) 44 547-71.e1. DOI: 10.1016/j.jogc.2022.03.002 2. Cífková R. Cardiovascular sequels of hypertension in pregnancy. J Am Heart Assoc (2018) 7. DOI: 10.1161/JAHA.118.009300 3. Ye C, Ruan Y, Zou L, Li G, Li C, Chen Y. The 2011 survey on hypertensive disorders of pregnancy (HDP) in China: prevalence, risk factors, complications, pregnancy and perinatal outcomes. PloS One (2014) 9. DOI: 10.1371/journal.pone.0100180 4. Li F, Qin J, Zhang S, Chen L. Prevalence of hypertensive disorders in pregnancy in China: A systematic review and meta-analysis. Pregnancy hypertens (2021) 24 13-21. DOI: 10.1016/j.preghy.2021.02.001 5. Rowe JW, Young JB, Minaker KL, Stevens AL, Pallotta J, Landsberg L. Effect of insulin and glucose infusions on sympathetic nervous system activity in normal man. Diabetes (1981) 30. DOI: 10.2337/diab.30.3.219 6. Doria A, Fioretto P, Avogaro A, Carraro A, Morocutti A, Trevisan R. Insulin resistance is associated with high sodium-lithium countertransport in essential hypertension. Am J Physiol (1991) 261. DOI: 10.1152/ajpendo.1991.261.6.E684 7. Gibbons GH, Dzau VJ. The emerging concept of vascular remodeling. New Engl J Med (1994) 330. DOI: 10.1056/NEJM199405193302008 8. Reaven GM, Lithell H, Landsberg L. Hypertension and associated metabolic abnormalities–the role of insulin resistance and the sympathoadrenal system. New Engl J Med (1996) 334. DOI: 10.1056/NEJM199602083340607 9. Seely EW, Solomon CG. Insulin resistance and its potential role in pregnancy-induced hypertension. J Clin Endocrinol Metab (2003) 88. DOI: 10.1210/jc.2003-030241 10. Li X, Li G, Cheng T, Liu J, Song G, Ma H. Association between triglyceride-glucose index and risk of incident diabetes: a secondary analysis based on a Chinese cohort study : TyG index and incident diabetes. Lipids Health Dis (2020) 19 236. DOI: 10.1186/s12944-020-01403-7 11. Xuan X, Hamaguchi M, Cao Q, Okamura T, Hashimoto Y, Obora A. U-Shaped association between the triglyceride-glucose index and the risk of incident diabetes in people with normal glycemic level: A population-base longitudinal cohort study. Clin Nutr (Edinburgh Scotland) (2021) 40. DOI: 10.1016/j.clnu.2021.02.037 12. Pranata R, Huang I, Irvan MA, Lim R. The association between triglyceride-glucose index and the incidence of type 2 diabetes mellitus-a systematic review and dose-response meta-analysis of cohort studies. Endocrine (2021) 74. DOI: 10.1007/s12020-021-02780-4 13. Vasques AC, Novaes FS, de Oliveira Mda S, Souza JR, Yamanaka A, Pareja JC. TyG index performs better than HOMA in a Brazilian population: a hyperglycemic clamp validated study. Diabetes Res Clin Pract (2011) 93 e98-e100. DOI: 10.1016/j.diabres.2011.05.030 14. Wang S, Shi J, Peng Y, Fang Q, Mu Q, Gu W. Stronger association of triglyceride glucose index than the HOMA-IR with arterial stiffness in patients with type 2 diabetes: a real-world single-centre study. Cardiovasc Diabetol (2021) 20 82. DOI: 10.1186/s12933-021-01274-x 15. Zheng R, Mao Y. Triglyceride and glucose (TyG) index as a predictor of incident hypertension: a 9-year longitudinal population-based study. Lipids Health Dis (2017) 16 175. DOI: 10.1186/s12944-017-0562-y 16. Bethesda Maryland. Report of the national high blood pressure education program working group on high blood pressure in pregnancy. Am J obstet gynecol (2000) 183 S1-s22. DOI: 10.1067/mob.2000.107928 17. Goldenberg RL, Culhane JF, Iams JD, Romero R. Epidemiology and causes of preterm birth. Lancet (London England) (2008) 371 75-84. DOI: 10.1016/S0140-6736(08)60074-4 18. Lewandowska M. Maternal obesity and risk of low birth weight, fetal growth restriction, and macrosomia: Multiple analyses. Nutrients (2021) 13. DOI: 10.3390/nu13041213 19. van Geijn HP, Copray FJ, Donkers DK, Bos MH. Diagnosis and management of intrapartum fetal distress. Eur J obstet gynecol Reprod Biol (1991) 42 20. Sentilhes L, Merlot B, Madar H, Sztark F, Brun S, Deneux-Tharaux C. Postpartum haemorrhage: prevention and treatment. Expert review of hematology (2016) 9. DOI: 10.1080/17474086.2016.1245135 21. Lacroix M, Kina E, Hivert MF. Maternal/fetal determinants of insulin resistance in women during pregnancy and in offspring over life. Curr Diabetes Rep (2013) 13. DOI: 10.1007/s11892-012-0360-x 22. Vejrazkova D, Vcelak J, Vankova M, Lukasova P, Bradnova O, Halkova T. Steroids and insulin resistance in pregnancy. J Steroid Biochem Mol Biol (2014) 139. DOI: 10.1016/j.jsbmb.2012.11.007 23. Sattar N, Greer IA, Rumley A, Stewart G, Shepherd J, Packard CJ. A longitudinal study of the relationships between haemostatic, lipid, and oestradiol changes during normal human pregnancy. Thromb haemost (1999) 81 24. Belo L, Caslake M, Gaffney D, Santos-Silva A, Pereira-Leite L, Quintanilha A. Changes in LDL size and HDL concentration in normal and preeclamptic pregnancies. Atherosclerosis (2002) 162. DOI: 10.1016/S0021-9150(01)00734-1 25. Caruso A, Ferrazzani S, De Carolis S, Lucchese A, Lanzone A, De Santis L. Gestational hypertension but not pre-eclampsia is associated with insulin resistance syndrome characteristics. Hum Reprod (Oxford England) (1999) 14. DOI: 10.1093/humrep/14.1.219 26. Fonseca VA. Insulin resistance, diabetes, hypertension, and renin-angiotensin system inhibition: reducing risk for cardiovascular disease. J Clin hypertens (Greenwich Conn) (2006) 8 713-20; quiz 21-2. DOI: 10.1111/j.1524-6175.2006.05583.x 27. Smith MM, Minson CT. Obesity and adipokines: effects on sympathetic overactivity. J Physiol (2012) 590. DOI: 10.1113/jphysiol.2011.221036 28. Jarvie JL, Metz TD, Davis MB, Ehrig JC, Kao DP. Short-term risk of cardiovascular readmission following a hypertensive disorder of pregnancy. Heart (British Cardiac Society) (2018) 104. DOI: 10.1136/heartjnl-2017-312299 29. Arnott C, Nelson M, Alfaro Ramirez M, Hyett J, Gale M, Henry A. Maternal cardiovascular risk after hypertensive disorder of pregnancy. Heart (British Cardiac Society) (2020) 106. DOI: 10.1136/heartjnl-2020-316541 30. Garovic VD, White WM, Vaughan L, Saiki M, Parashuram S, Garcia-Valencia O. Incidence and long-term outcomes of hypertensive disorders of pregnancy. J Am Coll Cardiol (2020) 75. DOI: 10.1016/j.jacc.2020.03.028 31. Guerrero-Romero F, Simental-Mendía LE, González-Ortiz M, Martínez-Abundis E, Ramos-Zavala MG, Hernández-González SO. The product of triglycerides and glucose, a simple measure of insulin sensitivity. comparison with the euglycemic-hyperinsulinemic clamp. J Clin Endocrinol Metab (2010) 95. DOI: 10.1210/jc.2010-0288 32. Wang Y, Yang W, Jiang X. Association between triglyceride-glucose index and hypertension: A meta-analysis. Front Cardiovasc Med (2021) 8. DOI: 10.3389/fcvm.2021.644035 33. Jian S, Su-Mei N, Xue C, Jie Z, Xue-Sen W. Association and interaction between triglyceride-glucose index and obesity on risk of hypertension in middle-aged and elderly adults. Clin Exp hypertens (New York NY 1993) (2017) 39. DOI: 10.1080/10641963.2017.1324477 34. Wang K, He G, Zhang Y, Yin J, Yan Y, Zhang Y. Association of triglyceride-glucose index and its interaction with obesity on hypertension risk in Chinese: a population-based study. J Hum hypertens (2021) 35. DOI: 10.1038/s41371-020-0326-4 35. Wang RN, Zhang DS, Bai Z, Yin C, Zhang R, Yang JL. [Prospective cohort study of relationship of triglyceride, fasting blood-glucose and triglyceride glucose product index with risk of hypertension]. Zhonghua liu xing bing xue za zhi = Zhonghua liuxingbingxue zazhi (2021) 42. DOI: 10.3760/cma.j.cn112338-20200401-00491 36. Wu D, Zhang J, Xiong Y, Wang H, Lu D, Guo M. Effect of maternal glucose and triglyceride levels during early pregnancy on pregnancy outcomes: A retrospective cohort study. Nutrients (2022) 14. DOI: 10.3390/nu14163295 37. Oben JA, Mouralidarane A, Samuelsson AM, Matthews PJ, Morgan ML, McKee C. Maternal obesity during pregnancy and lactation programs the development of offspring non-alcoholic fatty liver disease in mice. J Hepatol (2010) 52. DOI: 10.1016/j.jhep.2009.12.042 38. Yildiz Atar H, Baatz JE, Ryan RM. Molecular mechanisms of maternal diabetes effects on fetal and neonatal surfactant. Children (Basel Switzerland) (2021) 8. DOI: 10.3390/children8040281 39. Berhe AK, Ilesanmi AO, Aimakhu CO, Mulugeta A. Effect of pregnancy induced hypertension on adverse perinatal outcomes in tigray regional state, Ethiopia: a prospective cohort study. BMC pregnancy childbirth (2019) 20 7. DOI: 10.1186/s12884-019-2708-6
--- title: Red ginseng dietary fiber promotes probiotic properties of Lactiplantibacillus plantarum and alters bacterial metabolism authors: - Hyeon Ji Jeon - Seung-Hwan You - Eoun Ho Nam - Van-Long Truong - Ji-Hong Bang - Yeon-Ji Bae - Razanamanana H. G. Rarison - Sang-Kyu Kim - Woo-Sik Jeong - Young Hoon Jung - Minhye Shin journal: Frontiers in Microbiology year: 2023 pmcid: PMC10025373 doi: 10.3389/fmicb.2023.1139386 license: CC BY 4.0 --- # Red ginseng dietary fiber promotes probiotic properties of Lactiplantibacillus plantarum and alters bacterial metabolism ## Abstract Korean red ginseng has been widely used as an herbal medicine. Red ginseng dietary fiber (RGDF) is a residue of the processed ginseng product but still contains bioactive constituents that can be applied as prebiotics. In this study, we evaluated changes on fermentation profiles and probiotic properties of strains that belong to family Lactobacillaceae with RGDF supplementation. Metabolomic analyses were performed to understand specific mechanisms on the metabolic alteration by RGDF and to discover novel bioactive compounds secreted by the RGDF-supplemented probiotic strain. RGDF supplementation promoted short-chain fatty acid (SCFA) production, carbon source utilization, and gut epithelial adhesion of Lactiplantibacillus plantarum and inhibited attachment of enteropathogens. Intracellular and extracellular metabolome analyses revealed that RGDF induced metabolic alteration, especially associated with central carbon metabolism, and produced RGDF-specific metabolites secreted by L. plantarum, respectively. Specifically, L. plantarum showed decreases in intracellular metabolites of oleic acid, nicotinic acid, uracil, and glyceric acid, while extracellular secretion of several metabolites including oleic acid, 2-hydroxybutanoic acid, hexanol, and butyl acetate increased. RGDF supplementation had distinct effects on L. plantarum metabolism compared with fructooligosaccharide supplementation. These findings present potential applications of RGDF as prebiotics and bioactive compounds produced by RGDF-supplemented L. plantarum as novel postbiotic metabolites for human disease prevention and treatment. ## Introduction Ginseng is the root of plants in the genus Panax and has been widely used as an herbal medicine in Eastern Asia (So et al., 2018). It is typically characterized by the presence of ginsenosides, which are the main bioactive components with antioxidant, anti-proliferative, and neuroprotective properties (Tam et al., 2018). In recent years, the biological properties of ginseng have been extensively demonstrated; these include enhanced immune system performance and memory, and improved blood circulation (Geng et al., 2010; Cho et al., 2018; Su et al., 2022). Korean red ginseng (Panax ginseng C.A. Meyer) is a processed product made by the repetitive steaming and drying of fresh ginseng to extend shelf life, reduce toxic effects, and enhance biological benefits (He et al., 2018). Red ginseng is traditionally consumed as a water extract containing a high concentration of ginsenosides. The residues are usually discarded, but they still contain bioactive constituents, such as unextracted ginsenosides, acidic polysaccharides, mineral elements, and dietary fiber (Yu et al., 2020). Many attempts to make the most use of these residues have included pharmaceutical, health functional foods, and cosmetics applications (Truong and Jeong, 2022). Dietary fibers are carbohydrate polymers from plant-derived foods that are not digested by human enzymes or absorbed in the gut. Polymers contribute to human gut health by increasing stool weight and regularity, thickening the contents of the intestinal tract, and promoting growth of gut microbes (Makki et al., 2018). In particular, dietary fiber can be a good fermentable source for bacteria within the large intestine and influences the composition of bacterial communities as well as microbial metabolic activities producing fermentative end products, such as short-chain fatty acids (SCFAs). These prebiotic fermentable fibers promote metabolic interactions among bacterial communities that cross-feed probiotics and inhibit the proliferation of pathogens (Holscher, 2017). Lactobacillaceae (including newly defined Lactobacillus-associated genera by taxonomic changes such as Lactiplantibacillus and Limosilactobacillus) and Bifidobacteria are the most well-known genera of probiotic organisms that normally reside in human gastrointestinal tracts. Probiotics are live microorganisms which benefit the host by producing useful physiologically bioactive compounds. These compounds have immunomodulatory, anti-carcinogenic, anti-aging, and antimicrobial effects in hosts. However, the use of these compounds is currently limited by a lack of knowledge of their molecular mechanisms, strain specific behaviors, and safety (Bourebaba et al., 2022). To address these limitations, recent studies have focused on elucidating microbial metabolism and discovering postbiotic molecules, which are defined as metabolic products secreted by probiotics in cell-free supernatants (Nataraj et al., 2020). Metabolomics is the systematic study of unique chemical molecules, termed metabolites, generated by specific cellular processes (Jordan et al., 2009). Metabolomic data are used for phenotyping molecular interactions, identifying potential biomarkers, and discovering new therapeutic targets. In this study, we aimed to find an effective strategy for utilizing processed red ginseng residue as a prebiotic dietary fiber source and evaluated its prebiotic properties on the changes in growth, metabolism, and epithelial attachment ability of probiotic Lactobacillaceae strains. Comprehensive metabolomic analyses were performed to investigate the effects of red ginseng dietary fiber (RGDF) on bacterial metabolism and to discover novel bioactive compounds secreted by the RGDF-supplemented probiotic strain. ## Bacterial strains and media Limosilactobacillus reuteri KCTC 3594 and Lactiplantibacillus plantarum KCTC 3108 were obtained from the Korean Collection for Type Cultures (KCTC, Jeongeup, Republic of Korea). The strains were pre-cultured in 50 ml of MRS broth (BD Difco, Franklin Lakes, NJ, USA) in 50 ml conical tubes and were incubated at 37°C without shaking (Biofree, Seoul, Republic of Korea) overnight. Cultures of the probiotic strains were then generated at 37°C in 50 ml of MRS broth supplemented with 0.5, 1, or $2\%$ RGDF (Korea Ginseng Corporation, Daejeon, Republic of Korea). Composition of MRS broth is as follows: 10 g/L proteose peptone, 10 g/L beef extract, 5 g/L yeast extract, 20 g/L dextrose, 1 g/L polysorbate 80, 2 g/L ammonium citrate, 5 g/L sodium acetate, 0.1 g/L magnesium sulfate, 0.05 g/L manganese sulfate, and 2 g/L dipotassium phosphate. ## Preparation of RGDF The residue remaining after water extraction of red ginseng at 87°C for 24 h was provided by Korea Ginseng Corporation (Daejeon, Republic of Korea). RGDF was prepared from the residue by drying it at 115°C and pulverizing it to 50 mesh. The physicochemical characteristics of RGDF were analyzed as previously reported (Yu et al., 2022), and same RGDF material was used in this study. ## Measurement of bacterial growth and cell mass Colony forming units per ml of probiotic strains cultured in MRS broth or in MRS supplemented with 0.5, 1, or $2\%$ of RGDF were measured by serial dilution at 0, 3, 6, 12, and 24 h. Dry cell weight of strains at 24 h was measured by collecting cell pellets by centrifugation at 4,000 rpm and 4°C for 15 min, washing the pellets three times with 10 ml of $1\%$ (w/v) phosphate buffered saline, and drying in a dry oven (JS Research Inc., Natural Convection Oven, Gongju, Republic of Korea) at 70°C for 24 h. pH of the cultured media was measured using a pH meter (Ohaus, Parsippany, NJ, USA). ## Analysis of SCFAs The concentrations of formic, acetic, propionic, and butyric acids were measured by high-performance liquid chromatography (HPLC) using the LC-6000 system (FUTECS, Daejeon, Republic of Korea). Each 1.5 ml of culture medium was collected by centrifugation (Eppendorf, Hamburg, Germany) at 13,000 rpm for 5 min at 4°C and filtered through a 0.45 μm nylon membrane filter. HPLC analysis was performed using an Aminex HPX-87X organic acid column (Bio-Rad, Hercules, CA, USA) with 0.005 M H2SO4 as the mobile phase, with a constant elution flow of 0.5 ml/min at 55°C. ## Carbon source utilization analysis An API kit (BioMérieux, Marcy l’Étoile, France) was used to compare the ability of probiotic strains to utilize the particular carbon source. Inoculation samples were prepared by collecting cultured strains from each medium that had a turbidity greater than a McFarland standard of 4. One hundred microliters of sample were inoculated into the API strip and incubated at 37°C for 4 h. After incubation, the reagents were added for reading, incubated for 10 min, and exposed to strong light at 1,000 W for 10 s to decolorize any excess reagent. Identification and interpretation were performed using the numerical profiles. ## Analysis of bacterial attachment to intestinal epithelial cells Caco-2 cell line was procured from the American Type Culture Collection (ATCC, Manassas, VA, USA) and cultured in Minimum Essential Medium (MEM) supplemented with $10\%$ fetal bovine serum, 100 U/ml penicillin, and 100 μg/ml streptomycin at 37°C in a $5\%$ CO2 atmosphere. Escherichia coli, purchased from ATCC, were grown in Luria Broth (LB) overnight. E. coli and RGDF-pretreated probiotic strains were harvested by centrifugation at 5,000 rpm for 10 min, washed twice with sterile PBS, and re-suspended in serum and antibiotic-free MEM. For adhesion assay, Caco-2 monolayer was inoculated with approximately 108 CFU/ml of L. reuteri or L. plantarum and incubated for 2 h in a $5\%$ CO2 incubator. After incubation, the monolayers were washed three times with sterile PBS to remove non-adherent bacteria. The Caco-2 cells with adherent bacteria were detached using trypsin-EDTA solution. Bacterial counts were performed by the colony counting method on MRS agar plates. Adhesion result was expressed as the percentage of the bacteria adhered divided by the initial count of bacteria added. For competition assay, approximately 108 CFU/ml of each probiotic strain and E. coli was co-incubated with Caco-2 monolayer for 1 h in a $5\%$ CO2 incubator. Non-bounded bacteria were then washed three times with sterile PBS and the Caco-2 cells with adherent bacteria were detached using trypsin-EDTA solution. The number of viable adhering E. coli was determined using the colony counting method on LB agar plates. The competition index was expressed as the percentage inhibition of E. coli adhesion in the presence of each probiotic strain divided by the adhesion of bacteria in the absence of probiotic strains. ## Metabolome analysis GC-MS has advantages of a greater chromatographic resolution compared to LC-MS and large spectral libraries, although the chemical range of metabolome coverage is narrower than LC-MS (Aretz and Meierhofer, 2016). Recently, more researches have used LC-MS to detect more peaks, but most of the identified metabolites by LC-MS are considerably overlapped with GC-MS except for lipid molecules having large molecular weights. GC-MS has been the most commonly used technique for metabolite profiling because of its hard ionization method which is highly reproducible and easy for metabolite annotation, and it still has been widely applied for metabolite profiling and identification (Baiges-Gaya et al., 2023; Kurbatov et al., 2023; Neag et al., 2023). For metabolome analysis, each intracellular and extracellular metabolites were measured in L. plantarum and L. reuteri grown in MRS medium with different supplementation of RGDF, fructooligosaccharides, or without addition. To extract intracellular and extracellular metabolites from the probiotic strains, each strain was cultured in 15 ml of medium until the mid-exponential phase determined by measuring its growth curve. Fifteen milliliters of each probiotic culture was centrifuged at 4,000 rpm for 15 min at 4°C. The supernatant was filtered through a 0.2 μm syringe filter composed of polyvinylidene fluoride for the extraction of extracellular metabolites. Aliquots (750 μL) of the filtered supernatants was mixed with 2.25 ml of 4°C methanol (GC-grade $100\%$; Sigma-Aldrich, St. Louis, MO, USA) and vortexed for 1 min. The mixtures were centrifuged at 13,000 rpm and 4°C for 10 min, and 0.1 ml of each supernatant was collected and completely dried using a Spin Driver Lite VC-36R (TAITEC Corporation, Koshigaya City, Saitama, Japan) at 2,000 rpm for 24 h. To extract intracellular metabolites from the cell pellet, 1 ml of $0.9\%$ cold NaCl (w/v) was added to the pellet and filtered through a 0.2 μm syringe filter. Then, it was transferred to a 15 ml conical tube and washed twice with 10 ml of $0.9\%$ cold NaCl (w/v). The final washed pellet was mixed with 2 ml methanol, vortexed for 10 min, and sonicated for 1 min on ice. The material was mixed with 2 ml of chloroform, vortexed for 10 min, and sonicated for 1 min with ice. Water (1.8 ml) was added and vortexing and sonication were repeated. The final mixtures were centrifuged at 13,000 rpm and 4°C for 10 min, and 0.1 ml the upper supernatant layer of each was collected and completely dried using the aforementioned Spin Driver Lite VC-36R under same conditions to extract extracellular metabolites. Methoxymation and silylation were performed for the derivatization of intracellular and extracellular metabolites. For methoxymation, 10 μL containing 20,000 ppm methyl hydroxyl chloride amine in pyridine was mixed with each dried sample and incubated at 30°C for 90 min. Next, 45 μL of N-methyl-N-trimethylsilyl-trifluoroacetamide (Fluka, Buchs, Switzerland) and 30 μL of fluoranthene as internal standard were added, vortexed for silylation, and incubated at 37°C for 30 min. The derivatized sample was transferred to a gas chromatography (GC) vial with an insert. Gas chromatography was performed using a Crystal 9000 chromatograph (Chromatotec, Val-de-Virvée, France) coupled with a Chromatotec-crystal mass spectrometer (photomultiplier detector) for the analysis of untargeted metabolites. One microliter of the derivatized sample was injected into a VF-5MS GC column (Agilent, Santa Clara, CA, USA). The oven temperature was initially 50°C for 2 min, then increased to 320°C at a rate of 5°C/min, and held at 320°C for 10 min. The helium carrier gas flowed at a rate of 1.5 ml/min. ## Statistical analysis For the deconvolution of the mass spectrometry (MS) data and identification of metabolites, MS-DIAL ver. 4.70 was used. All records of the Fiehn RI Library were used to identify metabolites by matching the MS peaks. Based on n-alkane mixture, the calculation of retention index was conducted using Kovats retention index formula: where RI, retention index of a metabolite “i”; n, carbon number of the alkane which elutes before “i”; m, number of carbons of the alkane which elutes after “i”; tri, retention time of “i”; trn, retention time of the alkane which elutes before “i”; and trm, retention time of the alkane which elutes after “i”. Retention index of each metabolite was compared with the value of standards registered in NIST 2020 Mass Spectral Library (NIST, Gaithersburg, MD, USA), and metabolites were identified based on the retention indices and mass fragmentation profiles. Uni- and multi-variance analyses, principal component analysis (PCA), hierarchical clustering analysis, and metabolite set enrichment analysis (MSEA) were performed using MetaboAnalyst (Ver. 5.0). Network analysis, such as MetaMapp, was performed using Cytoscape software. ## RGDF supplementation promotes SCFA production and carbon source utilization in L. plantarum Red ginseng contains ginsenosides that have important pharmacological roles in cancer, diabetes, and aging (Yuan et al., 2012; Yu et al., 2020; Hong et al., 2022). The by-products of the processing red ginseng still contain several types of bioactive components, such as acidic polysaccharides and dietary fiber, as well as the remaining ginsenosides (Park and Kim, 2006). RGDF is a byproduct composed of approximately $31\%$ dietary fiber (314.3 mg/g) and $0.66\%$ ginsenoside (6.63 mg/g of total ginsenosides) (Yu et al., 2022). Since dietary fibers are well-known prebiotic ingredients for bacterial growth promotion and probiotic functionality, we first screened the effects of RGDF on metabolic profiles of probiotic strains, including L. reuteri, L. plantarum, Lactobacillus acidophilus, Lacticaseibacillus casei, and Lactococcus lactis (Supplementary Table 1). We selected two probiotic strains, L. plantarum and L. reuteri, which were most positively and negatively affected, respectively, by RGDF supplementation. Although RGDF supplementation slightly enhanced the growth of both probiotic Lactobacillaceae strains, the difference was not significant compared with control (Figures 1A, B). The pH change of cultured media also was not different between control and RGDF supplementation. To reveal possible associations between RGDF and probiotic functionality, we next measured the production of SCFAs and carbon source utilization profiles with RGDF. L. plantarum enhanced the production of SCFAs, specifically lactate and acetate, with RGDF supplementation in a dose-dependent manner. L. reuteri reduced the production of these metabolites (Figures 1C, D). RGDF also improved the carbon source utilization ability of L. plantarum but had no effect on L. reuteri (Figure 1E). Thus, RGDF supplementation can promote the production of beneficial metabolites (lactate and acetate) and carbon source utilization by L. plantarum. FIGURE 1: Fermentation profiles of L. plantarum and L. reuteri. (A,B) Bacterial growth, (C,D) pH of cultured media, (E,F) lactate and acetate production, and (G) carbon source utilization. Differences were indicated at a significance level of 95% () and 99% (*), as determined by one-way ANOVA with Dunnett’s post-hoc analysis. Error bars represent standard deviation (SD). RGDF, red ginseng dietary fiber; RIB, D-ribose; ARA, L-arabinose; MAN, D-mannitol; SOR, D-sorbitol; LAC, D-lactose; TRE, D-trehalose; INU, inulin; RAF, D-raffinose; AMD, starch; GLYG, glycogen; +, positive; –, negative. ## RGDF supplementation promotes gut epithelial adhesion of L. plantarum and protects against enteropathogens Dietary fibers help maintain intestinal homeostasis by promoting probiotics, limiting the growth and adhesion of pathogenic microbes, and stimulating fiber-derived SCFA production (Cai et al., 2020). RGDF supplementation significantly increased the adhesion of L. plantarum to gut epithelial cells compared to the control. The adhesion was most pronounced in the presence of $0.5\%$ RGDF (Figure 2A). Adhesion of L. plantarum and L. reuteri to the gut epithelium was decreased by adding RGDF (Figure 2B). L. reuteri is a probiotic that has a well-documented adhesive ability (approximately $30\%$ in the control) (Gao et al., 2016). This behavior was confirmed in the present study; a high percentage of adhesion in the control was evident compared with L. plantarum (approximately $2\%$ in the control). FIGURE 2: Gut epithelial adhesion (A,B) and inhibition of E. coli attachment (C,D) of L. plantarum and L. reuteri. Differences were indicated at a significance level of 95% () and 99% (*), as determined by one-way ANOVA with Dunnett’s post-hoc analysis. To evaluate the competitive inhibitory effects of RGDF-supplemented strains on binding of enteropathogenic bacteria to the host epithelium, E. coli and RGDF-pretreated probiotic strains were co-incubated with Caco-2 monolayer (Figures 2C, D). Supplementation with RGDF increasingly reduced the E. coli attachment in the presence of both L. plantarum and L. reuteri; greater differences were observed in L. plantarum. Similar to the epithelial adhesion of L. reuteri, the strain showed a higher basal level of competitiveness against pathogen attachment than L. plantarum. However, addition of RGDF significantly improved adhesion of the gut epithelium and protected against E. coli attachment of L. plantarum, which can broaden the applicability of the strain as a probiotic. It is noted that several factors would affect epithelial adhesion of the strains including presence of surface proteins, auto-aggregation and bacterial surface hydrophobicity. Bacterial adhesion is based on non-specific physical interactions and aggregation abilities that also form a barrier preventing colonization of pathogens (Kos et al., 2003). Dell’Anno et al. [ 2021], showed that both L. plantarum and L. reuteri showed auto-aggregation and epithelial adhesion. L. plantarum and L. reuteri had higher hydrophobicity and greater auto-aggregation, respectively, reflecting their different colonizing ability. The collective findings indicate that RGDF supplementation promoted gut epithelial adhesion and had a protective role against enteropathogens in the presence of L. plantarum. ## RGDF supplementation alters intracellular metabolic profiles of L. plantarum, but not L. reuteri Although both L. plantarum and L. reuteri utilize dietary fibers as prebiotics, our results indicate that RGDF supplementation was effective in L. plantarum, but not in L. reuteri. To identify the effects of RGDF on bacterial metabolism, we first determined the intracellular metabolome changes between RGDF supplementation and control in L. plantarum and L. reuteri. Total 106 of metabolites were identified including sugars, amino acids, fatty acids, organic acids, and polyamines (Supplementary Table 2). PCA results clearly showed metabolic alterations with $0.5\%$ (w/v) RGDF supplementation in L. plantarum, while the metabolic profile of L. reuteri with RGDF was not different (Figure 3A). Loading of PC1 and PC2 indicated that fumaric acid (−0.834 at PC1), uracil (−0.924 at PC1), picolinic acid (0.791 at PC1), and 2-hydroxybutanoic acid (0.763 at PC1) were important metabolites determining the metabolic differences between L. plantarum and L. reuteri. FIGURE 3: Intracellular metabolomic analysis of L. plantarum and L. reuteri cultured with 0.5% (w/v) RGDF compared to the control MRS broth. (A) Principle component analysis (PCA) score and loading plots. (B) Volcano plot of L. plantarum. Significantly decreased metabolites are indicated by blue triangles. (C) MetaMapp of L. plantarum culture with RGDF compared to the control MRS broth. Each node is a structurally identified metabolite. Blue nodes are decreased metabolites, and yellow nodes are unchanged metabolites. The size of nodes and labels reflect fold-changes and p-values by t-test, respectively. (D) MSEA of L. plantarum. (E) Normalized abundance of intracellular metabolites of L. plantarum cultured with 0.5% (w/v) RGDF compared to the control MRS broth. Data are expressed as violin plots of six determinations. Differences between metabolite abundances were all significant at a significance level of 95% () and 99% (*), as determined by the Student’s t-test. MetaMapp, a network graph of metabolites based on biochemical pathways and chemical and mass spectral similarities, displayed significantly altered metabolites ($p \leq 0.05$) with RGDF compared to the control in L. plantarum (Figure 3C). MSEA also supported the results of significantly altered bacterial metabolism, especially sugar (galactose, starch, and sucrose) metabolism and unsaturated fatty acid biosynthesis (Figure 3D). Considering the significant increase in lactate and acetate production and carbohydrate utilization in L. plantarum with RGDF (Figure 1), we suggest that glycolytic metabolic flow and membrane flexibility, respectively, can be affected by RGDF supplementation. In addition, we compared the effect of RGDF on the intensity of each metabolite with that of the control using a volcano plot (Figure 3B). The intensities of the four metabolites (oleic acid, nicotinic acid, uracil, and glyceric acid) decreased after RGDF supplementation in L. plantarum (Figure 3E). The relative abundance of these metabolites was also significantly reduced by RGDF, verifying that metabolic processes associated with the four metabolites were specifically altered by RGDF (Supplementary Figure 1). Together, these findings suggest that L. plantarum, but not L. reuteri, is specifically affected by RGDF supplementation via central carbon metabolism. ## RGDF supplementation promotes biosynthesis of specific metabolites in L. plantarum Postbiotics are nonviable bacterial metabolic products with biological activity in the host (Nataraj et al., 2020). These molecules have several advantages over probiotics with respect to safety and effectiveness, such as triggering only targeted responses by a defined mechanism, better accessibility of microbe-associated molecular patterns, and ease of production and storage (Nataraj et al., 2020). To systemically characterize postbiotic metabolites specifically produced by RGDF supplementation in L. plantarum, we further analyzed the extracellular metabolome in L. plantarum and L. reuteri grown with $0.5\%$ RGDF, defined as the relative metabolite intensity in spent medium from bacterial culture to metabolite intensity in baseline medium (Jain et al., 2012). As shown in the PCA results, exometabolome profiles were clearly separated between the bacterial strains, as well as between the RGDF supplement and control (Figure 4A). FIGURE 4: Extracellular metabolomic analysis of L. plantarum and L. reuteri cultured with 0.5% (w/v) RGDF compared to the control MRS broth. (A) Principle component analysis (PCA) score and loading plots. Normalized abundance of intracellular metabolites of L. plantarum (B–E) cultured with 0.5% (w/v) RGDF compared to the control MRS broth. Data are expressed as violin plots of six determinations. Differences between metabolite abundances were all significant at a significance level of 95% () and 99% (*), as determined by the Student’s t-test. Red ginseng dietary fiber-specific bacteria-derived metabolites were distinguished from the media components based on three criteria: [1] the averaged value of metabolite intensity in the spent medium subtracted from its intensity in the uncultured medium should be positive; [2] the statistical significance between RGDF and control should be under the level of $95\%$ confidence; and [3] the absolute change in metabolite intensity with RGDF compared to the control should be >2. Based on these criteria, we identified four L. plantarum metabolites (oleic acid, 2-hydroxybutanoic acid, hexanol, and sec-butyl acetate) biosynthesized specifically in response to the RGDF supplement (Figures 4B–E). The collective findings indicate that RGDF supplementation promoted the biosynthesis of specific metabolites in L. plantarum. These metabolites included oleic acid, 2-hydroxybutanoic acid, hexanol, and butyl acetate. ## RGDF supplementation has distinct effects on L. plantarum metabolism compared with fructooligosaccharide supplementation Dietary fiber, a plant-derived component that cannot be completely digested by human enzymes, consists of non-starch polysaccharides, including cellulose and oligosaccharides (Veronese et al., 2018). Fructooligosaccharides (FOS) are dietary fibers composed of linear chains of fructose units linked by β-[2,1] bonds (Sabater-Molina et al., 2009). They naturally occur in plants, such as onion, chicory, and banana, and are increasingly used in food products because of their prebiotic effect, which stimulates the growth of probiotic gut microbiota (Sabater-Molina et al., 2009). To compare the effects of different type of dietary fibers on metabolic alteration in L. plantarum, we cultured L. plantarum on control MRS, MRS with $0.5\%$ RGDF, and MRS with $0.5\%$ FOS. Similar to the growth results of RGDF shown in Figure 1, supplementation with either RGDF or FOS did not have an effect on bacterial growth (Supplementary Table 3). In contrast to the lack of observable differences in bacterial growth, the metabolome profile of L. plantarum supplemented with RGDF showed a transition between MRS and FOS in both intracellular and extracellular states (Figures 5A, B). Similar to the effect of RGDF shown in Figures 3C, D, FOS also decreased the abundance of specific metabolites in sugar and central carbon metabolism, while the abundance of leucine specifically increased with FOS supplementation compared to the control (Figure 5C). MSEA analysis indicated that the citrate cycle and its associated pathways, such as alanine, aspartate, and glutamate metabolism, as well as sugar metabolism, were altered by FOS treatment (Figure 5D). Comparison of the intracellular metabolite abundance of RGDF with FOS revealed that RGDF supplementation resulted in a decreased abundance of palmitic acid and stearic acid, while uracil, raffinose, ascorbic acid, and 2-hydroxybutanoic acid comparatively increased in RGDF (Figures 5E, F). FIGURE 5: Metabolomic analysis of L. plantarum cultured with 0.5% (w/v) RGDF or 0.5% (w/v) FOS. Principle component analysis (PCA) score and loading plots of intracellular (A) and extracellular (B) metabolome. MetaMapp of L. plantarum cultured with FOS compared to the control MRS broth (C) and cultured with RGDF compared to FOS (E). Each node is a structurally identified metabolite. Blue nodes are decreased metabolites, red nodes are increased metabolites, and yellow nodes are unchanged metabolites. The size of nodes and labels reflect fold-changes and p-values by t-test, respectively. MSEA of L. plantarum cultured with FOS compared to the control MRS broth (D) and cultured with RGDF compared to FOS (F). Next, we compared the extracellular metabolites differentially produced by FOS treatment to the control, applying the same criteria used for RGDF treatment (Figure 6). As expected, the culture supernatant of cells grown with FOS contained a significantly higher abundance of sugars and sugar derivatives than those grown with RGDF, including raffinose, D-glucosamine, and pinitol. Production of RGDF-specific metabolites, including oleic acid, 2-hydroxybutanoic acid, hexanol, and sec-butyl acetate, was not significantly induced by FOS, suggesting that the metabolism of these molecules is RGDF-specific. Thus, RGDF supplementation had distinct effects on L. plantarum metabolism compared with FOS supplementation. FIGURE 6: Normalized abundance of extracellular metabolites of L. plantarum cultured with 0.5% (w/v) RGDF or 0.5% (w/v) FOS. Data are expressed as violin plots of six determinations. Differences were indicated at a significance level of 95% () and 99% (*), as determined by one-way ANOVA with Tukey’s post-hoc analysis. ## Discussion Dietary fibers and the associated phytochemicals in ginseng-derived products provide various functional and health benefits. In this study, we evaluated the effects of RGDF as a prebiotic constituent on the physiological and metabolic alterations of probiotics. With RGDF supplementation in the growth media, L. plantarum showed the highest production of SCFAs, specifically lactate and acetate, and the most increased carbohydrate-fermenting capability compared with other probiotic Lactobacillaceae species, especially L. reuteri. In addition, RGDF improved gut epithelial adhesion of L. plantarum and protected against enteropathogens. Analysis of the intracellular metabolome of L. plantarum indicated decreases in metabolites of sugars and unsaturated fatty acids, and significant decreases in the abundance of oleic acid, nicotinic acid, uracil, and glyceric acid. RGDF supplementation also promoted the secretion of specific metabolites, such as oleic acid, 2-hydroxybutanoic acid, hexanol, and butyl acetate, in L. plantarum. Comparison of the metabolic alteration by red ginseng-derived dietary fiber with a representative dietary fiber, FOS, showed distinguishable effects between the two different types of fibers in L. plantarum. Although dietary fibers generally promote probiotic growth, their effects are strain specific. Our results consistently revealed that RGDF supplementation improved the probiotic properties of L. plantarum, but not of L. reuteri. L. plantarum, unlike most probiotic Lactobacillaceae species, exhibits ecological and metabolic flexibility and thus maintains a diverse functional genome that facilitates the flexibility to colonize a variety of environments (Fidanza et al., 2021). For example, L. plantarum strains exhibit acid tolerance by inducing alterations in the fatty acid composition of the bacterial membrane upon exposure to low-pH conditions (Huang et al., 2016). Genome analysis of 165 L. plantarum strains revealed the presence of a large number of carbohydrates metabolizing genes and two-component systems and signal transduction systems regulating physiological processes, facilitating the adaptability of the species in various environments compared to other lactic acid bacteria and even among probiotic Lactobacillaceae strains (Cui et al., 2021). In addition, L. plantarum produces bacteriocins termed plantaricins, which can effectively inhibit enteropathogenic bacteria, such as E. coli, under specific circumstances (Pal and Srivastava, 2014). These findings based on the diverse functional genetic characteristics support our results that L. plantarum greatly modulates and improves their metabolic functions, including acid production, carbohydrate utilization, and inhibition of pathogen growth in the presence of RGDF. To explain how RGDF promotes bacterial metabolic alterations in L. plantarum, but not in L. reuteri, and how the effects of RGDF are different from those of other dietary fibers, beyond the genetic flexibility of L. plantarum, we interpreted intracellular metabolic changes of L. plantarum and L. reuteri when supplied with RGDF and FOS. RGDF supplementation resulted in a significant decrease in the abundance of oleic acid, nicotinic acid, uracil, and glyceric acid. Oleic acid [cis-9-octadecenoic acid; 18:1(9c)] is the most common monounsaturated fatty acid in animals and vegetables. It is incorporated into the membranes of lactic acid bacteria grown in a medium, but is not synthesized (Johnsson et al., 1995). In L. plantarum, our metabolomic analysis indicated that the intracellular abundance of oleic acid decreased, while the extracellular abundance increased with RGDF supplementation. These findings suggest that oleic acid might be less incorporated from the medium, possibly by modified membrane rigidity by RGDF. Nicotinic acid, also known as niacin, is a form of vitamin B3 and is an essential human nutrient that can be supplied by plants and bacteria. Several cellular processes require the compound as a component of the coenzymes nicotinamide adenine dinucleotide (NAD) and NAD phosphate (NADP). In probiotic Lactobacillaceae spp., free nicotinic acid decrease with increasing cellular activity as it is largely incorporated in the form of cofactors (McIlwain et al., 1949). Nicotinic acid is also an important cofactor for lactate dehydrogenase, acting as the limiting factor for lactate production during fermentation, which might be associated with the reduced intracellular abundance and improved lactate production by RGDF (Colombié and Sablayrolles, 2004). Glyceric acid is a precursor of several phosphate derivatives that are important biochemical intermediates in glycolysis. 3-Phosphoglyceric acid is one derivative that is especially important for serine and cysteine biosynthesis. A recent study demonstrated that L. plantarum supplemented with $2\%$ RGDF upregulates the expression of genes involved in serine (sdhA, sdhB, and sdaC) and cysteine metabolism (cysE) (Yu et al., 2022). Although further verification of the changes in specific metabolic and physiologic mechanisms is required, our results support the view that RGDF supplementation alters cellular and metabolic processes. Lactobacilli are recognized for their ability to secrete many beneficial metabolites, such as SCFAs, indole-derivatives, and vitamins (Wang et al., 2018; Thompson et al., 2020; Sugimura et al., 2022). Our exometabolomic analysis revealed that 2-hydroxybutanoic acid, hexanol, and butyl acetate as metabolites that were secreted specifically in response to RGDF supplementation. These compounds are generally excreted as end products during propanoate biosynthesis and butanol metabolism. In mammalian tissues, 2-hydroxybutanoic acid, also known as α-hydroxybutyrate, is released as a byproduct when cystathionine is cleaved to cysteine for detoxification against oxidative stress. Although it has been used as a biomarker of type 2 diabetes and lactic acidosis, novel roles of 2-hydroxybutanoic acid have been suggested to protect against acetaminophen-induced liver injury and immune modulation against viral infection (Liu et al., 2018; Zheng et al., 2020; Shi et al., 2021). For example, the level of serum 2-hydroxybutanoic acid was reportedly enriched in patients with viral infections that included human papilloma virus or SARS-CoV-2 compared to healthy controls (Liu et al., 2018; Shi et al., 2021). It could be a result of the activation of antioxidant responses and control of cellular redox balance. Hexanol is an organic alcohol used in the perfume industry; its odor is that of freshly mown grass with a hint of strawberries. Its health-related functions are unclear, but it reportedly modulates the function of the actomyosin motor (Komatsu et al., 2004). Similar to hexanol, butyl acetate possesses characteristic flavors and a sweet odor of bananas or apples (Holland et al., 2005). It also has antimicrobial activity against undesirable microorganisms in cosmetic products, such as Staphylococcus aureus and E. coli (Lens et al., 2016). The specific mechanism of the secretion of these metabolites following stimulation by RGDF supplementation and comparative studies with FOS, would provide some evidence that metabolite production is highly specific to RGDF, but not to carbohydrate polymer-based dietary fiber. Further genetic investigations are required to elucidate the underlying mechanism. ## Conclusion Red ginseng dietary fiber supplementation promoted probiotic properties of L. plantarum, including production of SCFAs (lactate and acetate), carbohydrate utilization, epithelial attachment, and pathogen inhibition. Comparative metabolomic analyses suggested RGDF-related modification of cellular and metabolic processes, including membrane biology and central carbon metabolism. In addition, the potential applications of bioactive compounds produced by RGDF-supplemented L. plantarum have been proposed as novel postbiotic metabolites. ## Data availability statement The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation. ## Author contributions S-HY, YJ, and MS designed the study, and drafted and revised the manuscript. HJ, V-LT, J-HB, Y-JB, and RR performed the experiments, analyzed the data, and collected the samples and data interpretation. EN, S-KK, and W-SJ revised the manuscript and obtained the funding. All authors had read and approved the final manuscript. ## Conflict of interest S-HY and S-KK were employed by Korea Ginseng Corporation. 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/fmicb.2023.1139386/full#supplementary-material ## References 1. Aretz I., Meierhofer D.. Advantages and pitfalls of mass spectrometry based metabolome profiling in systems biology.. (2016) 17. DOI: 10.3390/ijms17050632 2. Baiges-Gaya G., Iftimie S., Castañé H., Rodríguez-Tomàs E., Jiménez-Franco A., López-Azcona A. F.. Combining semi-targeted metabolomics and machine learning to identify metabolic alterations in the serum and urine of hospitalized patients with COVID-19.. (2023) 13. DOI: 10.3390/biom13010163 3. Bourebaba Y., Marycz K., Mularczyk M., Bourebaba L.. Postbiotics as potential new therapeutic agents for metabolic disorders management.. (2022) 153. DOI: 10.1016/j.biopha.2022.113138 4. Cai Y., Folkerts J., Folkerts G., Maurer M., Braber S.. Microbiota-dependent and -independent effects of dietary fibre on human health.. (2020) 177 1363-1381. DOI: 10.1111/bph.14871 5. Cho I.-H., Kang B.-W., Yun-Jae P., Lee H.-J., Park S., Lee N.. Ginseng berry extract increases nitric oxide level in vascular endothelial cells and improves cGMP expression and blood circulation in muscle cells.. (2018) 22 6-13. DOI: 10.20463/jenb.2018.0018 6. Colombié S., Sablayrolles J.-M.. Nicotinic acid controls lactate production by K1-LDH: A Saccharomyces cerevisiae strain expressing a bacterial LDH gene.. (2004) 31 209-215. DOI: 10.1007/s10295-004-0138-5 7. Cui Y., Wang M., Zheng Y., Miao K., Qu X.. The carbohydrate metabolism of. (2021) 22. DOI: 10.3390/ijms222413452 8. Dell’Anno M., Giromini C., Reggi S., Cavalleri M., Moscatelli A., Onelli E.. Evaluation of adhesive characteristics of. (2021) 9. DOI: 10.3390/microorganisms9081587 9. Fidanza M., Panigrahi P., Kollmann T. R.. (2021) 12. DOI: 10.3389/fmicb.2021.712236 10. Gao K., Liu L., Dou X., Wang C., Liu J., Zhang W.. Doses. (2016) 6. DOI: 10.1038/srep28332 11. Geng J., Dong J., Ni H., Lee M. S., Wu T., Jiang K.. Ginseng for cognition.. (2010) 2010. DOI: 10.1002/14651858.CD007769.pub2 12. He M., Huang X., Liu S., Guo C., Xie Y., Meijer A. H.. The Difference between white and red ginseng: Variations in ginsenosides and immunomodulation.. (2018) 84 845-854. DOI: 10.1055/a-0641-6240 13. Holland D., Larkov O., Bar-Ya’akov I., Bar E., Zax A., Brandeis E.. Developmental and varietal differences in volatile ester formation and acetyl-coA: Alcohol acetyl transferase activities in apple (. (2005) 53 7198-7203. DOI: 10.1021/jf050519k 14. Holscher H. D.. Dietary fiber and prebiotics and the gastrointestinal microbiota.. (2017) 8 172-184. DOI: 10.1080/19490976.2017.1290756 15. Hong J., Gwon D., Jang C.-Y.. Ginsenoside Rg1 suppresses cancer cell proliferation through perturbing mitotic progression.. (2022) 46 481-488. DOI: 10.1016/j.jgr.2021.11.004 16. Huang R., Pan M., Wan C., Shah N. P., Tao X., Wei H.. Physiological and transcriptional responses and cross protection of. (2016) 99 1002-1010. DOI: 10.3168/jds.2015-9993 17. Jain M., Nilsson R., Sharma S., Madhusudhan N., Kitami T., Souza A. L.. Metabolite profiling identifies a key role for glycine in rapid cancer cell proliferation.. (2012) 336 1040-1044. DOI: 10.1126/science.1218595 18. Johnsson T., Nikkila P., Toivonen L., Rosenqvist H., Laakso S.. Cellular fatty acid profiles of. (1995) 61 4497-4499. DOI: 10.1128/aem.61.12.4497-4499.1995 19. Jordan K. W., Nordenstam J., Lauwers G. Y., Rothenberger D. A., Alavi K., Garwood M.. Metabolomic characterization of human rectal adenocarcinoma with intact tissue magnetic resonance spectroscopy.. (2009) 52 520-525. DOI: 10.1007/DCR.0b013e31819c9a2c 20. Komatsu H., Shigeoka T., Ohno T., Kaseda K., Kanno T., Matsumoto Y.. Modulation of actomyosin motor function by 1-hexanol.. (2004) 25 77-85. DOI: 10.1023/B:JURE.0000021350.85334.2f 21. Kos B., Šušković J., Vuković S., Šimpraga M., Frece J., Matošić S.. Adhesion and aggregation ability of probiotic strain. (2003) 94 981-987. DOI: 10.1046/j.1365-2672.2003.01915.x 22. Kurbatov I., Kiseleva O., Arzumanian V., Dolgalev G., Poverennaya E.. Some lessons learned on the impact of the storage conditions, syringe wash solvent, and the way of GC-MS injection on the reproducibility of metabolomic studies.. (2023) 13. DOI: 10.3390/metabo13010075 23. Lens C., Malet G., Cupferman S.. Antimicrobial activity of Butyl acetate, Ethyl acetate and Isopropyl alcohol on undesirable microorganisms in cosmetic products.. (2016) 38 476-480. DOI: 10.1111/ics.12314 24. Liu Y., Guo J.-Z., Liu Y., Wang K., Ding W., Wang H.. Nuclear lactate dehydrogenase A senses ROS to produce α-hydroxybutyrate for HPV-induced cervical tumor growth.. (2018) 9. DOI: 10.1038/s41467-018-06841-7 25. Makki K., Deehan E. C., Walter J., Bäckhed F.. The impact of dietary fiber on gut microbiota in host health and disease.. (2018) 23 705-715. DOI: 10.1016/j.chom.2018.05.012 26. McIlwain H., Stanley D. A., Hughes D. E.. The behaviour of. (1949) 44 153-158. DOI: 10.1042/bj0440153 27. Nataraj B. H., Ali S. A., Behare P. V., Yadav H.. Postbiotics-parabiotics: The new horizons in microbial biotherapy and functional foods.. (2020) 19. DOI: 10.1186/s12934-020-01426-w 28. Neag E. J., Collao V., Bhattacharya S. K.. Analysis of cholesterol lipids using gas chromatography mass spectrometry.. (2023) 2625 141-148. DOI: 10.1007/978-1-0716-2966-6_13 29. Pal G., Srivastava S.. Inhibitory effect of plantaricin peptides (Pln E/F and J/K) against. (2014) 30 2829-2837. DOI: 10.1007/s11274-014-1708-y 30. Park S.-H., Kim W.-J.. Study of hongsambak for medicinal foods applications –nutritional composition, antioxidants contents and antioxidative activity.. (2006) 20 449-454 31. Sabater-Molina M., Larqué E., Torrella F., Zamora S.. Dietary fructooligosaccharides and potential benefits on health.. (2009) 65 315-328. DOI: 10.1007/BF03180584 32. Shi D., Yan R., Lv L., Jiang H., Lu Y., Sheng J.. The serum metabolome of COVID-19 patients is distinctive and predictive.. (2021) 118. DOI: 10.1016/j.metabol.2021.154739 33. So S.-H., Lee J. W., Kim Y.-S., Hyun S. H., Han C.-K.. Red ginseng monograph.. (2018) 42 549-561. DOI: 10.1016/j.jgr.2018.05.002 34. Su F., Xu L., Xue Y., Xu W., Li J., Yu B.. Immune enhancement of nanoparticle-encapsulated ginseng stem-leaf saponins on porcine epidemic diarrhea virus vaccine in mice.. (2022) 10. DOI: 10.3390/vaccines10111810 35. Sugimura N., Li Q., Chu E. S. H., Lau H. C. H., Fong W., Liu W.. (2022) 71 2011-2021. DOI: 10.1136/gutjnl-2020-323951 36. Tam D., Truong D., Nguyen T., Quynh L., Tran L., Nguyen H.. Ginsenoside Rh1: A systematic review of its pharmacological properties.. (2018) 84 139-152. DOI: 10.1055/s-0043-124087 37. Thompson H. O., Önning G., Holmgren K., Strandler H. S., Hultberg M.. Fermentation of cauliflower and white beans with. (2020) 75 236-242. DOI: 10.1007/s11130-020-00806-2 38. Truong V.-L., Jeong W.-S.. Red ginseng (. (2022) 46 214-224. DOI: 10.1016/j.jgr.2021.12.006 39. Veronese N., Solmi M., Caruso M. G., Giannelli G., Osella A. R., Evangelou E.. Dietary fiber and health outcomes: An umbrella review of systematic reviews and meta-analyses.. (2018) 107 436-444. DOI: 10.1093/ajcn/nqx082 40. Wang J., Ji H., Wang S., Liu H., Zhang W., Zhang D.. Probiotic. (2018) 9. DOI: 10.3389/fmicb.2018.01953 41. Yu H.-Y., Rhim D.-B., Kim S.-K., Ban O.-H., Oh S.-K., Seo J.. Growth promotion effect of red ginseng dietary fiber to probiotics and transcriptome analysis of Lactiplantibacillus plantarum.. (2022) 47. DOI: 10.1016/j.jgr.2022.09.003 42. Yu S., Xia H., Guo Y., Qian X., Zou X., Yang H.. Ginsenoside Rb1 retards aging process by regulating cell cycle, apoptotic pathway and metabolism of aging mice.. (2020) 255. DOI: 10.1016/j.jep.2020.112746 43. Yuan H.-D., Kim J.-T., Kim S.-H., Chung S.-H.. Ginseng and diabetes: The evidences from in vitro, animal and human studies.. (2012) 36 27-39. DOI: 10.5142/jgr.2012.36.1.27 44. Zheng N., Gu Y., Hong Y., Sheng L., Chen L., Zhang F.. Vancomycin pretreatment attenuates acetaminophen-induced liver injury through 2-hydroxybutyric acid.. (2020) 10 560-570. DOI: 10.1016/j.jpha.2019.11.003
--- title: Ethyl-acetate fraction from a cinnamon-cortex extract protects pancreatic β-cells from oxidative stress damage authors: - Weiling Li - Jialu Qiao - Kuan Lin - Ping Sun - Yuansong Wang - Qian Peng - Xiansheng Ye - Wei Liu - Binlian Sun journal: Frontiers in Pharmacology year: 2023 pmcid: PMC10025376 doi: 10.3389/fphar.2023.1111860 license: CC BY 4.0 --- # Ethyl-acetate fraction from a cinnamon-cortex extract protects pancreatic β-cells from oxidative stress damage ## Abstract Background: The pathogenesis of diabetes mellitus is mediated mainly by oxidative stress produced by damaged pancreatic β-cells. We identified that an ethyl-acetate fraction (EA) from a cinnamon-cortex extract (CCE) is rich in flavonoid, and showed no toxicity to β cells. Objective: In this study, we evaluated the pharmacologic activities of EA on pancreatic β cells using a model of oxidative stress induced by H2O2 or alloxan. Results: The results showed that EA could significantly reduce reactive oxygen (ROS) accumulation to improve the survival of cells. Western blot showed that EA treatment upregulated expression of nuclear factor erythroid 2 related factor 2, heme oxygenase-1, and gamma glutamylcysteine synthetase. The same model study found that EA also can protect β cells against the apoptosis induced by oxidative stress. Furthermore, EA can enhance insulin secretion in rat and mouse β cell lines treated or not with alloxan or H2O2. The expression of the insulin transcription factor PDX-1 increased in an EA concentration-dependent manner. At last, the major functional compounds of EA analysis showed that three compounds, cinnamyl alcohol, coumarin, and cinnamic acid, had similar effects as EA. Conclusions: In sum, our data suggested that EA fraction from CCE can protect β cells from oxidative stress, and increase insulin secretion to improve the function of β cells. This function might be due to these three compounds found in EA. Our findings provide a theoretical basis and functional molecules for the use of CCE against diabetes mellitus. ## 1 Introduction Diabetes mellitus (DM) is a metabolic disease which results in uncontrolled high blood sugar. The prevalence of DM is increasing rapidly worldwide (Demir et al., 2021). Long-term hyperglycemia leads to the damage and dysfunction of various organs, which seriously affects the health and quality of life of patients. Hyperglycemic-induced ROS production (composed mostly of H2O2 and O2 −) takes part in DM developing and associated complications (Black, 2022). When the levels of ROS exceed the cell’s ability to detoxify it, the equilibrium is breached, then the cell enters into oxidative stress (Oguntibeju, 2019). One of the major mechanisms for DM is that high levels of ROS, which can induce β cell apoptosis, reduce β cell proliferation, and damage β cell function (Mathis et al., 2001; Baumel-Alterzon and Scott, 2022). Despite the advent of numerous anti-hyperglycemic agents, there are significant challenges in optimizing therapies and reducing the side-effects associated with them. Recently, extensive studies have been conducted on antioxidants to reduce the excessive production of ROS observed in DM (Darenskaya et al., 2021). Natural antioxidants showed in vitro and in vivo protective effects on the recovery and preservation of functional β cells (Zhang et al., 2020). These studies encourage further investigation on the beneficial effects of natural compounds in diabetic condition. Cinnamon is used as a spice worldwide. Aqueous or ethanol extracts of cinnamon have been shown to exert anti-DM effects in several preclinical and clinical investigations (Mang et al., 2006; Qureshi et al., 2019; Zhou et al., 2022). Diverse species of cinnamon have been reported to have different hypoglycemic effects (Lu et al., 2011), but their constituents and underlying molecular mechanism of action are incompletely understood. Studies have reported that cinnamon extracts can inhibit α-amylase and α-glucosidase activities (Adisakwattana et al., 2011; Mohamed Sham Shihabudeen et al., 2011; Hayward et al., 2019). The monomer procyanidin C1 and cinnamaldehyde isolated from cinnamon have been reported to enhance insulin sensitivity (Li et al., 2015; Sun et al., 2019). Several types of compounds, including flavonoids and phenylpropanoids, have been identified from cinnamon (Farag et al., 2022). It was proven that flavonoids have a crucial role in DM, hyperlipidemia, and neurodegenerative diseases (Al-Ishaq et al., 2019; Hussain et al., 2020). Recently phenylpropanoids have been reported to prevent post-prandial hyperglycemia and type 2 diabetes mellitus by promoting glucose uptake (Yoshioka et al., 2022). Studies have illustrated several benefits of specific dietary natural antioxidants on DM: reducing apoptosis, promoting proliferation of β cells, alleviating oxidative stress, inhibiting α-glucosidase activity, and increasing insulin production (Babu et al., 2013; Wang et al., 2019; El-Ouady et al., 2021) [17–19]. Therefore, antioxidants from cinnamon extract could be developed into drugs used for DM management. We wished to evaluate the anti-DM effect of an ethyl-acetate fraction (EA) from a cinnamon-cortex extract (CCE) on damaged pancreatic β cells treated with alloxan or H2O2. We found that β cells could be protected from damage and activated to produce insulin after EA treatment. Then, we identified the major functional compounds present in EA. These findings provide a theoretical basis for therapy of DM using cinnamon. ## 2.1 Preparation of CCE extract CCE was prepared from ground cinnamon purchased from a local pharmacy. Briefly, cinnamon powder (100 g) was extracted with $60\%$ aqueous ethanol (1 L) for 1.5 h at 70°C. The aqueous portion was partitioned by ethyl acetate to obtain EA. The concentrated solutions were adsorbed onto AB-8 macro porous resin (Cohesion, Beijing, China). Furthermore, water and ethanol were used to remove excess carbohydrates and low-molecular-weight compounds. The solution was eluted with $80\%$ ethanol and then concentrated and freeze-dried into powder to obtain purified EA. The total content of flavonoids and phenolic compounds was determined as milligrams of rutin equivalents (RE) and gallic-acid equivalents (GAE), respectively, per gram of extract. The chemical composition of EA was examined using high-performance liquid chromatography (HPLC) employing a 1,260 Infinity system (Agilent Technologies, USA) with an ultraviolet detector. The sample was separated using a C18 column (4.6 × 250 mm, 5 μm; PerkinElmer, USA), and analyzed using a mobile phase of $0.3\%$ acetic acid in water and acetonitrile. Absorbance was measured at 254 nm and 280 nm. ## 2.2 Evaluation of the antioxidant capacity of EA in vitro We conducted assays to measure the ability to scavenge radicals (hydroxyl and superoxide). These assays provided a basis for further research on the antioxidant capacity of EA from CCE. For the assays, EA and vitamin C (VC) of different concentrations (0.1, 0.25, 0.5, 1 mg/mL) were prepared. ## 2.2.1 Hydroxyl radical scavenging assay The test of the ability to scavenge hydroxyl radicals was carried out according to the method described by the previous study with minor modifications (Zhang et al., 2014). A mixture of solutions containing FeSO4 (9 mM, 100 μL), H2O2 (6 mM, 2 mL), and test sample (2 mL) were incubated for 10 min at 25°C. Then, salicylic acid (9 mM, 2 mL) was dripped into the mixture, followed by incubation for 30 min at 25°C. Absorbance was measured at 518 nm. Percent scavenging inhibition (%) was calculated using the following equation: Scavenging of hydroxyl radicals %=A0−A1/A2−A1×100 where A0, A1, and A2 are the absorbance of the sample, blank, and control, respectively. ## 2.2.2 Superoxide radical scavenging assay The test of the ability to scavenge superoxide radicals was carried out according to the method described by the previous study with minor modifications (Li, 2012). Briefly, the sample (1 mL) was introduced into a test tube containing 50 mM Tris-HCl buffer (pH 8.2, 4.5 mL) and 25 mM pyrogallol (0.4 mL) followed by incubation for 5 min at 25°C. Then, HCl (8 mM, 1 mL) was added to the mixture. Absorbance was measured at 420 nm. The scavenging inhibition (%) was calculated using the equation shown above. ## 2.3 Cell culture The reagents used in cell culture were purchased from Gibco. An insulinoma cell line (INS-1) from rats was cultured in RPMI-1640 medium. A pancreatic beta cell line (MIN6) from mice was grown in high-glucose Dulbecco’s modified Eagle’s medium (DMEM) containing $10\%$ fetal bovine serum, $1\%$ penicillin–streptomycin, and $0.2\%$ β-mercaptoethanol. Cells were maintained at 37°C in an atmosphere of $5\%$ CO2 in humidified air. To establish a model of oxidative stress in the β cell, INS-1 cells were cultured with alloxan (18 mM; Sigma, United States) or MIN6 cells were cultured with H2O2 (150 μM; Sigma) for 24h, which formed the oxidative stress (OS) model groups (Kimoto et al., 2003; Li et al., 2020). A stock solution of EA (40 mg/mL) was prepared. The main ingredients coumarin, cinnamyl alcohol, cinnamic acid, cinnamaldehyde (purity: HPLC ≥$98\%$) were purchased from Shanghai Yuanye Bio-Technology (Shanghai, China) and prepared as stock solutions (50 mM) and diluted to a final concentration of 50 µM. The concentration required for experimentation was prepared with medium immediately before use. ## 2.4 Measurement of cell viability Cell Counting Kit-8 (CCK-8; Beyotime Institute of Biotechnology, Shanghai, China) was used to determine the toxicity and protection of EA on INS-1 cells or MIN6 cells. INS-1 cells or MIN6 cells were seeded into 96-well plates at 5 × 104 cells/well and cultured overnight. Then, they were treated/not treated with alloxan/H2O2, and continued to be exposed to various concentrations of EA for 24 h. Then, CCK-8 solution (10 μL) was added. After incubation for 1 h at 37°C, absorbance was measured at 450 nm with a microplate reader. Each experiment was replicated in six wells. Cell viability was presented as a percentage of that of the control. Living cells can be stained with calcein-AM (Beyotime Institute of Biotechnology) but dead cells cannot. We rinsed cells with phosphate-buffered saline (PBS) and incubated them with calcein-AM (2 μM) for 30 min at 37°C. Images of cells were acquired with a fluorescence microscope (IX51 series; Olympus, Tokyo, Japan). ## 2.5 Biochemical analyses A Nitric Oxide Assay Kit (Beyotime Institute of Biotechnology) was used to quantify nitric oxide (NO) release according to manufacturer protocols. Briefly, standards of NaNO2 (50 μL; 0, 1, 2, 5, 10, 20, 40, 60, 100 μM) or cell-culture supernatants (50 μL) were dispensed into a 96-well plate. Then, Griess reagent I (50 μL) and Griess reagent II (50 μL) were added. After incubation for 3 min, absorbance was measured at 540 nm. NO production was calculated according to the standard curve and expressed as a percentage of that of the control. Commercial kits were used to assay for superoxide dismutase (SOD) activity (Beyotime Institute of Biotechnology), reduced glutathione (GSH) content (microplate method; Nanjing Jiancheng Bioengineering Institute, Nanjing, China), and malondialdehyde (MDA) content (colorimetric method; Nanjing Jiancheng Bioengineering Institute, Nanjing, China). Briefly, INS-1 cells or MIN6 cells were seeded at 1×106 cells/well in six-well plates and cultured overnight. After the indicated treatment, cells were washed twice with PBS, resuspended with PBS, sonicated on ice and centrifuged for 15 min at 4°C. Supernatants were used to measure the activity of SOD as well as levels of GSH and MDA. Finally, data were obtained under different wavelengths (532, 420, 405 nm) using a plate reader. Results were calculated with reference to a standard curve and expressed as nmol/mg protein for MDA, μmol/mg protein for GSH, and U/mg protein for SOD. ## 2.6 ROS determination The fluorescent probe 2′,7′-dichlorofluoresceindiacetate (DCFH-DA, Sigma-Aldrich, St. Louis, MO) was used to measure the ROS level. INS-1 cells or MIN6 cells were seeded 1 × 106 cells/well in six-well plates and cultured overnight. After the indicated treatment, cells were washed with Hank’s balanced salt solution (HBSS) and incubated with DCFH-DA (10 μM) for 30 min at 37°C in the dark. The fluorescence absorption of DCF was measured using a microplate reader (Fluoroskan™; Thermo Fisher Scientific, USA) at an excitation wavelength of 485 nm and emission wavelength of 538 nm. The fluorescence intensity was proportional to the amount of ROS generated intracellularly. ## 2.7 Flow cytometry An Annexin V-FITC Apoptosis Detection Kit (Beyotime Institute of Biotechnology) was used to quantify apoptosis according to standard procedures. Briefly, after the indicated treatment, binding buffer (195 μL) containing annexin V-fluorescein isothiocyanate (5 μL) and propidium iodide (10 μL) was added to a single-cell suspension and the reaction allowed to proceed for 15 min at room temperature in the dark. Then, percent apoptosis was evaluated by flow cytometry using a C6 Plus system (BD Biosciences, United States). A minimum of 10,000 cells were detected for each sample. ## 2.8 Caspase activity assay The activity of caspase-3 and caspase-9 was assessed using the corresponding Caspase Activity Assay Kit (Beyotime Institute of Biotechnology). In brief, cells in a six-well plate underwent the indicated treatment, were washed in PBS, suspended in lysis buffer (150 μL; provided in the kit) and centrifuged for 15 min at 4°C. The supernatant was harvested and mixed with peptide substrate (Ac-DEVD-pNA for caspase-3 and Ac-LEHD-pNA for caspase-9). Absorbance was measured using a microplate reader at 405 nm. ## 2.9 Western blotting Proteins were extracted using RIPA buffer (Beyotime Institute of Biotechnology) followed by addition of a fresh protease-inhibitor cocktail and phenylmethylsulfonyl fluoride. The protein concentration was determined using a BCA Protein Assay Kit (Boster Biological Technology, China). Proteins (30 μg) were separated by sodium dodecyl sulfate–polyacrylamide gel electrophoresis using $10\%$ gels, followed by transfer to polyvinylidene fluoride (PVDF) membranes (Millipore, USA). After blockade with $5\%$ non-fat milk in Tris-buffered saline–Tween buffer, the blot was probed with target-specific primary antibodies. The primary antibodies we used were those against nuclear factor erythroid 2 related factor 2 (Nrf2; catalog number: abs130481; Absin Bioscience, Shanghai, China), heme oxygenase 1 (HO-1; abs131494; Absin Bioscience), gamma glutamylcysteine synthetase (γ-GCSc; abs138070; Absin Bioscience), inducible nitric oxide synthase (iNOS; 22226-1-AP; Proteintech, United States), cleaved caspase-3 (9661S; Cell Signaling Technology, United States), Bax (AF1270; Beyotime Institute of Biotechnology), glyceraldehyde 3-phosphate dehydrogenase (GAPDH; 60004-1-Ig; Proteintech), and lamin B1 (abs155437; Absin Bioscience). Subsequently, PVDF membranes were incubated with the appropriate horseradish peroxidase-conjugated secondary antibody (Boster Biological Technology). Target proteins were visualized with BeyoECL Moon (Beyotime Institute of Biotechnology) using the Chemidoc XRS Gel Imaging System (Bio-Rad Laboratories, United States). ## 2.10 Insulin secretion assay INS-1 cells or MIN6 cells were seeded at 1 × 105 cells/well in 48-well plates, followed by treatment/non-treatment with EA or alloxan/H2O2 for 24 h. Then, cells were incubated in HBSS containing $0.2\%$ bovine serum albumin for 2 h. Then, the buffer was changed to HBSS containing basal glucose (5.5 mM) or glucose (5.5 mM) plus KCl (30 mM) followed by incubation for 1 h, respectively. The insulin content in the culture supernatant was measured by an Insulin ELISA Kit for rat or mice (Bio-Swamp, Wuhan, China). ## 2.11 Quantitative real-time PCR analysis TRIzol® Reagent was employed to purify the total RNA obtained from cells according to manufacturer instructions (Invitrogen, Carlsbad, CA, United States). RNA and (2 μg) was used to synthesize complimentary (c) DNA using the M-MLV Reverse Transcriptase Kit (Takara Biotechnology, Shiga, Japan). The obtained cDNAs were employed as templates to determine mRNA expression of the insulin transcription factor PDX-1 by real-time RT-qPCR using the TB Green® Premix Ex Taq™ Kit (Takara Biotechnology). Amplification was carried out using a thermocycler (CFX 96; Bio-Rad Laboratories). The primers for PDX-1 were 5′-CAAAGCTCACGCGTGGAAAA-3′ (sense) and 5′-CGAGGTTACGGCACAATCCT-3′ (antisense). The primers for β-actin were 5′-CACCCGCGAGTACAACCTTC-3′ (sense) and 5′- CCCATACCCACCCATCACACC-3′ (antisense). Relative expression of PDX-1 was calculated using the 2−ΔΔCT method. ## 2.12 Statistical analyses All assays were carried out in triplicate. Measurement data are the mean ± SEM. The Student’s t-test or one-way ANOVA was conducted to estimate significant differences using Prism 5 (GraphPad, La Jolla, CA, United States), $p \leq 0.05$ was considered significant. ## 3.1 EA fraction from CCE is rich in flavonoids We obtained EA fraction from CCE. Then, we determined the total flavonoid content and phenol content of EA: 89.93 ± 7.96 mg per gram RE equivalent and 34.26 + 0.78 mg per gram GAE equivalent (Table 1), which indicated that the total flavonoid content of the EA extract was higher than the total phenol content. Chemical assays were used to detect the antioxidant capacity of EA. The inhibition percent radical scavenging of EA indicated that EA containing abundant flavonoids had antioxidant capacity. TABLE 1 \| Sample \| Phenolic (mg GAE/g) \| Flavonoid (mg RE/g) \| Hydroxyl inhibition (%) \| Superoxide anion inhibition (%) \| \| --- \| --- \| --- \| --- \| --- \| \| EA \| 34.26 + 0.78 \| 89.93 + 7.96 \| 40.6 + 1.1 \| 68.1 + 2.0 \| \| VC \| — \| — \| 50.5 + 0.7 \| 71.8 + 0.9 \| ## 3.2 Protective effects of EA on β cell lines To determine the possible cytotoxicity of EA, cell viability was determined using the CCK-8 assay. INS-1 cells and MIN6 cells were incubated with EA (0–200 μg/mL) for 24 h. The CCK8 assay showed that EA did not cause cytotoxicity up to 100 μg/mL (Figure 1A). It has been reported that treatment with alloxan (18 mM) or H2O2 (150 μM) can induce β cells damage (Kimoto et al., 2003; Li et al., 2020). Hence, to evaluate if EA exerted protective effects upon β cells, we co-treated INS-1 cells or MIN6 cells with alloxan/H2O2 and EA. The proliferation of INS-1 cells was enhanced to $12.5\%$, $25.8\%$, $36.5\%$, and $36.25\%$ at concentrations of 12.5, 25, 50, and 100 μg/mL of EA, respectively, compared with that in the OS-model group (Figure 1B). For MIN6 cells treated with H2O2, the viability was enhanced to $10.2\%$, $15.1\%$, $17.2\%$, and $20.2\%$ at concentrations of 12.5, 25, 50, and 100 μg/mL of EA, respectively, compared with that in the OS-model group. Similarly, stimulation with alloxan/H2O2 in β cells resulted in a reduced mean fluorescence intensity of calcein-AM, which was reversed by treatment with EA (Figures 1C, D), which indicated that EA could increase β cells viability. These results indicated that EA was not toxic to β cells but also protected them from the damage induced by alloxan or H2O2. FIGURE 1: Cytotoxicity and cytoprotective effects of EA in β cell lines. INS-1 cells and MIN6 cells were exposed to EA (0–200 μg/mL) for 24 h (A), INS-1 cells were co-treated with the indicated concentrations of EA and alloxan (18 mM), or MIN6 cells were co-treated with the indicated concentrations of EA and H2O2 (150 μM) for 24 h, respectively (B), and then the cell viability was determined by the CCK-8 assay. (C,D) Morphologic observation of INS-1 cells and MIN6 cells treated as indicated and stained with calcein-AM. Representative images were acquired by a fluorescence microscope. Treatment of cells with EA markedly improved the morphologic changes of INS-1 cells caused by alloxan and in MIN6 cells caused by H2O2. Data are the mean ± SEM (n = 3). # p < 0.05 compared with control, ## p < 0.01 compared with control, p < 0.05 compared with the OS-model group, *p < 0.01 compared with the OS-model group. ## 3.3 Protective effects of EA against oxidative stress in β cells Studies have demonstrated that oxidative stress caused by increased ROS generation is the primary cause of β cell damage (Dinić et al., 2022). We wondered if EA affects ROS production under alloxan/H2O2 treatment. We detected the effect of EA on ROS production with the fluorescent probe DCFH-DA in the β cells co-treated with EA and alloxan or H2O2 for 24 h. EA suppressed ROS production significantly (Figure 2A). The antioxidant capacity of EA was evaluated by measuring the activity or level of markers of ROS-mediated injury: MDA, SOD, and GSH. In the β cells co-treated with EA, and alloxan or H2O2 for 24 h, the MDA level was suppressed, and SOD activity and GSH content recovered, by dose-dependent treatment with EA. These data implied that EA treatment improved endogenous antioxidant activities and inhibited lipid peroxidation significantly ($p \leq 0.05$ vs. OS-model group) (Figures 2B–D). These findings suggested that EA appeared to have a protective action via the improvement of antioxidant capacities and reducing oxidative stress by suppressing ROS and MDA production in β cells. FIGURE 2: EA protects β cells from oxidative stress damage. INS-1 cells were incubated with EA (12.5–100 μg/mL) and alloxan for 24 h, MIN6 cells were incubated with EA (12.5–100 μg/mL) and H2O2 for 24 h. Levels of oxidative stress-related makers were determined. (A) Analysis of the mean fluorescence intensity of DCF (indicator of reactive oxygen species). Effect of EA on MDA production (B), SOD activity (C), and on the level of reduced GSH (D) in oxidative stress-treated β cells. (E,F) Protein expression of Nrf 2, γ-GCSc, and HO-1 of alloxan (18 mM)-treated cells (E) and H2O2 (150 μM)-treated cells (F) was measured by western blotting. Data are the mean ± SEM (n = 3). # p < 0.05 compared with control, ## p < 0.01 compared with control, p < 0.05 compared with the OS-model group, *p < 0.01 compared with the OS-model group. Nrf2/HO-1 is a vital signaling pathway that regulates oxidative stress in cells. We wished to ascertain whether the antioxidant effect of EA correlated with regulation of the Nrf2/HO-1 signaling pathway. Hence, we measured expression of Nrf2 and its downstream proteins (HO-1, γ-GCSc) in INS-1 cells and MIN6 cells co-treated with alloxan or H2O2 and the indicated EA concentration. Western blotting showed that protein expression of Nrf2, γ-GCSc, and HO-1 was decreased in alloxan- or H2O2-treated cells, but was increased significantly under EA treatment (Figures 2E, F). These results implied that EA could suppress the oxidative stresses induced by alloxan/H2O2 by activating the Nrf2/HO-1 pathway. ## 3.4 EA protects β cells from apoptosis Apoptosis is an important result of OS. Hence, we investigated if EA could suppress oxidative stress-induced apoptosis. Flow cytometry showed that percent apoptosis of β cells was increased significantly under treatment with alloxan or H2O2 (Figures 3A, B). However, the addition of EA reduced the number of apoptotic cells markedly. To confirm these results, we measured the activity of caspase-3 and caspase-9 in treated cells: EA reversed the higher activity induced by treatment with alloxan or H2O2 (Figures 3C, D). Excessive production of NO is regarded as one of the critical molecular mechanisms leading to apoptosis of pancreatic β cells (Oyadomari et al., 2001). EA treatment reduced NO production in pancreatic β cells induced by treatment with alloxan or H2O2 (Figure 3F). NO release and apoptosis induction were further confirmed by protein expression of iNOS, cleaved caspase-3, and Bax by western blotting. Addition of EA (12.5–100 μg/mL) could suppress expression of iNOS, cleaved caspase-3, and Bax induced by treatment with alloxan or H2O2 (Figures 3G, H). Collectively, these data suggested that EA had protective effects against the apoptosis of pancreatic β cells induced by oxidative stress. FIGURE 3: Effect of EA on apoptosis of β cells. INS-1 cells were incubated with EA (12.5–100 μg/mL) and alloxan for 24 h. MIN6 cells were incubated with EA (12.5–100 μg/mL) and H2O2 for 24 h. Levels of apoptosis-related makers were determined. (A–C) Effect of EA on apoptosis was detected with flow cytometry, (D,E) Effect of EA on the activity of caspase-3 and caspase-9. (F) Effect of EA on NO release into the supernatant. (G,H) Western blots showing protein expression of iNOS, Bax, and cleaved caspase-3. Data are the mean ± SEM, n = 3. # p < 0.05 compared with control, ## p < 0.01 compared with control, p < 0.05 compared with the OS-model group, *p < 0.01 compared with the OS-model group. ## 3.5 EA treatment enhances insulin secretion of β cells Insulin content is an essential indicator for evaluating pancreatic β cells function. Hence, we measured the level of insulin secretion of the β cells under EA protection. We employed a common model (high insulin secretion induced by treatment with KCl (30 mM) or low insulin secretion induced by treatment with glucose (5.5 mM)) (Yang et al., 2021) to investigate the effect of treatment with EA and alloxan or H2O2 upon insulin secretion by measuring insulin content in the supernatant of β cells using ELISAs. Insulin secretion decreased in oxidative stress-induced pancreatic β cells, but recovered under co-treatment with EA (Figures 4A, B). Flavonoids upregulate the protein expression (including PDX-1 expression) involved in pancreatic β cell function (Lee et al., 2021). Protein expression of PDX-1 increased markedly following EA treatment (Figures 4C, D). These results indicated that EA might stimulate insulin secretion by increasing PDX-1 expression. To test this hypothesis, we investigated the effects of EA treatment on basal glucose-stimulated insulin secretion of β cells. We treated INS-1 cells with EA alone and measured the mRNA expression of PDX-1 and insulin level. Surprisingly, the level of insulin of glucose-treated cells increased significantly upon EA treatment. Simultaneously, PDX-1 expression increased in an EA concentration-dependent manner (Figures 4E, F). Collectively, these results demonstrated that EA could protect β cells from apoptosis but also improve pancreatic β-cell function by activating insulin synthesis. FIGURE 4: EA increases insulin secretion via enhancing PDX-1 expression. INS-1 cells were incubated with EA (12.5–100 μg/mL) and alloxan for 24 h. MIN6 cells were incubated with EA (12.5–100 μg/mL) and H2O2 for 24 h (A,B) Effect of EA on KCl-stimulated insulin secretion in INS-1 cells and MIN6 cells under oxidative stress-induced toxicity was detected using ELISA. (C,D) Effect of EA on protein expression of PDX-1 in INS-1 cells and MIN6 cells under oxidative stress-induced toxicity was detected by western blotting. (E) Effect of EA on basal glucose-stimulated insulin secretion in INS-1 cells was determined by ELISA. (F) Effect of EA on mRNA expression of PDX-1 in INS-1 cells was detected by qPCR. Data are the mean ± SEM. n = 3. # p < 0.05 compared with control, ## p < 0.01 compared with control, p < 0.05 compared with the OS-model group, *p < 0.01 compared with the OS-model group. ## 3.6 Bioactive components of EA We carried out HPLC to elucidate the major functional compounds of EA: four major peaks appeared (Figures 5A, B). A previous study demonstrated that the chemical constituents of EA were primarily cinnamyl alcohol, coumarin, cinnamic acid, and cinnamaldehyde (Al-Dhubiab, 2012). Our HPLC results showed that EA mainly comprised (mg/g) coumarin (7.32), cinnamyl alcohol (7.87), cinnamic acid (4.37), and cinnamaldehyde (0.65). Therefore, we treated INS-1 cells and MIN6 cells with these four compounds individually (50 µM) for 24 h. Cell-viability assays showed that none of these four compounds had obvious toxicity towards β cells (Figure 5C). Treatment with coumarin, cinnamic acid, or cinnamic alcohol could recover the cell proliferation damaged by alloxan (Figure 5D). Moreover, these three compounds provoked an increase in KCl-stimulated insulin secretion compared with the OS-model group, which was similar to the effect of EA (Figure 5E). Coumarin, cinnamic acid, and cinnamic alcohol also increased basal glucose-stimulated insulin secretion in INS-1 cells. Then, the protein and mRNA expression of PDX1 was measured: treatment with any of these three compounds elicited a significant increase compared with that of the control (Figures 5F, G). To confirm the effect of coumarin, cinnamic acid, and cinnamic alcohol on alloxan- or H2O2 induced oxidative stress, the parameters of related markers (NO production, SOD activity, and level of reduced GSH) were measured. Similar to EA treatment, therapy with any of these three compounds suppressed NO release (Figures 5H–J), and maintained SOD activity and the GSH level. Furthermore, treatment with coumarin, cinnamic acid, or cinnamic alcohol increased expression of the antioxidant proteins Nrf2, γ-GCSc, and HO-1, while suppressing expression of the apoptosis-associated proteins iNOS, cleaved caspase 3, and BAX, in INS-1 cells treated with alloxan (Figure 5K). Notably, these three compounds promoted remarkable antioxidative capacity. Taken together, these data suggested that treatment with coumarin, cinnamic acid, or cinnamic alcohol from EA had an obvious protective effect on oxidative stress-induced β cell damage. FIGURE 5: Major compounds of EA and their protective effect on β cells. Chemical profiles of EA were analyzed through HPLC and the effects of major functional compounds of EA on INS-1 cells were detected. (A,B) Chromatographic patterns from HPLC of EA shows peaks corresponding to retention times (min). INS-1 cells were incubated/not incubated with the main compounds (50 μM) of EA and alloxan for 24 h. The effect of compounds on the viability of INS-1 cells was determined by the CCK-8 assay (C,D). The effect of three compounds (coumarin, cinnamic acid, cinnamic alcohol) in EA on KCl-stimulated insulin secretion in INS-1 cells under alloxan-induced toxicity (E), basal glucose-stimulated insulin secretion in INS-1 cells (F), mRNA expression of PDX1 in INS-1 cells (G), NO release (H), activity of SOD (I), level of reduced GSH (J), expression of antioxidant- and apoptosis-related proteins (K). Data are the mean ± SEM. n = 3. # p < 0.05 compared with control, ## p < 0.01 compared with control, p < 0.05 compared with the OS-model group, *p < 0.01 compared with the OS-model group. ## 4 Discussion T2DM is influenced primarily by the eating habits and sedentary lifestyles of some people. It is characterized by disorders of the metabolism of glucose and lipids, with deactivation of the insulin signaling pathway and abnormally increased blood glucose levels (Kahn, 2003). Pancreatic β cells are crucial for maintaining glucose homeostasis in the body by generating insulin. Studies have indicated that various extracts of cinnamon (ether, aqueous, methanolic) could be antioxidants (Mancini-Filho et al., 1998). Different flavonoids isolated from cinnamon have free radical-scavenging activities (Okawa et al., 2001). We discovered that EA constituents from cinnamon were rich in flavonoids. We demonstrated that EA has antioxidant activity in two models. Hyperglycemia promotes oxidative stress (Bravard et al., 2011; Choudhuri et al., 2013), which causes the oxidation of lipids, proteins, and DNA, thereby leading to β cell damage. Enhanced oxidative stress plays a crucial part in DM pathogenesis. It contributes to cellular damage by increasing levels of oxidative-stress markers and reducing antioxidant levels in rodents and patients suffering from DM (Čater and Krizancic Bombek, 2022). Therefore, inhibiting ROS is important for preventing DM. It has been reported that Cinnamomum zeylanicum extracts reduced Aβ(1–42)-induced ROS production (Althobaiti et al., 2022). Recently, flavonoid compounds have attracted great interest for their potential use in DM treatment due to their remarkable antioxidant capacity. The proposed mechanism of flavonoids on antioxidant or anti-inflammatory activities might be: reducing ROS levels; activating anti-apoptosis pathways; inhibiting generation of NO (Ghorbani et al., 2019). We showed that EA from cinnamon reduced ROS production significantly in damaged β cells. The MDA level can reflect the level of lipid-associated ROS. There are two types of antioxidant systems in the body: enzyme antioxidant system (including SOD) and the non-enzymatic antioxidant system (including GSH). In the present study, a significant increase in ROS accumulation and remarkable reductions in levels of antioxidative enzymes (e.g., SOD, reduced GSH) were observed in INS-1 cells and MIN6 cells after exposure to oxidative stress induced by alloxan or H2O2. Nonetheless, these oxidative stress-induced cellular events were blocked to a great extent when β cells were co-incubated with EA. These results suggested that enhancement of the endogenous antioxidant system, inhibition of intracellular release of ROS, and attenuation of lipid peroxidation may represent important mechanisms of cellular protection elicited by EA, which might be due mainly to its chemical characteristics. A growing body of work has indicated that activation of Nrf2 could regulate redox homeostasis, thus blocking DM pathogenesis (Behl et al., 2021). Our results demonstrated that EA treatment activated Nrf2 and its downstream cytoprotective genes (HO-1, γ-GCSc), which were verified to be Nrf2 activators. Hence, activating Nrf2 by EA could protect and activate β cells. A simultaneous increase in ROS accumulation is associated with an increase in NO production. Excessive and sustained generation of NO derived from iNOS plays an important part in induction of β cell apoptosis in DM pathogenesis (Wang et al., 2014). A previous study found that cinnamaldehyde could prevent NO production and iNOS expression (Lee et al., 2002). We found that EA could significantly decrease the production of NO and expression of iNOS induced by treatment with alloxan or H2O2. In addition, it has been suggested that an increase in the level of pro-apoptotic proteins is associated with apoptosis. To further elucidate the anti-apoptosis effect of EA, we explored the possible effect of EA treatment on the activity of caspase-3 and caspase-9, as well as expression of cleaved caspase-3 and Bax. We demonstrated that EA treatment downregulated expression of iNOS. Our results further support the notion that EA has a pivotal anti-apoptotic role in DM. It is believed that PDX-1 is an important target in the fight against DM. There is evidence that EA has significant protective effects upon β cell viability, possibly due to a shielding effect on insulin synthesis. PDX-1 has been reported to control the secretion of insulin as well as other β cell-specific genes. As expected, we found that EA treatment increased PDX-1 expression and insulin secretion, which implied that EA increases insulin synthesis in β cells. The beneficial effect of EA on promoting PDX-1 expression needs further investigation. Our study suggested that EA has a robust protective role against damage of β cells, so we explored which monomer in EA is responsible. HPLC results showed the more abundant monomers are cinnamyl alcohol, coumarin, cinnamic acid, and cinnamaldehyde, they are phenylpropanoids. Further study revealed coumarin, cinnamic acid, and cinnamyl alcohol had protective effects upon β cells. Another compound from coumarin, scopoletin, has been reported to regulate glucose metabolism by enhancing the activities of antioxidant enzymes (Kalpana et al., 2019). In addition, cinnamic acid isolated from the hydro-alcohol extract of Cinnamomum cassia has been found to activate glucose transport in L6 myotubes through involvement of glucose transporter-4 via a phosphoinositide 3-kinase-independent pathway (Sangeetha et al., 2017). In a related study, cinnamyl alcohols were found to interact with the peroxisome proliferator-activated receptor gamma receptor, which is a key regulator of the metabolism and storage of lipids and glucose (Genovese et al., 2021). The studies mentioned above indicated that coumarin, cinnamic acid, and cinnamyl alcohol could have protective roles in DM, and our results provide more detailed explanations for these function. Future studies should focus on the detailed mechanism of action of these three monomers. Moreover, the chemical composition of cinnamon is relatively complex, whether the analogs of these ingredients in cinnamon also have similar activity warrants investigation. Our results in Table1 showed that the EA is rich in flavonoids, and we also found many small peaks in the HPLC analysis, which might be flavonoids with variety and low content. The flavonoids in this extract will be identified through UHPLC-MS/MS and further to elucidate the effect of cinnamon on antidiabetic activity in the next study. We demonstrated the antioxidant function and possible mechanism of action of EA and the main compounds from cinnamon using two cell lines, it is still necessary to confirm our findings with in vivo research. Although many natural antioxidants from plants such as phenylpropanoids and flavonoids have been shown to be helpful to human health from different perspectives, their low solubility, poor absorption, rapid metabolism, and low bioavailability limit their clinical application (Teng et al., 2023). Hence, further structural modifications and technological approaches are needed to overcome this issue (Teng et al., 2021; Teng et al., 2021; Lyu et al., 2022). ## 5 Conclusion Our study explained how EA protects β-cell from oxidative stress damage and improves β-cell function. EA treatment could prevent oxidative stress-induced apoptosis by blocking ROS accumulation and enhancing antioxidant capacities via activation of the Nrf2/HO-1 signaling pathway in two pancreatic β cell line models. EA might also be able to increase PDX-1 expression and insulin secretion. This study contributes to deeper understanding of the molecular and biological mechanisms of EA from CCE and its anti-DM functions. ## 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 BS and WL conceived and designed the experiments. WLL, JQ, KL, PS, YW, XY, and QP conducted the experiments and statistical analyses. WLL and JQ prepared the original draft of the manuscript. WLL, JQ, KL, PS, and YW prepared figures and tables. BS, QP, XY, and WL provided advice and reviewed the manuscript. All authors contributed to the manuscript 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. ## Abbreviations CCE, Cinnamon cortex extract; CCK-8, Cell Counting Kit-8; DCFH-DA, 2′,7′-dichlorofluoresceindiacetate; DM, Diabetes mellitus; DMEM, Dulbecco’s modified Eagle’s medium; EA, ethyl acetate fraction; ELISA, Enzyme-linked immunosorbent assay; GSH, reduced glutathione; GCS-γ, gamma glutamylcysteine synthetase; HBSS, Hank’s balanced salt solution; HO-1, heme oxygenase 1; iNOS, inducible nitric oxide synthase; MDA, Malondialdehyde; NO, Nitric oxide; Nrf2, nuclear factor erythroid 2 related factor 2; PVDF, polyvinylidene fluoride; PBS, phosphate-buffered saline; ROS, Reactive oxygen species; SOD, Superoxide dismutase; VC, vitamin C. ## References 1. Adisakwattana S., Lerdsuwankij O., Poputtachai U., Minipun A., Suparpprom C.. Inhibitory activity of cinnamon bark species and their combination effect with acarbose against intestinal α-glucosidase and pancreatic α-amylase. Plant Foods Hum. Nutr. (2011) 66 143-148. DOI: 10.1007/s11130-011-0226-4 2. Al-Dhubiab B. E.. Pharmaceutical applications and phytochemical profile of Cinnamomum burmannii. Pharmacogn. Rev. (2012) 6 125-131. DOI: 10.4103/0973-7847.99946 3. Al-Ishaq R. K., Abotaleb M., Kubatka P., Kajo K., Büsselberg D.. Flavonoids and their anti-diabetic effects: Cellular mechanisms and effects to improve blood sugar levels. Biomolecules (2019) 9 430. DOI: 10.3390/biom9090430 4. Althobaiti N. A., Menaa F., Dalzell J. J., Albalawi A. E., Ismail H., Alghuthaymi M. A.. Ethnomedicinal plants with protective effects against beta-amyloid peptide (Aβ)1-42 indicate therapeutic potential in a new. Antioxidants (Basel). (2022) 11 1865. DOI: 10.3390/antiox11101865 5. Babu P. V., Liu D., Gilbert E. R.. Recent advances in understanding the anti-diabetic actions of dietary flavonoids. J. Nutr. Biochem. (2013) 24 1777-1789. DOI: 10.1016/j.jnutbio.2013.06.003 6. Baumel-Alterzon S., Scott D. K.. Regulation of Pdx1 by oxidative stress and Nrf2 in pancreatic beta-cells. Front. Endocrinol. (Lausanne) (2022) 13 1011187. DOI: 10.3389/fendo.2022.1011187 7. Behl T., Kaur I., Sehgal A., Sharma E., Kumar A., Grover M.. Unfolding Nrf2 in diabetes mellitus. Mol. Biol. Rep. (2021) 48 927-939. DOI: 10.1007/s11033-020-06081-3 8. Black H. S.. A synopsis of the associations of oxidative stress, ROS, and antioxidants with diabetes mellitus. Antioxidants (Basel) (2022) 11 2003. DOI: 10.3390/antiox11102003 9. Bravard A., Bonnard C., Durand A., Chauvin M. A., Favier R., Vidal H.. Inhibition of xanthine oxidase reduces hyperglycemia-induced oxidative stress and improves mitochondrial alterations in skeletal muscle of diabetic mice. Am. J. Physiol. Endocrinol. Metab. (2011) 300 E581-E591. DOI: 10.1152/ajpendo.00455.2010 10. Čater M., Križančić Bombek L.. Protective role of mitochondrial uncoupling proteins against age-related oxidative stress in type 2 diabetes mellitus. Antioxidants (Basel) (2022) 11 1473. DOI: 10.3390/antiox11081473 11. Choudhuri S., Dutta D., Chowdhury I. H., Mitra B., Sen A., Mandal L. K.. Association of hyperglycemia mediated increased advanced glycation and erythrocyte antioxidant enzyme activity in different stages of diabetic retinopathy. Diabetes Res. Clin. Pract. (2013) 100 376-384. DOI: 10.1016/j.diabres.2013.03.031 12. Darenskaya M. A., Kolesnikova L. I., Kolesnikov S. I.. Oxidative stress: Pathogenetic role in diabetes mellitus and its complications and therapeutic approaches to correction. Bull. Exp. Biol. Med. (2021) 171 179-189. DOI: 10.1007/s10517-021-05191-7 13. Demir S., Nawroth P. P., Herzig S., Ekim Üstünel B.. Emerging targets in type 2 diabetes and diabetic complications. Adv. Sci. (Weinh). (2021) 8 e2100275. DOI: 10.1002/advs.202100275 14. Dinić S., Arambašić Jovanović J., Uskoković A., Mihailović M., Grdović N., Tolić A.. Oxidative stress-mediated beta cell death and dysfunction as a target for diabetes management. Front. Endocrinol. (Lausanne). (2022) 13 1006376. DOI: 10.3389/fendo.2022.1006376 15. El-Ouady F., Bachir F., Eddouks M.. Flavonoids extracted from asteriscus graveolens improve glucose metabo-lism and lipid profile in diabetic rats. Endocr. Metab. Immune Disord. Drug Targets (2021) 21 895-904. DOI: 10.2174/1871530320999200818103709 16. Farag M. A., Kabbash E. M., Mediani A., Döll S., Esatbeyoglu T., Afifi S. M.. Comparative metabolite fingerprinting of four different cinnamon species analyzed via UPLC-MS and GC-MS and chemometric tools. Molecules (2022) 27 2935. DOI: 10.3390/molecules27092935 17. Genovese S., Epifano F., Rollinger J. M., Fiorito S.. PPARγ transcription effect on naturally occurring O-prenyl cinnamaldehydes and cinnamyl alcohol derivatives. Future Med. Chem. (2021) 13 1175-1183. DOI: 10.4155/fmc-2021-0054 18. Ghorbani A., Rashidi R., Shafiee-Nick R.. Flavonoids for preserving pancreatic beta cell survival and function: A mechanistic review. Biomed. Pharmacother. (2019) 111 947-957. DOI: 10.1016/j.biopha.2018.12.127 19. Hayward N. J., McDougall G. J., Farag S., Allwood J. W., Austin C., Campbell F.. Cinnamon shows antidiabetic properties that are species-specific: Effects on enzyme activity inhibition and starch digestion. Plant Foods Hum. Nutr. (2019) 74 544-552. DOI: 10.1007/s11130-019-00760-8 20. Hussain T., Tan B., Murtaza G., Liu G., Rahu N., Saleem Kalhoro M.. Flavonoids and type 2 diabetes: Evidence of efficacy in clinical and animal studies and delivery strategies to enhance their therapeutic efficacy. Pharmacol. Res. (2020) 152 104629. DOI: 10.1016/j.phrs.2020.104629 21. Kahn S. E.. The relative contributions of insulin resistance and beta-cell dysfunction to the pathophysiology of Type 2 diabetes. Diabetologia (2003) 46 3-19. DOI: 10.1007/s00125-002-1009-0 22. Kalpana K., Sathiya Priya C., Dipti N., Vidhya R., Anuradha C. V.. Supplementation of scopoletin improves insulin sensitivity by attenuating the derangements of insulin signaling through AMPK. Mol. Cell Biochem. (2019) 453 65-78. DOI: 10.1007/s11010-018-3432-7 23. Kimoto K., Suzuki K., Kizaki T., Hitomi Y., Ishida H., Katsuta H.. Gliclazide protects pancreatic beta-cells from damage by hydrogen peroxide. Biochem. Biophys. Res. Commun. (2003) 303 112-119. DOI: 10.1016/s0006-291x(03)00310-3 24. Lee D., Kim K. H., Jang T., Kang K. S.. (-)-Leucophyllone, a tirucallane triterpenoid from cornus walteri, enhances insulin secretion in INS-1 cells. Plants (Basel) (2021) 10 431. DOI: 10.3390/plants10030431 25. Lee H. S., Kim B. S., Kim M. K.. Suppression effect of Cinnamomum cassia bark-derived component on nitric oxide synthase. J. Agric. Food Chem. (2002) 50 7700-7703. DOI: 10.1021/jf020751f 26. Li J. E., Futawaka K., Yamamoto H., Kasahara M., Tagami T., Liu T. H.. Cinnamaldehyde contributes to insulin sensitivity by activating PPARδ, PPARγ, and RXR. Am. J. Chin. Med. (2015) 43 879-892. DOI: 10.1142/S0192415X15500512 27. Li W., Lin K., Zhou M., Xiong Q., Li C., Ru Q.. Polysaccharides from Opuntia milpa alta alleviate alloxan-induced INS-1 cells apoptosis via reducing oxidative stress and upregulating Nrf2 expression. Nutr. Res. (2020) 77 108-118. DOI: 10.1016/j.nutres.2020.02.004 28. Li X.. Improved pyrogallol autoxidation method: A reliable and cheap superoxide-scavenging assay suitable for all antioxidants. J. Agric. Food Chem. (2012) 60 6418-6424. DOI: 10.1021/jf204970r 29. Lu Z., Jia Q., Wang R., Wu X., Wu Y., Huang C.. Hypoglycemic activities of A- and B-type procyanidin oligomer-rich extracts from different Cinnamon barks. Phytomedicine (2011) 18 298-302. DOI: 10.1016/j.phymed.2010.08.008 30. Lyu Q., Chen L., Lin S., Cao H., Teng H.. A designed self-microemulsion delivery system for dihydromyricetin and its dietary intervention effect on high-fat-diet fed mice. Food Chem. (2022) 390 132954. DOI: 10.1016/j.foodchem.2022.132954 31. Mancini-Filho J., Van-Koiij A., Mancini D. A., Cozzolino F. F., Torres R. P.. Antioxidant activity of cinnamon (Cinnamomum zeylanicum, breyne) extracts. Boll. Chim. Farm (1998) 137 443-447. PMID: 10077878 32. Mang B., Wolters M., Schmitt B., Kelb K., Lichtinghagen R., Stichtenoth D. O.. Effects of a cinnamon extract on plasma glucose, HbA, and serum lipids in diabetes mellitus type 2. Eur. J. Clin. Invest. (2006) 36 340-344. DOI: 10.1111/j.1365-2362.2006.01629.x 33. Mathis D., Vence L., Benoist C.. beta-Cell death during progression to diabetes. Nature (2001) 414 792-798. DOI: 10.1038/414792a 34. Mohamed Sham Shihabudeen H., Hansi Priscilla D., Thirumurugan K.. Cinnamon extract inhibits α-glucosidase activity and dampens postprandial glucose excursion in diabetic rats. Nutr. Metab. (Lond). (2011) 8 46. DOI: 10.1186/1743-7075-8-46 35. Oguntibeju O. O.. Type 2 diabetes mellitus, oxidative stress and inflammation: Examining the links. Int. J. Physiol. Pathophysiol. Pharmacol. (2019) 11 45-63. PMID: 31333808 36. Okawa M., Kinjo J., Nohara T., Ono M.. DPPH (1,1-diphenyl-2-picrylhydrazyl) radical scavenging activity of flavonoids obtained from some medicinal plants. Biol. Pharm. Bull. (2001) 24 1202-1205. DOI: 10.1248/bpb.24.1202 37. Oyadomari S., Takeda K., Takiguchi M., Gotoh T., Matsumoto M., Wada I.. Nitric oxide-induced apoptosis in pancreatic beta cells is mediated by the endoplasmic reticulum stress pathway. Proc. Natl. Acad. Sci. U. S. A. (2001) 98 10845-10850. DOI: 10.1073/pnas.191207498 38. Qureshi A. S., Ghaffor J., Usman M., Ehsan N., Umar Z., Sarfraz A.. Effect of ethanolic preparations of cinnamon (Cinnamomum zeylanicum) extract on hematologic and histometric parameters of selected organs in Alloxan® induced diabetic female albino rats. J. Diabetes Metab. Disord. (2019) 18 505-512. DOI: 10.1007/s40200-019-00457-4 39. Sangeetha K. N., Sujatha S., Muthusamy V. S., Anand S., Shilpa K., Kumari P. J.. Current trends in small molecule discovery targeting key cellular signaling events towards the combined management of diabetes and obesity. Bioinformation (2017) 13 394-399. DOI: 10.6026/97320630013394 40. Sun P., Li K., Wang T., Ji J., Wang Y., Chen K. X.. Procyanidin C1, a component of cinnamon extracts, is a potential insulin sensitizer that targets adipocytes. J. Agric. Food Chem. (2019) 67 8839-8846. DOI: 10.1021/acs.jafc.9b02932 41. Teng H., Mi Y., Cao H., Chen L.. Enzymatic acylation of raspberry anthocyanin: Evaluations on its stability and oxidative stress prevention. Food Chem. (2021) 372 130766. DOI: 10.1016/j.foodchem.2021.130766 42. Teng H., Zheng Y., Cao H., Huang Q., Xiao J., Chen L.. Enhancement of bioavailability and bioactivity of diet-derived flavonoids by application of nanotechnology: A review. Crit. Rev. Food Sci. Nutr. (2023) 63 378-393. DOI: 10.1080/10408398.2021.1947772 43. Wang T., Sun P., Chen L., Huang Q., Chen K., Jia Q.. Cinnamtannin D-1 protects pancreatic β-cells from palmitic acid-induced apoptosis by attenuating oxidative stress. J. Agric. Food Chem. (2014) 62 5038-5045. DOI: 10.1021/jf500387d 44. Wang Y., Li M., Yu X., He S., Wu X., Wang Y.. Mulberry leaf flavonoids protect against glucotoxicity-induced INS-1 cell apoptosis. J. Tradit. Chin. Med. (2019) 39 153-159. PMID: 32186036 45. Yang S., Qu Y., Chen J., Chen S., Sun L., Zhou Y.. Bee pollen polysaccharide from rosa rugosa thunb. (Rosaceae) promotes pancreatic β-cell proliferation and insulin secretion. Front. Pharmacol. (2021) 12 688073. DOI: 10.3389/fphar.2021.688073 46. Yoshioka Y., Kono R., Kuse M., Yamashita Y., Ashida H.. Phenylpropanoids and neolignans isolated from Myristica fragrans enhance glucose uptake in myotubes. Food Funct. (2022) 13 3879-3893. DOI: 10.1039/d1fo04408g 47. Zhang P., Li T., Wu X., Nice E. C., Huang C., Zhang Y.. Oxidative stress and diabetes: Antioxidative strategies. Front. Med. (2020) 14 583-600. DOI: 10.1007/s11684-019-0729-1 48. Zhang Y., Fang H., Xie Q., Sun J., Liu R., Hong Z.. Comparative evaluation of the radical-scavenging activities of fucoxanthin and its stereoisomers. Molecules (2014) 19 2100-2113. DOI: 10.3390/molecules19022100 49. Zhou Q., Lei X., Fu S., Li Z., Chen Y., Long C.. Efficacy of cinnamon supplementation on glycolipid metabolism in T2DM diabetes: A meta-analysis and systematic review. Front. Physiol. (2022) 13 960580. DOI: 10.3389/fphys.2022.960580
--- title: Compared effectiveness of sodium zirconium cyclosilicate and calcium polystyrene sulfonate on hyperkalemia in patients with chronic kidney disease authors: - Takashin Nakayama - Shintaro Yamaguchi - Kaori Hayashi - Kiyotaka Uchiyama - Takaya Tajima - Tatsuhiko Azegami - Kohkichi Morimoto - Tadashi Yoshida - Jun Yoshino - Toshiaki Monkawa - Takeshi Kanda - Hiroshi Itoh journal: Frontiers in Medicine year: 2023 pmcid: PMC10025387 doi: 10.3389/fmed.2023.1137981 license: CC BY 4.0 --- # Compared effectiveness of sodium zirconium cyclosilicate and calcium polystyrene sulfonate on hyperkalemia in patients with chronic kidney disease ## Abstract Hyperkalemia is a well-recognized electrolyte abnormality in patients with chronic kidney disease (CKD). Potassium binders are often used to prevent and treat hyperkalemia. However, few studies have evaluated the difference in serum potassium (K+) level-lowering effect during the post-acute phase between the novel potassium binder, sodium zirconium cyclosilicate (ZSC), and conventional agents. This retrospective study included patients who received potassium binders (either ZSC or calcium polystyrene sulfonate [CPS]) in our hospital between May 2020 and July 2022. The patients were divided into the ZSC and CPS groups. After propensity score matching, we compared changes from baseline to the first follow-up point, at least 4 weeks after initiating potassium binders, in electrolytes including K+ level between the two groups. Of the 132 patients, ZSC and CPS were administered in 48 and 84 patients, respectively. After matching, 38 patients were allocated to each group. The ZSC group showed greater reduction in K+ levels than did the CPS group ($P \leq 0.05$). Moreover, a significant increase in serum sodium minus chloride levels, a surrogate marker for metabolic acidosis, was observed in the ZSC group ($P \leq 0.05$). Our results demonstrated that ZSC could potentially improve hyperkalemia and metabolic acidosis in patients with CKD. ## Introduction Hyperkalemia is a common complication in patients with chronic kidney disease (CKD) and heart failure, particularly with the use of renin-angiotensin system (RAS) inhibitors, which are the cornerstone of treatment of these diseases (1–4). This electrolyte abnormality is strongly associated with increased mortality and healthcare costs in the short and long term. Therefore, optimal management of serum potassium (K+) levels, which reduces clinical complications and economic burden, is crucially important (5–9). Dietary potassium restriction has been recommended for preventing and treating hyperkalemia in patients with CKD [10]. However, adherence to renal diet restrictions is likely to be inadequate even in patients with sufficient knowledge [11]. As such, use of potassium binders, which bind potassium in the gastrointestinal tract to increase its fecal elimination, is also a therapeutic option for hyperkalemia. Although organic polymer resins such as calcium polystyrene sulfonate (CPS) and sodium polystyrene sulfonate have been conventionally used, their efficacy and safety have been concerned. Utilizing these resins, gastrointestinal toxicity is common as treatment-related adverse events. Furthermore, recurrent hyperkalemia episodes over a short period of time are frequent [12, 13]. Therefore, clinicians often make decision with compromise to discontinue or down-titrate RAS inhibitors, which could increase the risk of death and cardiovascular events [7, 8, 14, 15]. Sodium zirconium cyclosilicate (ZSC), a novel potassium binder, was approved for use in Japan on March 26, 2020. Compared with conventional nonspecific organic polymer resins, this inorganic cation exchange compound has a capacity for more selectively and efficiently capturing monovalent cations, particularly excess potassium and ammonium (NH4+) [16, 17]. Additionally, ZSC does not absorb water, which may reduce the risk of potassium binder-related constipation. Importantly, unlike conventional potassium binders requiring frequent medications, ZSC is essentially administered as a once-daily dose, leading to improved adherence to medication [18]. Although these profiles imply that ZSC could be an attractive therapeutic agent, few studies have evaluated its efficacy and safety in real–world practice settings. Therefore, we conducted this retrospective study to compare the effect of ZSC and conventional potassium binders on changes in electrolytes and kidney function in patients with CKD. ## Study participants Considering the prescription status at our hospital, we defined CPS as a conventional potassium binder. All patients aged ≥18 years who have newly started ZSC or CPS in our hospital between May 2020 and July 2022 were included in the study. CPS comprised ARGAMATE $20\%$ JELLY 25 g (Astellas Pharma Inc.), CALCIUM POLYSTYRENE SULFONATE $20\%$ Oral Jelly “SANWA” (Sanwa Kagaku Kenkyusho Co., Ltd.) or Kalimate Oral Solution $20\%$ (Kowa Pharmaceuticals Co., Ltd.). ZSC only included Lokelma 5 g or 10 g powder for oral suspension (AstraZeneca, plc.). Precisely assessing the effect of potassium binders in patients with acute hyperkalemia may be difficult because insulin-glucose infusion or beta-2 adrenergic agonist inhalation is often involved. Therefore, we excluded patients who had been treated for <4 weeks to evaluate the effects of potassium binders in the post-acute phase. The other exclusion criteria were: estimated glomerular filtration rate (eGFR) > 60 mL/min/1.73 m2; patients on dialysis; and missing data on K+ level before or after administration. This was a single-center retrospective cohort study. All protocols were reviewed and approved by the Keio University School of Medicine Ethics Committee (approval no.: 20221131). Informed consent was obtained using the opt-out method available on the website. ## Clinical parameters We collected the following demographic and anthropometric data from medical records: age, sex, disposition status (inpatient or outpatient), comorbid conditions, and primary cause of CKD. Data of medications associated with serum potassium level (RAS inhibitors, mineralocorticoid receptor [MR] blockers, loop diuretics, thiazide diuretics, beta blockers, or sodium bicarbonate), body weight (BW) (kg), and blood pressure (mmHg) were also collected. The comorbidity score was assessed using the Charlson Comorbidity Index (CCI). Additionally, we collected biochemical data on potassium binder initiation and the first follow-up, 4 weeks after administration. These biochemical data included serum levels of albumin (g/dl), urea nitrogen (mg/dl), creatinine (mg/dl), sodium (mEq/l), chloride (mEq/l), potassium (mEq/l), calcium (mg/dl), and phosphorus (mg/dl). Additionally, eGFR (ml/min/1.73 m2) was calculated using the 3-variable Japanese equation: eGFR = 194 × serum Creatine−1.094 × Age−0.287 (×0.739 if woman) [19]. The corrected calcium level was obtained for the lower range of albumin using Payne's formula: corrected total calcium = total calcium + (4.0—albumin) [20]. Since data of serum bicarbonate level was unavailable, sodium minus chloride level was used as a surrogate marker for evaluating metabolic acidosis [21]. Daily drug cost for CPS and ZSC was calculated using the following values: 68.4 JPY/pack for ARGAMATE $20\%$ JELLY 25 g, 61.4 JPY/pack for CALCIUM POLYSTYRENE SULFONATE $20\%$ Oral Jelly “SANWA”, 66.1 JPY/pack for Kalimate Oral Solution $20\%$, 1,069.3 JPY/pack for Lokelma 5 g and 1,567.0 JPY/pack for Lokelma 10 g [22, 23]. To evaluate tolerability, we reviewed electronic medical records to determine whether the patients continued each potassium binder beyond the follow-up point. Additionally, if discontinued, we evaluated the reasons for cessation. ## Statistical analyses Continuous variables are expressed as medians (25th to 75th percentiles). Binary variables are expressed as percentages. Differences between the patients receiving ZSC and those administered CPS in normally and non-normally distributed continuous variables (assessed using the Kolmogorov-Smirnov test) and binary variables were evaluated using the unpaired Student's t-test, Mann–Whitney U test, and Fisher's exact test, respectively. Propensity score matching was employed to adjust for confounding variables and reduce treatment selection bias. We included the following clinically potential predictors of changes in K+ levels and selection of potassium binder type as independent variables: age, sex, BW, CCI, baseline eGFR, baseline K+ level, and use of RAS inhibitors. Propensity scores were calculated using a logistic regression model based on these variables. Subsequently, the ZSC and CPS groups were matched 1:1 on the logit of the propensity score using calipers with 0.2 standard deviation of the logit of the propensity score [24]. In a sensitivity analysis, a linear mixed effect model analysis was performed to evaluate the K+-lowering effect of ZSC. Fixed effects were time (baseline and follow-up) and time × potassium binder type besides variables that were selected for the above-mentioned propensity score, and random effect was participant number. Moreover, changes in sodium minus chloride levels, corrected calcium levels, phosphorus levels, urea nitrogen levels, creatinine levels, and eGFR were evaluated using a linear mixed effect model with each baseline value added to the fixed effects. Outcomes of these models are expressed as estimated marginal means ($95\%$ confidence interval). All statistical analyses were performed using SPSS version 27 (IBM Inc., Armonk, NY, USA) and EZR, which is a graphical user interface for R (The R Foundation for Statistical Computing) [25]. All P-values were two-sided, and those < 0.05 indicated statistically significance. ## Patients' characteristics Among a total of 572 patients who newly received potassium binders (ZSC or CPS) during the study period, 440 patients were excluded. Those excluded were patients using ZSC or CPS for <4 weeks ($$n = 353$$), on dialysis ($$n = 69$$), with eGFR > 60 mL/min/1.73 m2 ($$n = 12$$), and missing values of pre and/or post K+ levels ($$n = 6$$) (Figure 1). Consequently, 132 patients administered ZSC ($$n = 48$$) and CPS ($$n = 84$$) were included in the study: 38 patients were matched in both groups. Table 1 summarizes the baseline characteristics of the two groups, categorized according to the type of potassium binder used. Before propensity score matching, in the ZSC and CPS groups, the median age was 75 and 79 years and the female proportion was 18.8 and $33.3\%$, respectively. The follow-up period did not significantly differ between the two groups (6 vs. 5 weeks, $$P \leq 0.52$$). Regarding medication-related burden, the number of doses per day was significantly lower in the ZSC group than in the CPS group (1 vs. 2, $P \leq 0.01$), whereas the ZSC group had a significantly higher daily medication cost (1,069 vs. 132 JPY, $P \leq 0.01$). The ZSC group tended to be more likely to receive RAS inhibitors than did the CPS group ($$P \leq 0.07$$). However, no significant difference was observed in the frequency of RAS inhibitor use after matching between the two groups. Figure 1: Flow chart of the study enrollment. TABLE_PLACEHOLDER:Table 1 ## Efficacy in lowering K+ level The K+ levels are listed in Table 2. Compared with the CPS group, the ZSC group tended to have high baseline K+ level (5.9 vs. 5.7 mEq/l, $$P \leq 0.07$$). However, propensity score matching eliminated that significant difference (5.8 vs. 5.8 mEq/l, $$P \leq 0.76$$). After matching, the follow-up K+ level was lower in the ZSC group than that in the CPS group (4.5 vs. 5.0 mEq/l, $P \leq 0.05$). Moreover, the ZSC group showed a greater reduction in K+ levels (−1.2 vs. −0.8 mEq/l, $P \leq 0.05$) compared with the CPS group. Linear mixed effect model analysis revealed that K+ levels with estimated marginal mean in the ZSC group declined from 5.8 at baseline to 4.5 mEq/l at follow-up, whereas it declined from 5.7 to 4.9 mEq/l in the CPS group (Figure 2A; Supplementary Table S1), indicating a mean difference of −0.5 mEq/l between the two group ($P \leq 0.01$). ## Effects on other electrolyte levels and kidney function Table 2 summarizes the changes in serum levels of sodium minus chloride, corrected calcium, phosphorus, creatinine and urea nitrogen, and eGFR. After propensity score matching, no significant differences were observed in any of the baseline values between the two groups. Serum sodium minus chloride level, a surrogate marker for metabolic acidosis, was significantly elevated in the ZSC group compared with the CPS group (1.9 vs. 1.0 mEq/l, $P \leq 0.05$), indicating that metabolic acidosis could have been more potentially improved in the ZSC group. Additionally, the CPS group showed slightly higher elevation in serum corrected calcium level (0.1 vs. 0.0 mg/dl, $P \leq 0.05$), probably due to its calcium content. No significant differences were observed regarding changes in serum phosphorus levels and all markers of kidney function. Similar results were obtained in the linear mixed effects model analysis (Figures 2B–G; Supplementary Table S1). ## Medication retention rate and reasons for discontinuation Thirteen of 48 patients ($27.1\%$) in the ZSC group and 19 of 84 patients ($22.6\%$) in the CPS group discontinued each potassium binder after the follow-up point, with no significant differences ($$P \leq 0.67$$) (Table 3). The most common reason for ZSC cessation was unknown ($$n = 7$$), followed by hypokalemia ($$n = 3$$), unpleasant taste ($$n = 2$$), and other adverse events ($$n = 1$$). The reasons for CPS cessation were unknown ($$n = 7$$), constipation ($$n = 5$$), hypokalemia ($$n = 2$$), unpleasant taste ($$n = 2$$), ineffectiveness ($$n = 2$$), and other adverse events ($$n = 1$$). Unknown was the most common reason for discontinuation in both groups, and none of the patients in the ZSC group discontinued treatment due to constipation which was the second common reason in the CPS group. Table 3 \| Outcomes \| ZSC group (n = 48) \| CPS group (n = 84) \| P value \| \| --- \| --- \| --- \| --- \| \| Continuous administration \| Continuous administration \| Continuous administration \| Continuous administration \| \| Yes \| 35 (72.9) \| 65 (77.4) \| 0.67 \| \| No \| 13 (27.1) \| 19 (22.6) \| \| \| Reasons for discontinuation \| Reasons for discontinuation \| Reasons for discontinuation \| Reasons for discontinuation \| \| Constipation \| 0 (0.0) \| 5 (6.0) \| 0.16 \| \| Unpleasant taste \| 2 (4.2) \| 2 (2.4) \| 0.62 \| \| Hypokalemia \| 3 (6.3) \| 2 (2.4) \| 0.35 \| \| Ineffectiveness \| 0 (0.0) \| 2 (2.4) \| 0.53 \| \| Other adverse events \| 1 (2.1) \| 1 (1.2) \| 1.00 \| \| Unknown \| 7 (14.6) \| 7 (8.3) \| 0.38 \| ## Discussion Huda et al. have recently reported that ZSC and CPS were equally effective in reducing K+ level in patients with hyperkalemia who were admitted to the hospital in an emergency, with comparable cost differences between the two groups [26]. However, the differences between these two agents in potassium control and cost after an acute phase including outpatient management remain unclear. To the best of our knowledge, this is the first study to evaluate the difference in the post-acute phase K+-lowering effect between ZSC and CPS in patients with non-dialysis dependent CKD. Propensity score methods and linear mixed effect models demonstrated that ZSC controlled K+ levels more effectively than did CPS. In addition, a greater increase in sodium minus chloride level, which is only a surrogate indicator for metabolic acidosis, was observed in the ZSC group. These results indicated that ZSC could be a promising therapeutic agent for treatment and prevention of hyperkalemia in patients with CKD. Patients with CKD are susceptible to developing hyperkalemia due to reduced functioning nephrons, use of RAS inhibitors, concomitant heart failure, and diabetes mellitus (1, 7–9, 27). The prevalence of hyperkalemia increases as kidney function declines, with a rate of $14.6\%$ in patients with non-dialysis dependent CKD and $33.3\%$ in those with end-stage kidney disease [1]. The mechanism by which hyperkalemia is associated with high mortality could be attributed not only to life-threatening cardiac arrhythmias by hyperkalemia itself, but also to RAS inhibitor down-titration or discontinuation as a result of hyperkalemia [14, 15]. In patients with CKD, restriction of potassium intake is often recommended for managing K+ levels. However, several studies have demonstrated that potassium-rich diets including fresh vegetables and fruits were associated with delayed progression of CKD as well as cardiovascular health problems. This beneficial effect might be attributed to other mineral elements, vitamins, and dietary fibers of a high-potassium diet [10, 28, 29]. Although there is a lack of evidence for potassium binders to directly provide reno- and cardio-protective effects, these drugs are likely to be useful for maintaining appropriate RAS inhibitor therapy and avoiding excessive restriction of potassium intake. Particularly, RAS inhibitors are critical for preventing progression of diabetic kidney disease (DKD) [30, 31]. Moreover, a recent study has demonstrated that a combination of finerenone, a novel MR blocker, and RAS inhibitors could act synergistically to reduce the risk of DKD progression and cardiovascular events in patients with diabetes [32]. Given that diabetes is the underlying disease for type 4 renal tubular acidosis with hyperkalemia, RAS inhibitors and MR blockers plus potassium binders could become the standard therapy in patients with DKD in the future. Although K+-lowering effects of ZSC in the present study appeared to be generally equivalent or superior to those in clinical trials, CPS showed only weaker effects than expected, possibly for two reasons [18, 33]. First, the lower dose of CPS in the present study compared with the randomized control study (RCT) (15 g per day) may be partially responsible for the modest efficacy of CPS [33]. The dose of potassium binders in real-world practice has been reported to tend to be lower probably due to concerns about the gastrointestinal side effects of CPS including constipation, colonic necrosis, and intestinal perforation [12, 13, 34]. Second, the potential poor adherence to CPS might have also contributed to the reduction in its efficacy [6]. The unpleasant taste and recommended twice-daily or thrice-daily treatment regimens could have a negative impact on patient compliance [33, 35]. These clinical shortcomings of CPS could lead to the results of the present study, showing that ZSC was more effective in improving hyperkalemia. CKD is associated with the development of metabolic acidosis due to impaired NH4+ excretion or reduced tubular bicarbonate reabsorption [36]. The increase in sodium minus chloride difference was also significant in the ZSC group compared with the CPS group, indicating that ZSC could have the potential to improve metabolic acidosis. Although the mechanism for improved metabolic acidosis with ZSC is poorly understood, it may be due to the effects of direct binding and removal of NHNH4+ from the gastrointestinal tract and/or increased renal ammoniagenesis with correction of hyperkalemia [37]. Since metabolic acidosis is a risk factor for an accelerated decline in kidney function, cardiovascular events, and impaired physical function, patients with advanced CKD often receive sodium bicarbonate supplements [36]. Each 5 g dose of ZSC contains approximately 400 mg of sodium, and mild-to-moderate edema has been reported as an adverse event [38]. However, considering that ZSC treatment could improve metabolic acidosis and possibly reduce the requirement for sodium bicarbonate, the concern regarding sodium loading by ZSC might be counterbalanced. A large-scale retrospective study has reported a low continuation rate of conventional potassium binders: approximately $60\%$ of patients with hyperkalemia who had received potassium binders discontinued them for some reason within 1 year [6]. It has also been reported that only $10\%$ of new sodium polystyrene sulfonate users continued therapy for more than 60 days [39]. Regarding this, we hypothesized that the potentially improved tolerability of ZSC could have facilitated its continuous use [17]. However, the medication retention rate beyond the follow-up period was comparable between the ZSC and CPS groups. Although the details were unclear because more than half of the reasons for ZSC discontinuation were not described in the medical record, the high medication cost of ZSC, about eight times that of CPS, might have contributed to its cessation. Kim et al. reported that the cost of prescribing ZSC could be almost fully compensated by savings accompanying the reduced risk of hyperkalemia events requiring hospitalization [40]. However, considering that ZSC is less prevalent than anticipated, it is desirable that its drug price be more accessible to many patients. The present study has several limitations. First, since this was not an RCT but a retrospective observational study, our results should be interpreted with caution. The type of potassium binder was not blinded, which might have led to information bias. Although propensity score matching model and linear mixed effect model analyses were performed to adjust for covariates clinically associated with response to potassium binder and selection of its type, the impact of potential confounders could not be completely ruled out. Second, the nature of this single-center study may have caused selection bias, limiting generalizability. Third, metabolic acidosis was only indirectly assessed using a surrogate marker of serum sodium minus chloride level. Therefore, the results should be interpreted with caution. Additionally, the short follow-up period and small sample size did not allow for adequate evaluation of the continuation rate of ZSC and its adverse events. The above issues should be addressed by robust evidence from future multicenter prospective studies. However, we suggest that the real-world findings of this observational study might be useful for managing hyperkalemia, in which medication and diet adherence play an important role. In summary, our study demonstrated that ZSC could more potentially improve hyperkalemia and possibly metabolic acidosis in patients with CKD than CPS. Future RCTs are encouraged to clarify the effectiveness of ZSC. ## 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 Keio University School of Medicine Ethics Committee (approval no.: 20221131). Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements. ## Author contributions TN and SY contributed to design of the study, collection of the data, and preparation of the initial draft. KH, KU, TT, TA, KM, TY, JY, and TK contributed to interpretation of the data and revision of the manuscript. TM and HI supervised the manuscript. All authors 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. ## Supplementary material The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmed.2023.1137981/full#supplementary-material ## References 1. Humphrey T, Davids MR, Chothia MY, Pecoits-Filho R, Pollock C, James G. How common is hyperkalaemia? A systematic review and meta-analysis of the prevalence and incidence of hyperkalaemia reported in observational studies. Clin Kidney J. (2021.0) 15 727-37. DOI: 10.1093/ckj/sfab243 2. Strippoli GF, Bonifati C, Craig M, Navaneethan SD. Craig JC: Angiotensin converting enzyme inhibitors and angiotensin II receptor antagonists for preventing the progression of diabetic kidney disease. Cochrane Database Syst Rev. (2006.0) 2006 CD006257. DOI: 10.1002/14651858.CD006257 3. Mishima E, Haruna Y, Arima H. Renin-angiotensin system inhibitors in hypertensive adults with non-diabetic CKD with or without proteinuria: a systematic review and meta-analysis of randomized trials. Hypertens Res. (2019.0) 42 469-82. DOI: 10.1038/s41440-018-0116-3 4. Murphy SP, Ibrahim NE, Januzzi JL Jr. Heart failure with reduced ejection fraction: a review. JAMA. (2020.0) 324 488-504. DOI: 10.1001/jama.2020.10262 5. An JN, Lee JP, Jeon HJ, Kim DH, Oh YK, Kim YS. Severe hyperkalemia requiring hospitalization: predictors of mortality. Crit Care. (2012.0) 16 R225. DOI: 10.1186/cc11872 6. Kashihara N, Kohsaka S, Kanda E, Okami S, Yajima T. Hyperkalemia in real-world patients under continuous medical care in Japan. Kidney Int Rep. (2019.0) 4 1248-60. DOI: 10.1016/j.ekir.2019.05.018 7. Collins AJ, Pitt B, Reaven N, Funk S, McGaughey K, Wilson D. Association of serum potassium with all-cause mortality in patients with and without heart failure, chronic kidney disease, and/or diabetes. Am J Nephrol. (2017.0) 46 213-21. DOI: 10.1159/000479802 8. Kanda E, Kashihara N, Kohsaka S, Okami S, Yajima T. Clinical and economic burden of hyperkalemia: a nationwide hospital-based cohort study in Japan. Kidney Med. (2020.0) 2 742-52. DOI: 10.1016/j.xkme.2020.09.003 9. Thomsen RW, Nicolaisen SK, Hasvold P, Sanchez RG, Pedersen L, Adelborg K. Elevated potassium levels in patients with chronic kidney disease: occurrence, risk factors and clinical outcomes—a Danish population-based cohort study. Nephrol Dial Transplant. (2018.0) 33 1610-20. DOI: 10.1093/ndt/gfx312 10. KDIGO 2012 clinical practice guideline for the evaluation and management of chronic kidney disease. Kidney Int Suppl. (2013.0) 3 S1-S150 11. Betz M, Steenes A, Peterson L, Saunders M. Knowledge does not correspond to adherence of renal diet restrictions in patients with chronic kidney disease stage 3-5. J Ren Nutr. (2021.0) 31 351-60. DOI: 10.1053/j.jrn.2020.08.007 12. Sterns RH, Rojas M, Bernstein P, Chennupati S. Ion-exchange resins for the treatment of hyperkalemia: are they safe and effective?. J Am Soc Nephrol. (2010.0) 21 733-5. DOI: 10.1681/ASN.2010010079 13. Buraphat P, Niyomnaitham S, Pongpaibul A, Maneerattanaporn M. Calcium polystyrene sulfonate-induced gastrointestinal tract necrosis and perforation. Acta Gastroenterol Belg. (2019.0) 82 542-3. PMID: 31950813 14. Xu Y, Fu EL, Trevisan M, Jernberg T, Sjölander A, Clase CM. Stopping renin-angiotensin system inhibitors after hyperkalemia and risk of adverse outcomes. Am Heart J. (2022.0) 243 177-86. DOI: 10.1016/j.ahj.2021.09.014 15. Ren H, Leon SJ, Whitlock R, Rigatto C, Komenda P, Bohm C. Prescription patterns of sodium and calcium polystyrene sulfonate in patients with hyperkalemia and chronic kidney disease receiving RAAS inhibitors. Clin Kidney J. (2022.0) 15 1713-9. DOI: 10.1093/ckj/sfac077 16. Rosano GMC, Spoletini I, Agewall S. Pharmacology of new treatments for hyperkalaemia: patiromer and sodium zirconium cyclosilicate. Eur Heart J Suppl. (2019.0) 21 A28-33. DOI: 10.1093/eurheartj/suy035 17. Stavros F, Yang A, Leon A, Nuttall M, Rasmussen HS. Characterization of structure and function of ZS-9, a K. PLoS ONE. (2014.0) 9 e114686. DOI: 10.1371/journal.pone.0114686 18. Zannad F, Hsu BG, Maeda Y, Shin SK, Vishneva EM, Rensfeldt M. Efficacy and safety of sodium zirconium cyclosilicate for hyperkalaemia: the randomized, placebo-controlled HARMONIZE-Global study. ESC Heart Fail. (2020.0) 7 54-64. DOI: 10.1002/ehf2.12561 19. Matsuo S, Imai E, Horio M, Yasuda Y, Tomita K, Nitta K. Collaborators developing the Japanese equation for estimated GFR. Revised equations for estimated GFR from serum creatinine in Japan. Am J Kidney Dis. (2009.0) 53 982-92. DOI: 10.1053/j.ajkd.2008.12.034 20. Payne RB, Little AJ, Williams RB, Milner JR. Interpretation of serum calcium in patients with abnormal serum proteins. Br Med J. (1973.0) 4 643-6. DOI: 10.1136/bmj.4.5893.643 21. Havlin J, Matousovic K, Schück O. Sodium-chloride difference as a simple parameter for acid-base status assessment. Am J Kidney Dis. (2017.0) 69 707-8. DOI: 10.1053/j.ajkd.2016.12.019 22. 22.Pmda. Pharmaceuticals and Medical Devices Agency. Available online at: https://www.pmda.go.jp/ (accessed October 27, 2022).. Pharmaceuticals and Medical Devices Agency 23. 23.ARGAMATE 20% JELLY 25 g (calcium polystyrene sulfonate). Available online at: https://med.skk-net.com/supplies/generic/products/item/ARG_if_1601.pdf (accessed October 27, 2022). 24. Wang Y, Cai H, Li C, Jiang Z, Wang L, Song J. Optimal caliper width for propensity score matching of three treatment groups: a Monte Carlo study. PLoS ONE. (2013.0) 8 e81045. DOI: 10.1371/journal.pone.0081045 25. Kanda Y. Investigation of the freely available easy-to-use software 'EZR' for medical statistics. Bone Marrow Transplant. (2013.0) 48 452-8. DOI: 10.1038/bmt.2012.244 26. Huda AB, Langford C, Lake J, Langford N. Hyperkalaemia and potassium binders: retrospective observational analysis looking at the efficacy and cost effectiveness of calcium polystyrene sulfonate and sodium zirconium cyclosilicate. J Clin Pharm Ther. (2022.0) 47 2170-5. DOI: 10.1111/jcpt.13766 27. Gennari FJ, Segal AS. Hyperkalemia: an adaptive response in chronic renal insufficiency. Kidney Int. (2002.0) 62 1-9. DOI: 10.1046/j.1523-1755.2002.00350.x 28. Aaron, KJ. Sanders PW: Role of dietary salt and potassium intake in cardiovascular health and disease: a review of the evidence. Mayo Clin Proc. (2013.0) 88 987-95. DOI: 10.1016/j.mayocp.2013.06.005 29. Clegg DJ, Headley SA, Germain MJ. Impact of dietary potassium restrictions in ckd on clinical outcomes: benefits of a plant-based diet. Kidney Med. (2020.0) 2 476-87. DOI: 10.1016/j.xkme.2020.04.007 30. Brenner BM, Cooper ME, de Zeeuw D, Keane WF, Mitch WE, Parving HH. Effects of losartan on renal and cardiovascular outcomes in patients with type 2 diabetes and nephropathy. N Engl J Med. (2001.0) 345 861-9. DOI: 10.1056/NEJMoa011161 31. Lewis EJ, Hunsicker LG, Clarke WR, Berl T, Pohl MA, Lewis JB. Collaborative Study Group. Renoprotective effect of the angiotensin-receptor antagonist irbesartan in patients with nephropathy due to type 2 diabetes. N Engl J Med. (2001.0) 345 851-60. DOI: 10.1056/NEJMoa011303 32. Pitt B, Filippatos G, Agarwal R, Anker SD, Bakris GL, Rossing P. Cardiovascular events with finerenone in kidney disease and Type 2 diabetes. N Engl J Med. (2021.0) 385 2252-63. DOI: 10.1056/NEJMoa2110956 33. Nakayama Y, Ueda K, Yamagishi SI, Sugiyama M, Yoshida C, Kurokawa Y. Compared effects of calcium and sodium polystyrene sulfonate on mineral and bone metabolism and volume overload in pre-dialysis patients with hyperkalemia. Clin Exp Nephrol. (2018.0) 22 35-44. DOI: 10.1007/s10157-017-1412-y 34. Yu MY, Yeo JH, Park JS, Lee CH, Kim GH. Long-term efficacy of oral calcium polystyrene sulfonate for hyperkalemia in CKD patients. PLoS ONE. (2017.0) 12 e0173542. DOI: 10.1371/journal.pone.0173542 35. Schröder CH, van den Berg AM, Willems JL, Monnens LA. Reduction of potassium in drinks by pre-treatment with calcium polystyrene sulphonate. Eur J Pediatr. (1993.0) 152 263-4. DOI: 10.1007/BF01956159 36. Adamczak M, Surma S. Metabolic acidosis in patients with CKD: epidemiology, pathogenesis, and treatment. Kidney Dis (Basel). (2021.0) 7 452-67. DOI: 10.1159/000516371 37. Ash SR, Batlle D, Kendrick J, Oluwatosin Y, Pottorf W, Brahmbhatt Y. Effect of sodium zirconium cyclosilicate on serum potassium and bicarbonate in patients with hyperkalemia and metabolic acidosis associated with chronic kidney disease: rationale and design of the NEUTRALIZE study. Nephron. (2022.0) 146 599-609. DOI: 10.1159/000523911 38. Roger SD, Spinowitz BS, Lerma EV, Singh B, Packham DK, Al-Shurbaji A. Efficacy and safety of sodium zirconium cyclosilicate for treatment of hyperkalemia: an 11-month open-label extension of HARMONIZE. Am J Nephrol. (2019.0) 50 473-80. DOI: 10.1159/000504078 39. Palmer BF. Potassium binders for hyperkalemia in chronic kidney disease-diet, renin-angiotensin-aldosterone system inhibitor therapy, and hemodialysis. Mayo Clin Proc. (2020.0) 95 339-54. DOI: 10.1016/j.mayocp.2019.05.019 40. Kim K, Fagerström J, Chen G, Lagunova Z, Furuland H. BMC Nephrol. (2022.0) 23 281. DOI: 10.1186/s12882-022-02903-7
--- title: Macromolecular crowding in animal component-free, xeno-free and foetal bovine serum media for human bone marrow mesenchymal stromal cell expansion and differentiation authors: - Stefanie H. Korntner - Alessia Di Nubila - Diana Gaspar - Dimitrios I. Zeugolis journal: Frontiers in Bioengineering and Biotechnology year: 2023 pmcid: PMC10025396 doi: 10.3389/fbioe.2023.1136827 license: CC BY 4.0 --- # Macromolecular crowding in animal component-free, xeno-free and foetal bovine serum media for human bone marrow mesenchymal stromal cell expansion and differentiation ## Abstract Background: Cell culture media containing undefined animal-derived components and prolonged in vitro culture periods in the absence of native extracellular matrix result in phenotypic drift of human bone marrow stromal cells (hBMSCs). Methods: Herein, we assessed whether animal component-free (ACF) or xeno-free (XF) media formulations maintain hBMSC phenotypic characteristics more effectively than foetal bovine serum (FBS)-based media. In addition, we assessed whether tissue-specific extracellular matrix, induced via macromolecular crowding (MMC) during expansion and/or differentiation, can more tightly control hBMSC fate. Results: Cells expanded in animal component-free media showed overall the highest phenotype maintenance, as judged by cluster of differentiation expression analysis. Contrary to FBS media, ACF and XF media increased cellularity over time in culture, as measured by total DNA concentration. While MMC with Ficoll™ increased collagen deposition of cells in FBS media, FBS media induced significantly lower collagen synthesis and/or deposition than the ACF and XF media. Cells expanded in FBS media showed higher adipogenic differentiation than ACF and XF media, which was augmented by MMC with Ficoll™ during expansion. Similarly, Ficoll™ crowding also increased chondrogenic differentiation. Of note, donor-to-donor variability was observed for collagen type I deposition and trilineage differentiation capacity of hBMSCs. Conclusion: Collectively, our data indicate that appropriate screening of donors, media and supplements, in this case MMC agent, should be conducted for the development of clinically relevant hBMSC medicines. ## 1 Introduction Mesenchymal stromal cells (MSCs) hold great potential for therapeutic and reparative use in tissue engineering and regenerative medicine due to their self-renewal, multipotency and immunomodulatory properties (Deans and Moseley, 2000; Di Nicola et al., 2002; Jorgensen et al., 2003). Regarding clinical translation of MSC medicines, animal-derived cell culture media components (i.e., animal sera) raise safety concerns related to xenogeneic contaminations and disease transfer through pathogens (e.g., mycoplasma, viruses and prions) (Selvaggi et al., 1997; Tekkatte et al., 2011; Hawkes, 2015; van der Valk, 2022). Also, antibodies against bovine antigens (when foetal bovine serum, FBS, is used, which is the most widely used serum in cell culture) may be elicited by repeated administration of cells, which will in turn directly affect the safety and efficacy of cell-based treatments to patients (Horwitz et al., 2002; Sundin et al., 2007). Moreover, the undefined composition of FBS results in inconsistent batch-to-batch performance, low reproducibility of experiments and ultimately jeopardises the therapeutic potential of MSC therapies (Honn et al., 1975; Heiskanen et al., 2007). As a tight control of cell behaviour in vitro is imperative for intended clinical use, recent efforts have been directed towards the development of more defined xeno-free (XF) and/or animal component-free (ACF) media formulations for translational research, development and regulatory compliant MSC medicines (Chase et al., 2012; Jung et al., 2012; Kinzebach and Bieback, 2013). Per definition, both XF and ACF media cannot contain animal-derived proteins or serum. While ACF media is entirely free of animal- and human-derived components and all elements are therefore chemically defined, XF media can contain human-derived supplements (de Soure et al., 2016; Karnieli et al., 2017). Another limiting factor in the clinical translation of MSC therapies, especially in the case of autologous therapies, is the prolonged in vitro expansion required to reach the high cell numbers needed for therapeutic effects, which is associated with phenotype, immunomodulatory capability and therapeutic losses (Bara et al., 2014; Elgaz et al., 2019; L'Heureux et al., 2006; Siddappa et al., 2007; Whitfield et al., 2013; Yao et al., 2006; Zhang et al., 2015). In artificial in vitro cell culture systems, cells are grown in liquid media in planar 2D cultures, which poorly resemble the native in vivo scenario where cells reside in a dense 3D microenvironment, in direct contact with the extracellular matrix (ECM). In vivo, the dynamic reciprocity between cells and their surrounding ECM determines their fate and function (Roskelley and Bissell, 1995; Thorne et al., 2015; van Helvert et al., 2018). Similarly, the presence of tissue-specific ECM has been shown to facilitate cell phenotype maintenance in vitro (Pei et al., 2011; Cheng et al., 2014; Gattazzo et al., 2014; Yang et al., 2018a). In eukaryotic cell culture systems, macromolecular crowding (MMC), following the principles of excluded volume effect, enhances and accelerates tissue-specific ECM deposition (Raghunath and Zeugolis, 2021; Tsiapalis and Zeugolis, 2021; Zeugolis, 2021), a phenomenon that has been well documented in both differentiated (Lareu et al., 2007; Satyam et al., 2014; Kumar et al., 2015a; Kumar et al., 2015b; Satyam et al., 2016; Kumar et al., 2018; Gaspar et al., 2019; Shendi et al., 2019; Tsiapalis et al., 2021) and progenitor cell cultures (Zeiger et al., 2012; Prewitz et al., 2015; Cigognini et al., 2016; Lee et al., 2016; Patrikoski et al., 2017; Graceffa and Zeugolis, 2019; De Pieri et al., 2020). However, to-date, only one study has assessed the influence of MMC in xeno-free and/or serum-free media formulations using human adipose-derived mesenchymal stromal cells (Patrikoski et al., 2017). Considering the above, herein we ventured to investigate the influence of MMC in FBS, XF and ACF media on human bone marrow mesenchymal stromal cell (hBMSC) expansion and differentiation. Cells from two donors were isolated in ACF media and expanded from passage 0 (p0) to passage 4 (p4) in FBS, XF and ACF media, in the absence and presence of MMC. At p4, phenotype, viability, metabolic activity, proliferation, collagen deposition and trilineage differentiation analyses were assessed (Figure 1 graphically illustrates the study design). FIGURE 1: Study design. Created with BioRender.com. ## 2.1 Materials Ficoll™ (Fc) 70 kDa and 400 kDa were purchased from Sigma Aldrich (Ireland). Polysucrose 1,000 kDa (Fc 1,000 kDa) was purchased from TdB Consultancy AB (Sweden). MesenCult™ ACF Plus Media and supplements were purchased from STEMCELL Technologies (United Kingdom). MSC NutriStem® XF Medium (Biological Industries) was purchased from Geneflow Ltd. (United Kingdom). Tissue culture consumables were purchased from Sarstedt (Ireland) and NUNC (Denmark). All other chemicals, cell culture media and reagents were purchased from Sigma Aldrich (Ireland), unless otherwise stated. ## 2.2 Isolation and expansion of hBMSCs Fresh human whole bone marrow from the iliac crest of two different donors (donor 1: female, 22 years old; donor 2: male, 25 years old) was purchased from AllCells® (United States) and hBMSCs were isolated using the MesenCult™-ACF Plus medium, according to the manufacturer’s protocol. In order to provide a complete ACF culture system, the MesenCult™-ACF Plus medium was used in conjunction with ACF Cell Attachment Substrate (STEMCELL Technologies, United Kingdom) and ACF Cell Dissociation Kit (STEMCELL Technologies, United Kingdom). hBMSCs were isolated using density gradient medium separation (Lymphoprep™, STEMCELL Technologies, United Kingdom). Briefly, phosphate buffered saline (PBS) containing 2 mM ethylenediaminetetraacetic acid (EDTA) and Lymphoprep™ were added to the bone marrow sample. After centrifugation at 300 g for 30 min, the mononuclear cell layer was collected at the plasma/Lymphoprep™ interface and washed with cold PBS containing 2 mM EDTA. After another centrifugation step at 300 x g for 10 min, the supernatant was discarded, and the cell pellet resuspended in complete MesenCult™-ACF Plus medium. Nucleated cells were counted using $3\%$ acetic acid with methylene blue (STEMCELL Technologies, United Kingdom) and seeded into pre-coated (ACF cell attachment substrate) culture flasks with complete MesenCult™-ACF Plus medium at a density of 50,000 freshly isolated cells/cm2. Flasks were incubated at 37°C until cells reached a confluency of approximately $80\%$. A half-medium change was performed on day 7. Cells up to this stage were considered to be at p0. From p1 onwards, hBMSCs were subjected to MMC treatment. For MMC conditions, a Fc cocktail, composed of 10 mg/mL Fc 70 kDa, 25 mg/mL Fc 400 kDa and 2.25 mg/mL Fc 1,000 kDa [Fc cocktail was previously optimised for maximum excluded volume effect (Gaspar et al., 2019)] was used, dissolved in the respective media. Cells were expanded in MesenCult™-ACF Plus Medium, STEMCELL Technologies, United Kingdom (referred to from now on as ACF) or MSC NutriStem® XF Medium, Biological Industries, United Kingdom (referred to from now on as XF) without and with MMC, according to the manufacturer’s protocols. $1\%$ penicillin/streptomycin (P/S) was added to both ACF and XF media formulations. For a serum-containing control, cells were expanded in alpha-Minimum Essential Medium (α-MEM) with GlutaMAX (Gibco Life Technologies, Ireland) supplemented with $10\%$ FBS, Life Technologies, Ireland (referred to from now on as FBS), $1\%$ (P/S), 1 ng/mL of basic fibroblast growth factor/fibroblast growth factor 2 and without and with MMC. FBS, XF and ACF media were supplemented with 100 μM L-ascorbic acid 2-phosphate sesquimagnesium salt hydrate, to induce collagen synthesis. Cells in each media (Supplementary Table S1) were expanded at 37°C in a humidified atmosphere of $5\%$ CO2 until p4. hBMSCs expanded with the ACF medium, XF medium and FBS medium were detached using the ACF Cell Dissociation Kit (STEMCELL Technologies, United Kingdom), the TrypLE Select (Life Technologies) and trypsin-EDTA (Life Technologies), respectively. ## 2.3 Flow cytometry analysis Flow cytometry (BD FACSCanto™ II, BD Biosciences, Belgium and BD Stemflow™, United Kingdom) was used to determine the immunophenotype of hBMSCs at p0 and to assess the effect of different expansion media and MMC on immunophenotype of BMSCs at p4 after 10 days of culture. The monoclonal antibodies against cluster of differentiation (CD) CD105, CD73, CD90, CD44, CD45, CD31 and CD146 and their respective isotype controls are provided in Supplementary Table S2. Briefly, cells were detached using respective detachment solutions, centrifuged and resuspended in $2\%$ FBS in PBS. After straining using a 40 μm cell strainer, cells were counted and diluted to a concentration of 1,000,000 cells/mL in $2\%$ FBS in PBS. Subsequently, ∼100,000 cells were placed in each tube and stained with the appropriate volume of fluorochrome-labelled antibodies for 30 min at 4°C. Cells were washed with PBS and resuspended in $2\%$ FBS in PBS. Analysis was performed on 100,000 cells per sample and unstained cell samples were used to correct for background autofluorescence. SYTOX™ Blue Dead Cell Stain (Invitrogen, United Kingdom) was used to label and exclude dead cells. Single stained samples were used to determine the level of spectral overlap between different fluorophores and for compensation. Fluorescence minus one (FMO) controls were used to determine gating boundaries. Isotype control antibodies were used to assess the level of background staining and non-specific binding. Cells were analysed using a BD FACSCanto™ II cytometer (BD Biosciences, United Kingdom) and Median Fluorescence Intensity of hBMSCs was calculated using FlowJo® software v10 (TreeStar Inc., United States). The gating strategy was as follows: a primary gate was placed on the area vs height signal of the forward scatter (FSC-A/FSC-H) dot plot to discriminate for doublets and cell aggregates. The single cell population was identified by defining the gated population on a side scatter area signal vs a forward scatter area (SSC-A/FSC-A) signal dot plot. Single parameter histograms were generated, overlayed with respective isotype controls, and range gates were used to determine the percentage of cells expressing the individual surface markers. ## 2.4 Phase contrast microscopy analysis To assess morphological changes of hBMSCs during cell expansion and during trilineage differentiation, cells were observed using an inverted brightfield microscope (Leica Microsystem, Germany). Phase contrast images were captured at different passages and during trilineage differentiation and were processed using ImageJ software (NIH, United States). ## 2.5 Cell viability analysis At p4 cells were seeded at a density 25,000 cells/cm2, and a Live/Dead assay, with calcein AM (ThermoFisher Scientific, United Kingdom) and ethidium homodimer I (ThermoFisher Scientific, United Kingdom) stainings, was performed at day 4 and day 10 of culture, as per manufacturer’s protocol. In live cells, non-fluorescent calcein-AM is converted to green fluorescent calcein after acetoxymethyl ester hydrolysis by intracellular esterases. Ethidium homodimer I can penetrate the disrupted cell membranes of dead or dying cells and binds to DNA, producing a red fluorescence. Briefly, at each time point, cells were washed with Hank’s Balanced Salt Solution (HBSS) and a solution of calcein AM (4 μM) and ethidium homodimer I (2 μM) was added. After 30 min incubation at 37°C in a humidified atmosphere of $5\%$ CO2, fluorescence images were obtained with an Olympus IX-81 inverted fluorescence microscope (Olympus Corporation, Japan). For each condition dimethyl sulfoxide treated cells were used as negative control. ## 2.6 DNA concentration analysis At p4, cells were seeded at a density of 25,000 cells/cm2, and a Quant-iT™ PicoGreen® dsDNA (ThermoFisher Scientific, United Kingdom) assay was performed to quantify the amount of dsDNA present in the respective samples at day 4 and day 10 of culture, as per manufacturer’s protocol. Briefly, 250 μl of nucleic acid-free water was added to each well (24-well plate), the well plate was frozen at −80°C and three freeze-thaw cycles were performed to lyse the cells and extract the DNA. 100 μl of each DNA sample were transferred into a 96-well plate. A standard curve was generated with 0, 200, 375, 500, 1,000 and 2,000 ng/mL DNA concentrations. 100 μl of PicoGreen® reagent at 1:200 dilution in 1X Tris-EDTA buffer was added to all standards and samples. Fluorescence values (excitation: 480 nm, emission: 520 nm) were obtained with a Varioskan Flash Spectral scanning multimode reader (ThermoFisher Scientific, United Kingdom). The DNA concentration was defined as a function of the standard curve and compared at day 4 and day 10. ## 2.7 Cell metabolic activity analysis At p4, cells were seeded at a density of 25,000 cells/cm2 and the alamarBlue® (Invitrogen, United Kingdom) assay was carried out at day 4 and day 10 of culture, according to the manufacturer’s protocol. Briefly, samples were washed with HBSS and left to incubate in HBSS containing $10\%$ alamarBlue® for 3 h at 37°C in a humidified atmosphere of $5\%$ CO2. After incubation, 100 μl of the alamarBlue® solution were transferred into a 96-well plate. Absorbance readings were measured at 550 nm excitation and 595 nm emission with a Varioskan Flash Spectral scanning multimode reader (ThermoFisher Scientific, United Kingdom). Metabolic activity was expressed in terms of % of reduced alamarBlue™ dye normalised to the DNA quantity (ng/mL) obtained from the Quant-iT™ PicoGreen® dsDNA assay. Each value was normalised to the value of α-MEM -MMC group. ## 2.8 Electrophoresis analysis To assess collagen type I deposition sodium dodecyl sulphate-polyacrylamide gel electrophoresis (SDS-PAGE) was performed at p4, at a seeding density of 25,000 cells/cm2, at day 4 and at day 10 of culture, as has been described previously (Capella-Monsonis et al., 2018). Briefly, at each time point media were aspired and cell layers were washed with HBSS. Subsequently, cell layers were digested with porcine gastric mucosa pepsin at a final concentration of 0.1 mg/mL in 0.05 M acetic acid (Fischer Scientific, Ireland) and incubated for 2 h at 37°C with gentle shaking. After digestion, cell layers were scraped and neutralised with 0.1 N sodium hydroxide. For electrophoresis, sample buffer (SDS, 1.25 M Tris HCl, glycerol, bromophenol blue) was added to the samples. Cell layers were analysed by SDS-PAGE under non-reducing conditions with a Mini-Protean® three electrophoresis system (Bio-Rad Laboratories, United Kingdom). Bovine collagen type I (125 μg/mL, Symatese Biomateriaux, France) was used as control on every gel. Protein bands were stained with the SilverQuest™ kit (Invitrogen) according to the manufacturer’s protocol. The gels were imaged with a HP PrecisionScan Pro scanner (HP, United Kingdom). Densitometric analysis of the α1 and α2 bands was performed with ImageJ software (NIH). ## 2.9 Immunocytochemistry analysis At p4, cells were seeded in 48-well plates (Sarstedt, Ireland) at a density of 25,000 cells/cm2. At each time point cells were washed with PBS and fixed in $4\%$ paraformaldehyde for 15 min at room temperature. Cells were washed with PBS and then non-specific sites were blocked with $3\%$ bovine serum albumin for 30 min. Afterwards, cells were incubated overnight at 4°C with primary antibodies against collagen type I (Supplementary Table S2). Cells washed 3 times with PBS and incubated for 45 min at room temperature with the secondary antibody (Supplementary Table S2). Nuclei were counterstained with Hoechst 33342 Fluorescent Stain (ThermoFisher Scientific, United Kingdom). Images were acquired using an Olympus IX-81 inverted fluorescence microscope (Olympus Corporation, Japan) and relative fluorescence intensity was analysed with ImageJ software (NIH, United States). ## 2.10 Trilineage differentiation analysis For all differentiation experiments (Supplementary Table S3 provides the groups), cells at p4 were subjected to MesenCult™ Adipogenic Differentiation Kit (STEMCELL Technologies, United Kingdom), MesenCult™ Osteogenic Differentiation Kit (STEMCELL Technologies, United Kingdom) and MesenCult™-ACF Chondrogenic Differentiation Kit (STEMCELL Technologies, United Kingdom), according to the manufacturer’s protocols. For adipogenic and osteogenic differentiation cells were seeded in 48-well plates at an initial density of 25,000 cells/cm2, differentiation was commenced when cells were approximately $90\%$–$98\%$ confluent, and media was changed every 3 days. For chondrogenic differentiation a 3D pellet culture system was used with 500,000 cells/pellet. For MMC conditions, the same as during stem cell expansion Fc cocktail was used. ## 2.10.1 Adipogenic differentiation, oil red O staining and quantification of uptake After 14 days of adipogenic differentiation, phase contrast images were captured using an inverted brightfield microscope (Leica Microsystem, Germany). Cells were then fixed for 20 min with $4\%$ paraformaldehyde, stained for 15 min with oil red O solution (oil red O $0.5\%$ in isopropanol, diluted 3:2 in deionised water) at room temperature and images were acquired using an inverted microscope (Leica Microsystems, Germany). For semi-quantitative analysis of oil red O staining, the dye was extracted with $100\%$ isopropanol, the solution was centrifuged at 500 g for 2 min, and absorbance was measured at 520 nm using a Varioskan Flash plate reader (ThermoFisher Scientific). ## 2.10.2 Osteogenic differentiation, alizarin red staining and quantification of uptake After 14 days of osteogenic differentiation, phase contrast images were captured using an inverted brightfield microscope (Leica Microsystem, Germany). Cells were then fixed with ice-cold methanol for 20 min, stained with $2\%$ alizarin red solution in deionised water for 15 min and washed three times with deionised water. Brightfield images were acquired using an inverted microscope (Leica Microsystems, Germany). Semi-quantitative analysis of alizarin red staining was performed by dissolving the bound stain with $10\%$ acetic acid. Samples were collected using a cell scraper and heated to 85°C for 10 min. Subsequently, $10\%$ solution of ammonium hydroxide was used to adjust the pH to 4.5, and absorbance at 405 nm was read using a micro-plate reader (Varioskan Flash, ThermoFisher Scientific, Ireland). ## 2.10.3 Chondrogenic differentiation and Alcian Blue staining For pellet culture, cells were directly resuspended in chondrogenic differentiation medium, 0.5 mL of the cell suspension was added to each 15 mL polypropylene tube and centrifuged at 300 g for 10 min. Cells were incubated at 37°C in a humidified atmosphere of $5\%$ CO₂. On day 3, 0.5 mL chondrogenic media was added to reach a final volume of 1 mL and subsequently media was changed every 3 days. After 21 days of differentiation, pellets were fixed with $4\%$ paraformaldehyde, cryoprotected with $15\%$ and $30\%$ solutions of sucrose in one x PBS (w/vol), cryo-embedded and cryo-sectioned (5 μm) with a Leica Cryostat (Leica Biosystems, Germany). To assess the presence of proteoglycans sections were stained with Alcian Blue 8GX solution (Sigma-Aldrich 66011) for 30 min at room temperature and counterstained with Nuclear fast red (Nuclear fast red–aluminium sulphate solution $0.1\%$, Merck Millipore, 1001210500) for 1 min at room temperature. Slides were dehydrated in $100\%$ ethanol, xylene and mounted. Brightfield images were acquired using an inverted microscope (Leica Microsystems, Germany). ## 2.11 Statistical analysis For both donors ($$n = 2$$ biological replicates), all experiments were conducted in three technical replicates ($$n = 3$$). Due to limitations in cell numbers, flow cytometry assays were performed one time ($$n = 1$$ technical replicate) for each donor ($$n = 2$$). Data were processed using MINITAB® version 17 (Minitab Inc., United States) and reported as mean ± standard deviation. One-way analysis of variance (ANOVA) was used for multiple comparisons and Tukey’s post hoc test was used for pairwise comparisons when the group distributions were normal (Anderson-Darling normality test) and the variances of populations were equal (Bonett’s test and Levene’s test). When either or both assumptions were violated, non-parametric analysis was conducted using Kruskal–Wallis test for multiple comparisons and Mann-Whitney test for pairwise comparisons. Results were considered statistically significant for $p \leq 0.05.$ ## 3.1 Cell immunophenotype Flow cytometry analysis at p0 (Supplementary Figures S1-S2; Supplementary Tables S4-S5) revealed that hBMSCs of both donors that were isolated in ACF, were positive for CD90, CD73, CD44, CD105 and CD146 and negative for the haematopoietic markers CD31 and CD45. However, cells of donor two exhibited an elevated population of CD45+ cells at p0. Flow cytometry analysis at p4 (Supplementary Figures S3-S4; Supplementary Tables S4-S5) revealed that hBMSCs of both donors continued to express CD90, CD44, CD73 and CD105 at high levels in most culture conditions. Reduced levels of CD105 were detected for donor one in FBS -MMC, XF -MMC and ACF + MMC, and for donor two in FBS -MMC, FBS + MMC, ACF -MMC and ACF + MMC. The addition of MMC retained high CD105 values in both FBS and XF groups. In contrast, ACF + MMC showed lower CD105 values compared to ACF-MMC. While cells of both donors showed relatively high CD146 expression in p0 (> $70\%$ for donor 1, >$50\%$ for donor 2), CD146 dramatically decreased in donor one cells in FBS -MMC (>$10\%$), and in donor two cells in FBS–MMC, FBS + MMC (>$15\%$). With respect to negative markers at p4, only donor one cells in XF + MMC and ACF -MMC, and donor two cells in XF -MMC and XF + MMC increased CD31 expression over $40\%$. While donor one cells did not upregulate CD45 at p4 (> $5\%$), donor two cells increased CD45 expression in most culture conditions over $40\%$ (except ACF -MMC). ## 3.2 Cell morphology analysis For both donors, qualitative cell morphology analysis (Supplementary Figure S5A-B) revealed that cell morphology was not affected as a function of ACF and XF media formulations and MMC supplementation, whilst cells expanded with serum-containing media adopted a slightly rounder, cuboidal shape at p4. ## 3.3 Cell viability analysis For cells isolated from both donors, no significant differences were observed in cell viability (Supplementary Figure S6A-D) as a function of media formulation and MMC supplementation. Percentages of live cells were ≥$90\%$ for all experimental conditions and for cells of both donors. ## 3.4 DNA concentration analysis DNA concentration analysis revealed that cells from the two donors exhibited different proliferation behaviours in the various media (Supplementary Figure S7A-B). For donor 1 (Supplementary Figure S7A), no statistically significant differences were evident between groups after 4 days of culture. At day 10, serum-containing conditions showed significantly lower ($p \leq 0.05$) DNA concentration when compared to all XF- and ACF conditions, regardless of MMC. DNA concentration analysis for donor 2 (Supplementary Figure S7B) revealed that the FBS + MMC induced significantly ($p \leq 0.05$) highest DNA concentration among all groups at day 4At day 10, DNA concentration in XF -MMC, XF + MMC, and ACF -MMC was significantly ($p \leq 0.05$) higher than DNA concentration in all other conditions. Overall, serum-containing conditions showed lower proliferation rates than XF and ACF conditions. ## 3.5 Cell metabolic activity analysis Cell metabolic activity analysis revealed that cells from the two donors exhibited different metabolic activity in the various media (Supplementary Figure S7C-D). For donor 1 (Supplementary Figure S7C), cell metabolic activity analysis at day 4 revealed that the FBS + MMC and XF + MMC induced significantly ($p \leq 0.05$) higher metabolic activity than the XF -MMC, ACF -MMC and ACF + MMC. At day 10, cell metabolic activity in FBS was significantly ($p \leq 0.001$) higher than cell metabolic activity in ACF and XF. For donor 2 (Supplementary Figure S7D), cell metabolic activity in FBS -MMC, was significantly ($p \leq 0.05$) higher than cell metabolic activity in all other conditions at day 4. At day 10, cell metabolic activity in FBS -MMC was significantly higher than cell metabolic activity in all other conditions ($p \leq 0.05$). Overall, a decrease in metabolic activity from day 4 to day 10 was observed for cells from both donors in all culture conditions. ## 3.6 Electrophoresis analysis SDS-PAGE and corresponding densitometric analysis (Figures 2A–D) revealed similar collagen deposition profiles for cells from both donors. FIGURE 2: SDS-PAGE and corresponding densitometric analysis of hBMSCs of donor 1 (A, B) and 2 (C, D) at p4 after 4 and 10 days of culture, expanded with or without MMC in ACF, XF and FBS containing media. Experiments for cells of each donor were performed in three technical replicates. # indicates the lowest statistically significant value (p < 0.05) at a given time point. For both donors at day 4, no significant ($p \leq 0.05$) differences in collagen deposition was observed among all conditions. For both donors at day 10, FBS -MMC showed significantly lower ($p \leq 0.05$) collagen deposition compared to all other groups, regardless of MMC. In all ACF and XF conditions, the presence of MMC did not significantly ($p \leq 0.05$) affect collagen deposition. No protein bands were detected in silver-stained SDS-PAGE when attachment solutions only were analysed (Supplementary Figure S11A). In summary, MMC increased collagen type I deposition in FBS (both donors) and in XF (donor 2) media at day 10, and XF and ACF conditions induced overall higher collagen deposition in donor two cells compared to donor one cells on day 10. ## 3.7 Immunocytochemistry analysis Immunocytochemistry (ICC) for collagen type I (Supplementary Figure S8A–C) and complementary relative fluorescence intensity analysis (Supplementary Figure S8B–D) show similar collagen type I deposition profiles for cells of both donors. While MMC did not significantly ($p \leq 0.05$) increase collagen type I deposition in all ACF and XF conditions for donor 1 (Supplementary Figure S8A-B), MMC significantly ($p \leq 0.05$) increased collagen type I deposition in FBS at day 4 and day 10. Overall, collagen type I deposition did not significantly ($p \leq 0.05$) increase between day 4 and day 10 in none of the groups. For donor 2 (Supplementary Figure S8C-D), the ACF with MMC at day 4 resulted in the highest ($p \leq 0.05$) collagen type I deposition across all groups and time points. At day 10, collagen type I deposition was significantly ($p \leq 0.05$) lowest in FBS -MMC. ICC for collagen type I and fibronectin of attachment solutions only did not show any positive staining (Supplementary Figure S11B). In summary, XF and ACF conditions induced overall higher collagen type I deposition in donor two cells compared to donor one cells on day 10. ## 3.8 Trilineage differentiation analysis Phase contrast images (Supplementary Figure S9A-B) and Oil Red O staining and corresponding absorbance analysis (Figures 3A–D) of hBMSCs expanded without (−) or with (+) MMC in the respective expansion media and differentiated with adipogenic induction media without (−) or with (+) MMC at p4 revealed differences with respect to donor, media, and MMC. For donor 1 (Figure 3A), FBS condition showed higher lipid droplet accumulation compared to ACF and XF conditions when analysed qualitatively, regardless of whether MMC was used during expansion or differentiation. However, these differences were not statistically significant ($p \leq 0.05$) (Figure 3B). No lipid droplet accumulation was detected in XF +/+ conditions. For donor 2 (Figures 3C,D), FBS conditions induced significantly ($p \leq 0.05$) higher lipid droplet accumulation compared to ACF and XF conditions, regardless of whether MMC was used during expansion or differentiation. Qualitatively (Figure 3C), FBS +/+ induced higher lipid deposition than FBS expanded without MMC, regardless of the presence of MMC in the differentiation media (FBS −/−, FBS −/+). Overall, cells of both donors expanded in FBS conditions showed higher adipogenic differentiation potential than all ACF and XF groups. FIGURE 3: Oil Red O staining and corresponding semi-quantitative absorbance analysis of hBMSCs of donor 1 (A, B) and 2 (C, D), expanded with or without MMC in ACF, XF and FBS containing media, and differentiated with adipogenic induction media with or without MMC supplementation at p4. Experiments were performed in three technical replicates. # indicates the lowest statistically significant value (p < 0.05) at a given time point. indicates the highest statistically significant value (p < 0.05) at a given time point. Scale bars: 100 μm.* Phase contrast images (Supplementary Figure S10A-B) and alizarin red staining and corresponding absorbance analysis (Figures 4A–D) of hBMSCs expanded without (−) or with (+) MMC in the respective expansion media and differentiated with osteogenic induction media without (−) or with (+) MMC in p4 revealed differences with respect to donor, media, and MMC. For both donors (Figures 4A–D), no calcium deposition was detected in FBS groups, regardless of whether MMC was used during expansion or differentiation. For donor 1 (Figures 4A,B), no calcium deposition was detected in XF groups, regardless of whether MMC was used during expansion or differentiation. Calcium deposition was detected in all ACF groups; the ACF −/+ group induced significantly ($p \leq 0.05$) higher calcium deposition than the ACF +/+ and ACF −/− groups. For donor 2 (Figures 4C,D), no calcium deposition was detected when hBMSCs were expanded in XF without MMC, independently on whether MMC was used during differentiation. When hBMSCs were expanded in XF with MMC, the XF +/+ induced significantly ($p \leq 0.05$) higher calcium deposition than the XF +/−. ACF groups showed calcium deposition only when hBMSCs were expanded without MMC and the ACF −/− induced significantly ($p \leq 0.05$) higher calcium deposition than ACF −/+ and all other ACF, XF and FBS conditions. Overall, cells of both donors showed high osteogenic potential when expanded in ACF -MMC conditions. FIGURE 4: Alizarin red staining and corresponding semi-quantitative absorbance analysis of hBMSCs of donor 1 (A, B) and 2 (C, D), expanded with or without MMC in ACF, XF and FBS containing media, and differentiated with osteogenic induction media with or without MMC supplementation at p4. Experiments were performed in three technical replicates. # indicates the lowest statistically significant value (p < 0.05) at a given time point. indicates the highest statistically significant value (p < 0.05) at a given time point. Scale bars: 100 μm.* Qualitative analysis of Alcian blue staining (Figures 5A,B) of hBMSCs expanded without (−) or with (+) MMC in the respective expansion media and differentiated with chondrogenic induction media without (−) or with (+) MMC in p4 for donor one revealed a proteoglycan-rich ECM in all FBS conditions, regardless of whether MMC was used during expansion or differentiation (Figure 5A). For ACF and XF conditions, only the XF ± and XF +/+ groups resulted in proteoglycan-rich ECM. For the ACF groups, proteoglycan-rich ECM was detected for ACF −/− and little to no chondrogenic differentiation was detected for ACF −/+. When hBMSCs were expanded in ACF with MMC, a proteoglycan-rich ECM was detected independently on whether MMC was used during differentiation. FIGURE 5: Alcian blue staining for sulphated proteoglycans of hBMSCs of donor 1 (A) and 2 (B), expanded with or without MMC in ACF, XF and FBS containing media, and differentiated in pellet culture with chondrogenic differentiation media with or without MMC supplementation at p4. Experiments were performed in three technical replicates. For donor 2, qualitative analysis of Alcian blue staining (Figure 5B) revealed a proteoglycan-rich ECM for all conditions, except for FBS −/+, FBS +/+ and XF −/+. Overall, donor two chondrogenic pellets showed a relatively lower proteoglycan content when compared to donor one groups. Overall, donor one cells expanded with MMC showed a higher proteoglycan deposition and chondrocyte-like cells with lacunae-formation across all groups compared to donor two cells. ## 4 Discussion Animal-derived cell culture media and absence of native ECM are associated with hBMSC phenotypic drift and loss of their therapeutic potential. To alleviate these issues in the developmental cycle of stem cell-based medicines, the use of either ACF or XF media formulations and MMC have been proposed. Interestingly, their combined effect has only been assessed once in human adipose-derived mesenchymal stromal cell cultures (Patrikoski et al., 2017). Herein, we ventured to assess whether ACF or XF media formulations supplemented with MMC (either/or during expansion and either/or during differentiation) can more effectively control hBMSCs fate than FBS-based media formulations supplemented with MMC (either/or during expansion and either/or during differentiation). To induce artificial polydispersity for maximum excluded volume effect, and consequently increased ECM deposition by cells, we used a previously optimised (Gaspar et al., 2019) MMC cocktail with different molecular weights of the same crowder (Ficoll®). ## 4.1 Surface marker analysis While hBMSCs of both donors exhibited a surface marker profile according to the International Society for Cellular Therapy criteria (Dominici et al., 2006) at p0, donor two cells showed elevated levels of the haematopoietic marker CD45. It was previously reported that freshly isolated hBMSCs expressed CD45, while in cultured MSCs and cells at later passages CD45 was downregulated. These cells were capable to differentiate into osteochondroblastic cells, adipocytes and stromacytes. These results are in agreement with our results for donor two hBMSCs expanded with ACF -MMC. Even though cells mildly expressed CD45 at p0, the hematopoietic marker was almost absent at p4 when kept in ACF media without MMC during expansion (Deschaseaux et al., 2003). At p4, a decrease in CD105 and CD146 and an increase in CD31 and CD45 levels was observed at p4 in some conditions. It is worth noting that one study has shown the expression of CD146 in hBMSCs to decrease with increasing passage number (from $95.1\%$ in p3 to $49.7\%$ in p8) (Yang et al., 2018b), whilst other studies have shown the expression of CD105 to increase in hADSCs in higher [e.g., p3 (Yoshimura et al., 2006), p4 (Varma et al., 2007)] passages. The different expression patterns for CD105 and CD146 could therefore be influenced by donor, culture media and MMC. However, it needs to be noted that hADSCs have a different surface marker profile than hBMSCs. CD31 increased to highest values in donor two cells expanded in XF media, regardless of the presence of MMC. While CD45 remained low in all donor one conditions, for donor 2, it remained low only for ACF -MMC and increased to highest values in both XF media conditions. These results clearly suggest donor variability and effects of culture media on stem cell phenotype. It is also worth noting that a study has shown high expression of CD45 in freshly isolated and commercially available hBMSCs from different donors throughout the expansion period (Okolicsanyi et al., 2015). Furthermore, increased expression levels of CD31 and CD45 in later passage adipose-derived mesenchymal stromal cells were associated with presence of endothelial and haematopoietic cells (Wan Safwani et al., 2011). All these data indicate that attention should be paid when surface markers are used to characterise MSCs, as variations in their expression is well documented in the literature (Mafi et al., 2011). Among both donors and all conditions assessed herein, cells expanded in ACF media showed overall high surface expression of positive MSC markers and relatively low expression of CD31 and CD45, indicative of phenotype maintenance, despite a relatively low CD105 expression in donor two cells. Across all media, donor one cells better maintained their stromal phenotype compared to donor two cells. This is in accordance with previous publications with hBMSCs (Russell et al., 2010; Siegel et al., 2013) and can be attributed to donor variability. To be consistent with other experiments at p4, immunophenotyping at p4 was performed after 10 days of culture when cells have had grown to confluency, which may have affected their surface marker profile. ## 4.2 Basic cellular function analysis Cell viability of both donors was not negatively affected by media conditions. With respect to DNA concentration, the only clear trend observed among both donors was that in the absence of MMC, the ACF and XF media formulations induced significantly higher DNA concentration than their FBS counterparts. This increased DNA concentration in ACF and XF media formulations can be attributed to the various additives used to replace animal sera. For example, a previous study showed that adipose-derived mesenchymal stromal cells cultured in serum-free media had a higher population doubling time compared to adipose-derived mesenchymal stromal cells cultured with FBS (Lee et al., 2022). Another study showed that hBMSCs cultured in serum-free media had higher proliferation rates compared to cells cultured in serum-containing media (Chase et al., 2010). With respect to metabolic activity, among both donors, the FBS without MMC almost across the board induced the highest metabolic activity. This is in agreement with a previous publication, in which adipose-derived mesenchymal stromal cells cultured with FBS had significantly higher metabolic activity compared to cells cultured in human serum or xeno-free conditions (Patrikoski et al., 2017). Further, alamarBlue® assay is based on the reduction of resazurin to resorufin by mitochondrial enzymes, like NADPH dehydrogenase (O'Brien et al., 2000). It has been demonstrated that culture expansion with serum leads to a progressive decline of intracellular NAD + levels and increase in NADH levels, which together change the redox cycle balance in high passages of hMSCs, connecting mitochondrial fitness with replicative senescence in hMSCs (Patrikoski et al., 2017; Yuan et al., 2020). Thus, the relatively higher metabolic activity of hBMSCs cultured with FBS could indicate senescence or phenotype loss. No clear trend with respect to MMC was observed for DNA concentration and metabolic activity. One should note that Ficoll™ is used extensively in stem cell purification (Jaatinen and Laine, 2007; Kawasaki-Oyama et al., 2008; Al Battah et al., 2011; Najar et al., 2014; Kakabadze et al., 2019) and as MMC agent (Zeiger et al., 2012; Rashid et al., 2014; Gaspar et al., 2019), and to-date, no negative data have been reported. Overall, differences in DNA concentration and metabolic activity between donors could also be a result of the differences observed in surface marker expression of positive and negative MSC markers. In addition, contact inhibition may have affected proliferation rates in confluent cultures on day 10. ## 4.3 Collagen deposition analysis SDS-PAGE analysis revealed that MMC increased collagen type I deposition in FBS (both donors) and in XF (donor 2) media formulations at day 10, but not at day 4. Ficoll™ cocktails have been shown repeatedly to enhance and accelerate ECM deposition in both permanently differentiated (Kumar et al., 2015b) and stem cell (Cigognini et al., 2016) cultures. Further, Ficoll™ cocktails are known to require longer periods of time to enhance ECM deposition than natural [such as carrageenan (Satyam et al., 2014; Gaspar et al., 2019)] or synthetic [such as dextran sulphate and polyvinylpyrrolidone (Chen et al., 2009; Rashid et al., 2014)] macromolecules. It is worth noting that FBS containing media induced significantly lower collagen synthesis and/or deposition (as assessed via SDS-PAGE) than the XF and ACF media formulations. We attribute these observations to the presence of matrix metalloproteinases that degrade collagen in FBS (Satyam et al., 2014; Kumar et al., 2015a) and the presence of growth factors that allow for cell attachment and growth in XF and ACF media formulations (Chase et al., 2010). Interestingly, the XF with MMC induced the highest collagen deposition, as judged by SDS-PAGE. We attribute this to the synergistic effect of the contained growth factors with MMC. Indeed, MMC combined with growth factor supplementation resulted in amplified (over cells with growth factors alone and cells with MMC alone) collagen deposition, as the growth factors enhanced collagen synthesis and MMC enhanced collagen deposition (Tsiapalis et al., 2021). Following the same reasoning, one would have expected the ACF with MMC to induce higher collagen deposition than the ACF alone, which was not the case, as judged by SDS-PAGE. For this, we believe that the concentration of the Ficoll™ cocktail used was not sufficient to effectively exclude volume and an optimisation study should be conducted. The immunofluorescence analysis only for donor one at day 10 validated the SDS-PAGE results with respect to higher ECM deposition when ACF and XF, as opposed to FBS, were used and higher ECM deposition when MMC was used in FBS cultures. MMC also effectively increased ECM deposition in ACF at both time points with donor two cells. The MMC data are in clear contradiction to previous studies (Gaspar et al., 2019), where increased ECM deposition has been comprehensively demonstrated in the presence of Ficoll™. The only logical explanation is the sensitivity of the assay, as opposed to the sensitivity of silver-stained gels than can reach 0.05–0.2 ng (Jin et al., 2004). ## 4.4 Adipogenic differentiation analysis Overall, hBMSCs isolated from both donors showed sufficient potential to differentiate into the adipogenic lineage. Even though CD31 and CD45 expression was relatively high in donor two cells at p4, previous studies reported that CD45-positive adipose-derived mesenchymal stromal cells possessed adipogenic potential in vitro (Yu et al., 2010). Qualitatively, hBMSCs expanded in serum-containing media showed higher adipogenic differentiation compared to ACF/XF conditions, which was further enhanced by MMC during expansion. An overall low lipid droplet formation in donor one cells resulted in no statistically significant differences in Oil red O uptake. Relatively high values in negative controls (no differentiation media) are due to higher cell proliferation and entrapment of Oil Red O in the deposited ECM. These results are in agreement with previous studies, where high serum concentrations ($10\%$, $20\%$ FBS) enhanced adipogenic differentiation of MSCs by activation of the MEK/ERK signalling pathway, ultimately promoting PPARγ expression and phosphorylation, compared to low serum culture ($2\%$ FBS) (Wu et al., 2010). Further, MMC (Ficoll™) was reported to increase adipogenic differentiation of hBMSCs by promoting a pro-adipogenic microenvironment (Levengood and Zhang, 2014); to facilitate brown adipocyte differentiation through MMC-enhanced collagen type IV formation in adult hBMSCs (Lee et al., 2016) and to promote the differentiation of adipocytes (Chen et al., 2023). ## 4.5 Osteogenic differentiation analysis As no calcium deposition was detected in FBS conditions for both donors, independent of MMC, cells possibly committed towards the adipogenic lineage during expansion, as evidenced by high adipogenic differentiation capacity. In XF media, only donor two cells deposited abundant calcium nodules when expanded with +MMC. Interestingly, these groups showed the highest expression of CD31 and CD45. Previous studies reported that CD45-positive adipose-derived mesenchymal stromal cells possessed osteogenic potential in vitro (Yu et al., 2010). For donor two cells expanded with MMC, MMC during differentiation significantly increased calcium nodule formation. Studies showed carrageenan (Cigognini et al., 2016; Graceffa and Zeugolis, 2019) and dextran sulphate 500 kDa (Assunção et al., 2020) enhanced osteogenic differentiation of BMSCs, in serum-containing culture. Other studies showed osteogenic potential of a Ficoll™ 70 kDa and Ficoll™ 400 kDa cocktail (Patrikoski et al., 2017) and sulphated seaweed polysaccharides (De Pieri et al., 2020) in serum-containing human adipose-derived mesenchymal stromal cell cultures. CD146 is considered one of the most appropriate stemness markers, as it is universally detected in MSCs isolated from various tissues and associated with higher multipotency (Crisan et al., 2008; Lv et al., 2014). CD146 was reported to be heterogeneous on a subset of BMSCs (Bühring et al., 2009). Osteoprogenitor cells have been reported to be highly positive for CD146 (Sacchetti et al., 2007) and thus a reduction in CD146 over passages may compromise their osteogenic potential (Yang et al., 2018b). Interestingly, cells of both donors expanded with ACF medium showed sufficient osteogenic potential, except for donor two cells when MMC was present during differentiation. Cells of both donors expanded with ACF/XF media showed higher expression of CD146, compared to FBS, indicating that ACF/XF media supported multipotency, and therefore possibly higher osteogenic potential. Differences in osteogenic potential observed in our study can further be attributed to donor variability and differences in media formulations. A previous study reported variability in osteogenic potential of MSCs from 19 different donors, irrespective of age, gender, and source of isolation, which was attributed to cellular heterogeneity among donors (Siddappa et al., 2007). ## 4.6 Chondrogenic differentiation analysis A lower chondrogenic and osteogenic potential of donor two cells corroborates with relatively high expression of CD31 and CD45, indicating phenotype loss. Even though donor one cells show decreased CD105 expression at p4, previous studies showed that chondrogenic differentiation potential of BMSCs was not linked to CD105 levels (Cleary et al., 2016). Across both donors, MMC during expansion increased proteoglycan-deposition in hBMSCs. MMC has previously been shown to increase chondrogenic differentiation in hBMSCs (Cigognini et al., 2016), and human adipose-derived mesenchymal stromal cells (De Pieri et al., 2020). CD146+ MSCs have been associated with enhanced chondrogenesis and greater therapeutic potential for collagen-induced arthritis (Hagmann et al., 2014; Wu et al., 2016; Li et al., 2019). Here, higher CD146 expression in donor one compared to donor two cells is reflected in its chondrogenic potential. Interestingly, high deposition of sulphated proteoglycans in MMC groups coincides with higher expression of CD146 in p4. Both supportive and inhibitory effects of MMC on human adipose-derived mesenchymal stromal cell culture have been shown to be culture condition dependent (Mittal et al., 2015; Patrikoski et al., 2017). ## 4.7 Limitations and future perspectives Due to unknown composition of ACF/XF media, some results could not be fully explained. Possible inhibitory interactions of Ficoll™ with certain media components cannot be ruled out. Changing media formulations between cell isolation and expansion may have affected cell phenotype and differentiation capacity. To limit the introduction of unknown variables to this study cells of different experimental groups were isolated using the same isolation media and protocol and differentiated with the same differentiation media and protocols. Therefore, the experimental groups only differed in media and protocols for expansion. Future studies could use compatible media and protocols for isolation, expansion, and differentiation derived from the same company, for each experimental group. An additional positive control group per donor should be added, in which hBMSCs are isolated and expanded with FBS-containing media, and trilineage experiments for this group are performed with established in-house protocols. In addition, future studies should use a higher number of donors to account for donor variability, and calculate cumulative population doubling (cPD) levels instead of passage number to appropriately track cellular aging across different conditions. ## 5 Conclusion Contemporary tissue engineered therapies require development of clinically relevant cell culture media that enhance and accelerate ECM deposition to reduce manufacturing costs, whilst maintaining cellular phenotype and function. In this context, herein we studied the influence of culture media (FBS, ACF, XF) without/with MMC in hBMSC (from two different donors) fate. Our data indicate that cell behaviour depends on donor and media formulation. Investigators should carefully select culture conditions for cell expansion and differentiation and consider potential cross-reactions between media supplements (e.g., macromolecular crowding molecules) and base media (e.g., chemically defined media). ## 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. ## Author contributions Conceptualization: SK, DG, and DZ; Methodology: SK, DG, AN, and DZ; Formal Analysis: SK and AN; Investigation: SK, AN, and DG; Writing Original Draft: SK and AN; Writing, Reviewing and Editing: DZ, SK, and AN; Funding Acquisition: DZ; Resources: DZ; Supervision: DZ and SK. ## 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.2023.1136827/full#supplementary-material ## References 1. Al Battah F., De Kock J., Ramboer E., Heymans A., Vanhaecke T., Rogiers V.. Evaluation of the multipotent character of human adipose tissue-derived stem cells isolated by Ficoll gradient centrifugation and red blood cell lysis treatment. Vitro (2011) 25 1224-1230. DOI: 10.1016/j.tiv.2011.05.024 2. Assunção M., Wong C. W., Richardson J. J., Tsang R., Beyer S., Raghunath M.. Macromolecular dextran sulfate facilitates extracellular matrix deposition by electrostatic interaction independent from a macromolecular crowding effect. Mater Sci. Eng. C Mater Biol. Appl. (2020) 106 110280. DOI: 10.1016/j.msec.2019.110280 3. Bara J. J., Richards R. G., Alini M., Stoddart M. J.. Concise review: Bone marrow-derived mesenchymal stem cells change phenotype following. Stem Cells (2014) 32 1713-1723. DOI: 10.1002/stem.1649 4. Bühring H. J., Treml S., Cerabona F., de Zwart P., Kanz L., Sobiesiak M.. Phenotypic characterization of distinct human bone marrow-derived MSC subsets. Ann. N. Y. Acad. Sci. (2009) 1176 124-134. DOI: 10.1111/j.1749-6632.2009.04564.x 5. Capella-Monsonis H., Coentro J. Q., Graceffa V., Wu Z., Zeugolis D. I.. An experimental toolbox for characterization of mammalian collagen type I in biological specimens. Nat. Protoc. (2018) 13 507-529. DOI: 10.1038/nprot.2017.117 6. Chase L. G., Lakshmipathy U., Solchaga L. A., Rao M. S., Vemuri M. C.. A novel serum-free medium for the expansion of human mesenchymal stem cells. Stem Cell Res. Ther. (2010) 1 8. DOI: 10.1186/scrt8 7. Chase L. G., Yang S., Zachar V., Yang Z., Lakshmipathy U., Bradford J.. Development and characterization of a clinically compliant xeno-free culture medium in good manufacturing practice for human multipotent mesenchymal stem cells. Stem Cells Transl. Med. (2012) 1 750-758. DOI: 10.5966/sctm.2012-0072 8. Chen C. Z., Peng Y. X., Wang Z. B., Fish P. V., Kaar J. L., Koepsel R. R.. The Scar-in-a-Jar: Studying potential antifibrotic compounds from the epigenetic to extracellular level in a single well. Br. J. Pharmacol. (2009) 158 1196-1209. DOI: 10.1111/j.1476-5381.2009.00387.x 9. Chen H., Yan X., Sun A., Zhang L., Zhang J., Yan Y.. Adipose extracellular matrix deposition is an indicator of obesity and metabolic disorders. J. Nutr. Biochem. (2023) 111 109159. DOI: 10.1016/j.jnutbio.2022.109159 10. Cheng C. W., Solorio L. D., Alsberg E.. Decellularized tissue and cell-derived extracellular matrices as scaffolds for orthopaedic tissue engineering. Biotechnol. Adv. (2014) 32 462-484. DOI: 10.1016/j.biotechadv.2013.12.012 11. Cigognini D., Gaspar D., Kumar P., Satyam A., Alagesan S., Sanz-Nogues C.. Macromolecular crowding meets oxygen tension in human mesenchymal stem cell culture - a step closer to physiologically relevant. Sci. Rep. (2016) 6 30746. DOI: 10.1038/srep30746 12. Cleary M. A., Narcisi R., Focke K., van der Linden R., Brama P. A., van Osch G. J.. Expression of CD105 on expanded mesenchymal stem cells does not predict their chondrogenic potential. Osteoarthr. Cartil. (2016) 24 868-872. DOI: 10.1016/j.joca.2015.11.018 13. Crisan M., Yap S., Casteilla L., Chen C. W., Corselli M., Park T. S.. A perivascular origin for mesenchymal stem cells in multiple human organs. Cell Stem Cell (2008) 3 301-313. DOI: 10.1016/j.stem.2008.07.003 14. De Pieri A., Rana S., Korntner S., Zeugolis D. I.. Seaweed polysaccharides as macromolecular crowding agents. Int. J. Biol. Macromol. (2020) 164 434-446. DOI: 10.1016/j.ijbiomac.2020.07.087 15. de Soure A. M., Fernandes-Platzgummer A., da Silva C. L., Cabral J. M.. Scalable microcarrier-based manufacturing of mesenchymal stem/stromal cells. J. Biotechnol. (2016) 236 88-109. DOI: 10.1016/j.jbiotec.2016.08.007 16. Deans R. J., Moseley A. B.. Mesenchymal stem cells: Biology and potential clinical uses. Exp. Hematol. (2000) 28 875-884. DOI: 10.1016/s0301-472x(00)00482-3 17. Deschaseaux F., Gindraux F., Saadi R., Obert L., Chalmers D., Herve P.. Direct selection of human bone marrow mesenchymal stem cells using an anti-CD49a antibody reveals their CD45med,low phenotype. Br. J. Haematol. (2003) 122 506-517. DOI: 10.1046/j.1365-2141.2003.04469.x 18. Di Nicola M., Carlo-Stella C., Magni M., Milanesi M., Longoni P. D., Matteucci P.. Human bone marrow stromal cells suppress T-lymphocyte proliferation induced by cellular or nonspecific mitogenic stimuli. Blood (2002) 99 3838-3843. DOI: 10.1182/blood.v99.10.3838 19. Dominici M., Le Blanc K., Mueller I., Slaper-Cortenbach I., Marini F., Krause D.. Minimal criteria for defining multipotent mesenchymal stromal cells. The International Society for Cellular Therapy position statement. Cytotherapy (2006) 8 315-317. DOI: 10.1080/14653240600855905 20. Elgaz S., Kuçi Z., Kuçi S., Bönig H., Bader P.. Clinical use of mesenchymal stromal cells in the treatment of acute graft-versus-host disease. Transfus. Med. Hemother (2019) 46 27-34. DOI: 10.1159/000496809 21. Gaspar D., Fuller K. P., Zeugolis D. I.. Polydispersity and negative charge are key modulators of extracellular matrix deposition under macromolecular crowding conditions. Acta Biomater. (2019) 88 197-210. DOI: 10.1016/j.actbio.2019.02.050 22. Gattazzo F., Urciuolo A., Bonaldo P.. Extracellular matrix: A dynamic microenvironment for stem cell niche. Biochim. Biophys. Acta (2014) 1840 2506-2519. DOI: 10.1016/j.bbagen.2014.01.010 23. Graceffa V., Zeugolis D. I.. Carrageenan enhances chondrogenesis and osteogenesis in human bone marrow stem cell culture. Eur. Cell Mater (2019) 37 310-332. DOI: 10.22203/eCM.v037a19 24. Hagmann S., Frank S., Gotterbarm T., Dreher T., Eckstein V., Moradi B.. Fluorescence activated enrichment of CD146+ cells during expansion of human bone-marrow derived mesenchymal stromal cells augments proliferation and GAG/DNA content in chondrogenic media. BMC Musculoskelet. Disord. (2014) 15 322. DOI: 10.1186/1471-2474-15-322 25. Hawkes P. W.. Fetal bovine serum: Geographic origin and regulatory relevance of viral contamination. Bioresour. Bioprocess (2015) 2 34. DOI: 10.1186/s40643-015-0063-7 26. Heiskanen A., Satomaa T., Tiitinen S., Laitinen A., Mannelin S., Impola U.. N-glycolylneuraminic acid xenoantigen contamination of human embryonic and mesenchymal stem cells is substantially reversible. Stem Cells (2007) 25 197-202. DOI: 10.1634/stemcells.2006-0444 27. Honn K. V., Singley J. A., Chavin W.. Fetal bovine serum: A multivariate standard. Proc. Soc. Exp. Biol. Med. (1975) 149 344-347. DOI: 10.3181/00379727-149-38804 28. Horwitz E. M., Gordon P. L., Koo W. K., Marx J. C., Neel M. D., McNall R. Y.. Isolated allogeneic bone marrow-derived mesenchymal cells engraft and stimulate growth in children with osteogenesis imperfecta: Implications for cell therapy of bone. Proc. Natl. Acad. Sci. U. S. A. (2002) 99 8932-8937. DOI: 10.1073/pnas.132252399 29. Jaatinen T., Laine J.. Isolation of mononuclear cells from human cord blood by Ficoll-Paque density gradient. Curr. Protoc. Stem Cell Biol. Chapter (2007) 2 Unit 2A.1. DOI: 10.1002/9780470151808.sc02a01s1 30. Jin L. T., Hwang S. Y., Yoo G. S., Choi J. K.. Sensitive silver staining of protein in sodium dodecyl sulfate-polyacrylamide gels using an azo dye, calconcarboxylic acid, as a silver-ion sensitizer. Electrophoresis (2004) 25 2494-2500. DOI: 10.1002/elps.200306002 31. Jorgensen C., Djouad F., Apparailly F., Noël D.. Engineering mesenchymal stem cells for immunotherapy. Gene Ther. (2003) 10 928-931. DOI: 10.1038/sj.gt.3302019 32. Jung S., Panchalingam K. M., Rosenberg L., Behie L. A.. Stem Cells Int. (2012) 2012 1-21. DOI: 10.1155/2012/123030 33. Kakabadze Z., Chakhunashvili D., Gogilashvili K., Ediberidze K., Chakhunashvili K., Kalandarishvili K.. Bone marrow stem cell and decellularized human amniotic membrane for the treatment of nonhealing wound after radiation therapy. Exp. Clin. Transpl. (2019) 17 92-98. DOI: 10.6002/ect.MESOT2018.O29 34. Karnieli O., Friedner O. M., Allickson J. G., Zhang N., Jung S., Fiorentini D.. A consensus introduction to serum replacements and serum-free media for cellular therapies. Cytotherapy (2017) 19 155-169. DOI: 10.1016/j.jcyt.2016.11.011 35. Kawasaki-Oyama R., Braile D., Caldas H., Leal J., Goloni-Bertollo E., Pavarino-Bertelli E.. Blood mesenchymal stem cell culture from the umbilical cord with and without Ficoll-Paque density gradient method. Rev. Bras. Cir. Cardiovasc (2008) 23 29-34. DOI: 10.1590/s0102-76382008000100006 36. Kinzebach S., Bieback K.. Expansion of mesenchymal stem/stromal cells under xenogenic-free culture conditions. Adv. Biochem. Eng. Biotechnol. (2013) 129 33-57. DOI: 10.1007/10_2012_134 37. Kumar P., Satyam A., Cigognini D., Pandit A., Zeugolis D. I.. Low oxygen tension and macromolecular crowding accelerate extracellular matrix deposition in human corneal fibroblast culture. J. Tissue Eng. Regen. Med. (2018) 12 6-18. DOI: 10.1002/term.2283 38. Kumar P., Satyam A., Fan X., Collin E., Rochev Y., Rodriguez B. J.. Macromolecularly crowded. Sci. Rep. (2015a) 5 8729. DOI: 10.1038/srep08729 39. Kumar P., Satyam A., Fan X., Rochev Y., Rodriguez B. J., Gorelov A.. Accelerated development of supramolecular corneal stromal-like assemblies from corneal fibroblasts in the presence of macromolecular crowders. Tissue Eng. Part C Methods (2015b) 21 660-670. DOI: 10.1089/ten.TEC.2014.0387 40. L'Heureux N., Dusserre N., Konig G., Victor B., Keire P., Wight T. N.. Human tissue-engineered blood vessels for adult arterial revascularization. Nat. Med. (2006) 12 361-365. DOI: 10.1038/nm1364 41. Lareu R. R., Subramhanya K. H., Peng Y., Benny P., Chen C., Wang Z.. Collagen matrix deposition is dramatically enhanced. FEBS Lett. (2007) 581 2709-2714. DOI: 10.1016/j.febslet.2007.05.020 42. Lee J. Y., Kang M. H., Jang J. E., Lee J. E., Yang Y., Choi J. Y.. Comparative analysis of mesenchymal stem cells cultivated in serum free media. Sci. Rep. (2022) 12 8620. DOI: 10.1038/s41598-022-12467-z 43. Lee M. H., Goralczyk A. G., Kriszt R., Ang X. M., Badowski C., Li Y.. ECM microenvironment unlocks Brown adipogenic potential of adult human bone marrow-derived MSCs. Sci. Rep. (2016) 6 21173. DOI: 10.1038/srep21173 44. Levengood S. L., Zhang M.. Chitosan-based scaffolds for bone tissue engineering. J. Mater Chem. B (2014) 2 3161-3184. DOI: 10.1039/C4TB00027G 45. Li X., Guo W., Zha K., Jing X., Wang M., Zhang Y.. Enrichment of CD146(+) adipose-derived stem cells in combination with articular cartilage extracellular matrix scaffold promotes cartilage regeneration. Theranostics (2019) 9 5105-5121. DOI: 10.7150/thno.33904 46. Lv F. J., Tuan R. S., Cheung K. M., Leung V. Y.. Concise review: The surface markers and identity of human mesenchymal stem cells. Stem Cells (2014) 32 1408-1419. DOI: 10.1002/stem.1681 47. Mafi P., Hindocha S., Mafi R., Griffin M., Khan W. S.. Adult mesenchymal stem cells and cell surface characterization - a systematic review of the literature. Open J. Orthop. (2011) 5 253-260. DOI: 10.2174/1874325001105010253 48. Mittal S., Chowhan R. K., Singh L. R.. Macromolecular crowding: Macromolecules friend or foe. Biochim. Biophys. Acta (2015) 1850 1822-1831. DOI: 10.1016/j.bbagen.2015.05.002 49. Najar M., Rodrigues R. M., Buyl K., Branson S., Vanhaecke T., Lagneaux L.. Proliferative and phenotypical characteristics of human adipose tissue-derived stem cells: Comparison of Ficoll gradient centrifugation and red blood cell lysis buffer treatment purification methods. Cytotherapy (2014) 16 1220-1228. DOI: 10.1016/j.jcyt.2014.05.021 50. O'Brien J., Wilson I., Orton T., Pognan F.. Investigation of the Alamar Blue (resazurin) fluorescent dye for theassessment of mammalian cell cytotoxicity. Eur. J. Biochem. (2000) 267 5421-5426. DOI: 10.1046/j.1432-1327.2000.01606.x 51. Okolicsanyi R. K., Camilleri E. T., Oikari L. E., Yu C., Cool S. M., van Wijnen A. J.. Human mesenchymal stem cells retain multilineage differentiation capacity including neural marker expression after extended. PLoS One (2015) 10 e0137255. DOI: 10.1371/journal.pone.0137255 52. Patrikoski M., Lee M. H. C., Makinen L., Ang X. M., Mannerstrom B., Raghunath M.. Effects of macromolecular crowding on human adipose stem cell culture in fetal bovine serum, human serum, and defined xeno-free/serum-free conditions. Stem Cells Int. (2017) 2017 1-14. DOI: 10.1155/2017/6909163 53. Pei M., He F., Kish V. L.. Expansion on extracellular matrix deposited by human bone marrow stromal cells facilitates stem cell proliferation and tissue-specific lineage potential. Tissue Eng. Part A (2011) 17 3067-3076. DOI: 10.1089/ten.TEA.2011.0158 54. Prewitz M. C., Stißel A., Friedrichs J., Träber N., Vogler S., Bornhäuser M.. Extracellular matrix deposition of bone marrow stroma enhanced by macromolecular crowding. Biomaterials (2015) 73 60-69. DOI: 10.1016/j.biomaterials.2015.09.014 55. Raghunath M., Zeugolis D.. Transforming eukaryotic cell culture with macromolecular crowding. Trends Biochem. Sci. (2021) 46 805-811. DOI: 10.1016/j.tibs.2021.04.006 56. Rashid R., Lim N. S., Chee S. M., Png S. N., Wohland T., Raghunath M.. Novel use for polyvinylpyrrolidone as a macromolecular crowder for enhanced extracellular matrix deposition and cell proliferation. Tissue Eng. Part C Methods (2014) 20 994-1002. DOI: 10.1089/ten.TEC.2013.0733 57. Roskelley C. D., Bissell M. J.. Dynamic reciprocity revisited: A continuous, bidirectional flow of information between cells and the extracellular matrix regulates mammary epithelial cell function. Biochem. Cell Biol. (1995) 73 391-397. DOI: 10.1139/o95-046 58. Russell K. C., Phinney D. G., Lacey M. R., Barrilleaux B. L., Meyertholen K. E., O'Connor K. C.. Stem Cells (2010) 28 788-798. DOI: 10.1002/stem.312 59. Sacchetti B., Funari A., Michienzi S., Di Cesare S., Piersanti S., Saggio I.. Self-renewing osteoprogenitors in bone marrow sinusoids can organize a hematopoietic microenvironment. Cell (2007) 131 324-336. DOI: 10.1016/j.cell.2007.08.025 60. Satyam A., Kumar P., Cigognini D., Pandit A., Zeugolis D. I.. Low, but not too low, oxygen tension and macromolecular crowding accelerate extracellular matrix deposition in human dermal fibroblast culture. Acta Biomater. (2016) 44 221-231. DOI: 10.1016/j.actbio.2016.08.008 61. Satyam A., Kumar P., Fan X., Gorelov A., Rochev Y., Joshi L.. Macromolecular crowding meets tissue engineering by self-assembly: A paradigm shift in regenerative medicine. Adv. Mater (2014) 26 3024-3034. DOI: 10.1002/adma.201304428 62. Selvaggi T. A., Walker R. E., Fleisher T. A.. Development of antibodies to fetal calf serum with arthus-like reactions in human immunodeficiency virus-infected patients given syngeneic lymphocyte infusions. Blood (1997) 89 776-779. DOI: 10.1182/blood.v89.3.776 63. Shendi D., Marzi J., Linthicum W., Rickards A. J., Dolivo D. M., Keller S.. Hyaluronic acid as a macromolecular crowding agent for production of cell-derived matrices. Acta Biomater. (2019) 100 292-305. DOI: 10.1016/j.actbio.2019.09.042 64. Siddappa R., Licht R., van Blitterswijk C., de Boer J.. Donor variation and loss of multipotency during. J. Orthop. Res. (2007) 25 1029-1041. DOI: 10.1002/jor.20402 65. Siegel G., Kluba T., Hermanutz-Klein U., Bieback K., Northoff H., Schäfer R.. Phenotype, donor age and gender affect function of human bone marrow-derived mesenchymal stromal cells. BMC Med. (2013) 11 146. DOI: 10.1186/1741-7015-11-146 66. Sundin M., Ringdén O., Sundberg B., Nava S., Götherström C., Le Blanc K.. No alloantibodies against mesenchymal stromal cells, but presence of anti-fetal calf serum antibodies, after transplantation in allogeneic hematopoietic stem cell recipients. Haematologica (2007) 92 1208-1215. DOI: 10.3324/haematol.11446 67. Tekkatte C., Gunasingh G. P., Cherian K. M., Sankaranarayanan K.. Humanized" stem cell culture techniques: The animal serum controversy. Stem Cells Int. (2011) 2011 1-14. DOI: 10.4061/2011/504723 68. Thorne J. T., Segal T. R., Chang S., Jorge S., Segars J. H., Leppert P. C.. Dynamic reciprocity between cells and their microenvironment in reproduction. Biol. Reprod. (2015) 92 25. DOI: 10.1095/biolreprod.114.121368 69. Tsiapalis D., Kearns S., Kelly J., Zeugolis D.. Growth factor and macromolecular crowding supplementation in human tenocyte culture. Biomaterials Biosyst. (2021) 1 100009. DOI: 10.1016/j.bbiosy.2021.100009 70. Tsiapalis D., Zeugolis D.. It is time to crowd your cell culture media – physicochemical considerations with biological consequences. Biomaterials (2021) 275 120943. DOI: 10.1016/j.biomaterials.2021.120943 71. van der Valk J.. Fetal bovine serum - a cell culture dilemma. Science (2022) 375 143-144. DOI: 10.1126/science.abm1317 72. van Helvert S., Storm C., Friedl P.. Mechanoreciprocity in cell migration. Nat. Cell Biol. (2018) 20 8-20. DOI: 10.1038/s41556-017-0012-0 73. Varma M. J., Breuls R. G., Schouten T. E., Jurgens W. J., Bontkes H. J., Schuurhuis G. J.. Phenotypical and functional characterization of freshly isolated adipose tissue-derived stem cells. Stem Cells Dev. (2007) 16 91-104. DOI: 10.1089/scd.2006.0026 74. Wan Safwani W. K., Makpol S., Sathapan S., Chua K. H.. The changes of stemness biomarkers expression in human adipose-derived stem cells during long-term manipulation. Biotechnol. Appl. Biochem. (2011) 58 261-270. DOI: 10.1002/bab.38 75. Whitfield M. J., Lee W. C., Van Vliet K. J.. Onset of heterogeneity in culture-expanded bone marrow stromal cells. Stem Cell Res. (2013) 11 1365-1377. DOI: 10.1016/j.scr.2013.09.004 76. Wu C. C., Liu F. L., Sytwu H. K., Tsai C. Y., Chang D. M.. CD146+ mesenchymal stem cells display greater therapeutic potential than CD146-cells for treating collagen-induced arthritis in mice. Stem Cell Res. Ther. (2016) 7 23. DOI: 10.1186/s13287-016-0285-4 77. Wu L., Cai X., Dong H., Jing W., Huang Y., Yang X.. Serum regulates adipogenesis of mesenchymal stem cells via MEK/ERK-dependent PPARγ expression and phosphorylation. J. Cell Mol. Med. (2010) 14 922-932. DOI: 10.1111/j.1582-4934.2009.00709.x 78. Yang Y., Lin H., Shen H., Wang B., Lei G., Tuan R. S.. Mesenchymal stem cell-derived extracellular matrix enhances chondrogenic phenotype of and cartilage formation by encapsulated chondrocytes. Acta Biomater. (2018a) 69 71-82. DOI: 10.1016/j.actbio.2017.12.043 79. Yang Y., Ogando C. R., Wang See C., Chang T. Y., Barabino G. A.. Changes in phenotype and differentiation potential of human mesenchymal stem cells aging. Stem Cell Res. Ther. (2018b) 9 131. DOI: 10.1186/s13287-018-0876-3 80. Yao L., Bestwick C. S., Bestwick L. A., Maffulli N., Aspden R. M.. Phenotypic drift in human tenocyte culture. Tissue Eng. (2006) 12 1843-1849. DOI: 10.1089/ten.2006.12.1843 81. Yoshimura K., Shigeura T., Matsumoto D., Sato T., Takaki Y., Aiba-Kojima E.. Characterization of freshly isolated and cultured cells derived from the fatty and fluid portions of liposuction aspirates. J. Cell Physiol. (2006) 208 64-76. DOI: 10.1002/jcp.20636 82. Yu G., Wu X., Dietrich M. A., Polk P., Scott L. K., Ptitsyn A. A.. Yield and characterization of subcutaneous human adipose-derived stem cells by flow cytometric and adipogenic mRNA analyzes. Cytotherapy (2010) 12 538-546. DOI: 10.3109/14653241003649528 83. Yuan X., Liu Y., Bijonowski B. M., Tsai A.-C., Fu Q., Logan T. M.. NAD+/NADH redox alterations reconfigure metabolism and rejuvenate senescent human mesenchymal stem cells. Commun. Biol. (2020) 3 774. DOI: 10.1038/s42003-020-01514-y 84. Zeiger A. S., Loe F. C., Li R., Raghunath M., Van Vliet K. J.. Macromolecular crowding directs extracellular matrix organization and mesenchymal stem cell behavior. PLoS One (2012) 7 e37904. DOI: 10.1371/journal.pone.0037904 85. Zeugolis D.. Bioinspired. Am. J. Physiol. Cell Physiol. (2021) 320 C842-C849. DOI: 10.1152/ajpcell.00380.2020 86. Zhang J., Huang X., Wang H., Liu X., Zhang T., Wang Y.. The challenges and promises of allogeneic mesenchymal stem cells for use as a cell-based therapy. Stem Cell Res. Ther. (2015) 6 234. DOI: 10.1186/s13287-015-0240-9
--- title: Altered gut microbiota in the early stage of acute pancreatitis were related to the occurrence of acute respiratory distress syndrome authors: - Xiaomin Hu - Ziying Han - Ruilin Zhou - Wan Su - Liang Gong - Zihan Yang - Xiao Song - Shuyang Zhang - Huijun Shu - Dong Wu journal: Frontiers in Cellular and Infection Microbiology year: 2023 pmcid: PMC10025409 doi: 10.3389/fcimb.2023.1127369 license: CC BY 4.0 --- # Altered gut microbiota in the early stage of acute pancreatitis were related to the occurrence of acute respiratory distress syndrome ## Abstract ### Background Acute respiratory distress syndrome (ARDS) is the most common cause of organ failure in acute pancreatitis (AP) patients, which associated with high mortality. Specific changes in the gut microbiota have been shown to influence progression of acute pancreatitis. We aimed to determine whether early alterations in the gut microbiota is related to and could predict ARDS occurrence in AP patients. ### Methods In this study, we performed 16S rRNA sequencing analysis in 65 AP patients and 20 healthy volunteers. The AP patients were further divided into two groups: 26 AP-ARDS patients and 39 AP-nonARDS patients based on ARDS occurrence during hospitalization. ### Results Our results showed that the AP-ARDS patients exhibited specific changes in gut microbiota composition and function as compared to subjects of AP-nonARDS group. Higher abundances of Proteobacteria phylum, Enterobacteriaceae family, Escherichia-Shigella genus, and Klebsiella pneumoniae, but lower abundances of Bifidobacterium genus were found in AP-ARDS group compared with AP-nonARDS groups. Random forest modelling analysis revealed that the Escherichia-shigella genus was effective to distinguish AP-ARDS from AP-nonARDS, which could predict ARDS occurrence in AP patients. ### Conclusions Our study revealed that alterations of gut microbiota in AP patients on admission were associated with ARDS occurrence after hospitalization, indicating a potential predictive and pathogenic role of gut microbiota in the development of ARDS in AP patients. ## Introduction Acute pancreatitis (AP), one of the most common gastrointestinal diseases, is an acute inflammatory disease with an increasing incidence worldwide (Tenner et al., 2013; Greenberg et al., 2016). In patients with AP, persistent organ failure (OF) can reach $35\%$ and is a key determinant of mortality (Johnson and Abu-Hilal, 2004; Shah and Rana, 2020). The most common cause of OF in AP is acute respiratory distress syndrome (ARDS) (Garg et al., 2005). ARDS is a type of acute, diffuse inflammatory lung injury that can lead to a high mortality rate of up to $48\%$ (Schmandt et al., 2021). Even after five years of rehabilitation, surviving ARDS patients still suffer from poor long-term quality of life; including exercise limitation, difficulty in returning to work, and high medical costs (Herridge et al., 2011). However, missed or delayed diagnosis of ARDS remains a common and challenging problem worldwide. Nearly two-thirds of patients had a delayed or missed diagnosis of ARDS. The miss rate was approximately $40\%$, and the diagnosis of half of mild ARDS patients was delayed (Bellani et al., 2020). Early recognition of ARDS ensures that patients receive appropriate treatment which relieves lung injury and improves prognosis, therefore, effective prediction methods for ARDS are urgently required (Fan et al., 2018; Pan et al., 2018; Bellani et al., 2020). AP is strongly associated with gut microbiota imbalance and an impaired epithelial barrier (Besselink et al., 2009). Compared with the healthy group, the diversity of the gut microbiota decreased; with a greater abundance of pathogenic bacteria in AP patients and lower numbers of commensal beneficial genera (Zhang et al., 2018; Zhu et al., 2019). According to the revised Atlanta classification 2012, AP can be divided into three grades: mild AP (MAP), moderately severe AP (MSAP) with transient OF, and severe AP (SAP) with persistent OF (Banks et al., 2013). In AP patients with different degrees of severity, the dominant gut microbiota also varied; with Bacteroides in MAP, Escherichia-Shigella in MSAP, and Enterococcus in SAP (Yu et al., 2020). Similar results have been reported in animal models; gut microbiota-depleted AP rats were found to have lower levels of inflammatory factors (Zheng et al., 2019; Li et al., 2020). The degree of gut barrier injury and bacterial translocation are important prognostic factors for AP (Besselink et al., 2009). Disorganized microbiota and damaged intestinal epithelium in AP patients make it easier for the endotoxin diffusion, immune cell migration, and bacteria translocation. The lung environment may be more susceptible to the gut microbiota in patients with AP. Owing to the increased gut permeability, inflammatory factors and activated trypsin could function as the gut-lung axis, thus triggering and promoting lung disease in patients with AP (Shah and Rana, 2020). In addition, gram-negative infections promote release of endotoxins and these can translocate through high-permeability gut mucosa and contribute to the development of ARDS in AP patients (AP-ARDS; “ARDS” mean for ARDS in general, “AP-ARDS” mean for ARDS in acute pancreatitis.) ( Gray et al., 2003). Bacteria can also translocate from the gut to the lung (Mukherjee and Hanidziar, 2018). Previous studies also found evidence of bacteria translocation in ARDS patients (Dickson et al., 2016; Dickson et al., 2020). The composition of gut-associated bacteria, especially Bacteroidetes and Enterobacteriaceae, increased in the lower respiratory tract of ARDS patients (Dickson et al., 2016; Siwicka-Gieroba and Czarko-Wicha, 2020). Further study found that the increase of Escherichia coli in lung was related to higher mortality of ARDS patients (Zhang et al., 2021). Studies have confirmed that the lung microbiota is associated with alveolar inflammation in ARDS (Dickson et al., 2016). Michihito et al. revealed that alterations in lung microbiota are correlated with serum IL-6 levels and hospital mortality in patients with ARDS (Kyo et al., 2019). Therefore, the gut microbiota may be involved in the pathogenesis of AP-ARDS, however, the relationship between the gut microbiota and AP-ARDS remains unknown. If early changes in gut microbiota in AP-ARDS patients can be found, they may help in the early recognition of AP-ARDS, promote early intervention, and even improve patient outcomes. Therefore, we wanted to investigate the relationship between gut microbiota and AP-ARDS by comparing the microbiota among three groups: healthy controls, AP patients without ARDS (AP-nonARDS), and AP-ARDS patients. By collecting the gut microbiota at the early stage of AP, we investigated whether gut microbiota was related to and could help predict and recognize AP-ARDS. Our study explored the potential effect of the gut-lung axis in AP-ARDS and provide identify biomarkers for prediction and early recognition of AP-ARDS. ## Study population This prospective and observational cohort study was conducted at Peking Union Medical College Hospital, Beijing, China. Twenty healthy volunteers and 75 patients were enrolled between June 2018 and July 2021. Ten AP patients were excluded due to history of comorbidities and medicine intake. All patients fulfilled the AP diagnostic criteria according to the 2012 revised Atlanta criteria and were admitted within 24 h of onset (Banks et al., 2013). The exclusion criteria were as follows: patients with chronic pancreatitis, immunosuppressive disease, inflammatory bowel disease, cancer, irritable bowel syndrome, gastroenteritis, or necrotizing enterocolitis; and use of antibiotics, probiotics, laxatives, or Chinese herbs within two months before symptom onset. Informed consent was obtained from all participants. This study was approved by the Ethics Committee of PUMCH (Identifier: JS1826; date of approval:20th February 2018. Period of validity: February 2018 to August 2020) For all patients, no ARDS was diagnosed during the first fecal sampling; however, some patients developed ARDS during hospitalization. Patients were diagnosed with ARDS according to the Berlin definition (Ranieri et al., 2012). According to PaO2/FiO2 levels, patients with ARDS were divided into three groups: mild ARDS (MARDS), moderate ARDS, and severe ARDS (Ranieri et al., 2012). Considering the higher rate of mechanical ventilation in moderate ARDS and severe ARDS patients (Fan et al., 2018) as well as the small sample size of severe ARDS group ($$n = 5$$), we combined moderate ARDS and severe ARDS as non-MARDS group for subgroup analysis. ## Collection and analysis of clinical characteristics Demographic and clinical data were collected from medical record libraries, including age, sex, body mass index (BMI), smoking history, drinking history, combined diseases, disease severity-related scores, local complications, systematic complications, and clinical outcomes. Definitions of local and systematic complications can be found in previous studies (Yu et al., 2020; Hu et al., 2021b; Yu et al., 2021). Statistical analysis of clinical characteristics was performed using SPSS Statistics 26.0 (IBM Corp., Armonk, NY, USA). The mean ± standard deviation (SD) was used to represent the data distribution. However, when the data did not fit a normal distribution, the median (interquartile range [IQR]) was used. For categorical variables, we performed the χ2 test or Fisher’s exact test; while for continuous variables, we performed the nonparametric Mann-Whitney test. A difference was considered significant when the two-sided p value was less than 0.05. ## Sample collection, DNA extraction, and 16S rRNA gene sequencing Patients with AP have difficulty defecating owing to fasting and water deprivation. Therefore, we used rectal swabs for fecal sampling, as previous studies have described (Yu et al., 2020; Yu et al., 2021). The fecal samples were immediately collected after admission, and all samples were collected within 24 h of AP onset. Then, these samples were stored at − 80°C, and microbial DNA was extracted as soon as possible. We then performed PCR amplification, library construction, Illumina (San Diego, CA, USA) MiSeq sequencing, and sequence quality control, using previously reported methods (Yu et al., 2020; Yu et al., 2021). ## Bioinformatics analysis Amplicon sequence variant (ASV) analysis was performed using EasyAmplicon (Version 1.10). We use the -derep_fullength command in VSEARCH (version 2.15) to create dereplication, denoised these unique sequences into ASVs by the -unoise3 algorithm in USEARCH (Version 10.0), created an ASVs table using the -usearch_global command, and then completed the ASVs classification using the Sintax algorithm command. ## Microbiota composition Alpha diversity analysis, including the Chao and Simpson indices, was performed using Mothur software (1.30.2). The dilution curve was plotted using R software to calculate the microbial diversity at different numbers of sequences. In the beta diversity analysis, principal coordinate analysis (PCoA) was performed using the R package vegan (v2.5-6). Based on taxonomic information, community structure analysis can be performed at various taxonomic levels. The composition of microbiota at the phylum, family, genus, and species levels was determined using the stat package in R software. Relevant analytical methods were used to detect variation in microbes between the different groups and pairwise comparisons were calculated using the Wilcoxon rank-sum test. ## Functional annotation Linear discriminant analysis (LDA) effect size (LefSe; http://huttenhower.sph.harvard.edu/galaxy) was performed to identify potential biomarkers in the different groups (LDA score>2, $p \leq 0.05$). Microbiota phenotypes were predicted using BugBase, based on normalized ASVs. The significance of the functional difference was evaluated using the Wilcoxon rank-sum test in the BugBase prediction analysis. The Random Forest R package was used to build a random forest regression model. We randomly divided the 65 samples into training sets ($70\%$) and testing sets ($30\%$) according to the 16S amplicon sequence and clinical characteristics. Bioinformatics analysis and visualization were performed using the R software. Detailed analysis methods can be found in our previous studies (Hu et al., 2021a; Hu et al., 2021b). ## Clinical characteristics of AP-ARDS patients Sixty-five AP patients and 20 healthy individuals were included in the study. Rectal swabs were collected before the occurrence of ARDS. Twenty-six patients with AP developed ARDS (AP-ARDS; mild ARDS: $$n = 12$$; moderate ARDS: $$n = 9$$; severe ARDS: $$n = 5$$) and 39 patients did not (AP-nonARDS). The average diagnosis time of ARDS was 3.46 ± 1.92 d after AP onset. Table 1 shows the demographic and clinical characteristics of the two groups. Demographic characteristics were generally balanced, however, patients with ARDS had more severe symptoms than those without ARDS (Table 1). The AP-ARDS group had a higher proportion of SAP ($2.56\%$ vs. $73.08\%$; $p \leq 0.001$) and higher disease severity-related scores compared to the AP-nonARDS group. The occurrence of acute peripancreatic fluid collection ($35.90\%$ vs. $92.31\%$; $p \leq 0.001$), systematic complications, organ failure ($10.26\%$ vs. $100.00\%$; $p \leq 0.001$), and ICU admission ($0.00\%$ vs. $73.08\%$; $p \leq 0.001$) was also significantly increased in the AP-ARDS group, except for bowel obstruction and mental status. Furthermore, the total duration of organ failure (median 0.00, IQR 0.00–0.00; vs median 85.00, IQR 41.50–276.00; $p \leq 0.001$); ICU stay (0.00 ± 0.00 vs. 7.15 ± 6.59; $p \leq 0.001$) and hospital stay (8.26 ± 6.65 vs. 23.04 ± 11.52; $p \leq 0.001$) were both longer in AP-ARDS group. Table 1 \| Unnamed: 0 \| CONTROL(n=20) \| AP-nonARDS(n=39) \| AP-ARDS(n=26) \| P value (AP-nonARDS vs AP-ARDS) \| \| --- \| --- \| --- \| --- \| --- \| \| Age (years), mean ± SD \| 37.20 ± 12.00 \| 44.15 ± 15.02 \| 48.69 ± 13.99 \| 0.135 \| \| Male, n (%) \| 11(55.00) \| 17(43.59) \| 17(65.38) \| 0.085 \| \| BMI (kg/m2), mean ± SD \| 22.80 ± 2.89 \| 26.19 ± 3.63 \| 26.48 ± 3.88 \| 0.794 \| \| Smoking, n (%) \| \| 9(23.07) \| 9(34.62) \| 0.308 \| \| Drinking, n (%) \| \| 9(23.07) \| 7(26.92) \| 0.724 \| \| Comorbid abnormalities, n (%) \| Comorbid abnormalities, n (%) \| Comorbid abnormalities, n (%) \| Comorbid abnormalities, n (%) \| Comorbid abnormalities, n (%) \| \| Hypertension \| \| 10(25.64) \| 13(50.00) \| 0.044 \| \| Diabetes \| \| 9 (23.07) \| 8(30.77) \| 0.489 \| \| Fatty liver \| \| 27(69.23) \| 16(61.54) \| 0.521 \| \| Etiology, n (%) \| \| \| \| 0.538 \| \| Biliary \| \| 18(46.15) \| 9(34.62) \| \| \| Hypertriglyceridemia \| \| 17(43.59) \| 15(57.69) \| \| \| Alcohol consumption \| \| 4(10.26) \| 2(7.69) \| \| \| Disease severity, n (%) \| \| \| \| <0.001 \| \| MAP \| \| 21(53.85) \| 0(0.00) \| \| \| MSAP \| \| 17(43.59) \| 7(26.92) \| \| \| SAP \| \| 1(2.56) \| 19(73.08) \| \| \| APACHE II, mean ± SD \| \| 3.13 ± 2.25 \| 9.96 ± 4.17 \| <0.001 \| \| SOFA score, mean ± SD; median (IQR) \| \| 0.54 ± 0.64;0.00 (0.00,1.00) \| 6.12 ± 4.09;4.00(3.00, 7.25) \| <0.001 \| \| Balthazar score E, mean ± SD \| \| 2.90 ± 1.02 \| 4.04 ± 0.72 \| <0.001 \| \| Local complications, n (%) \| Local complications, n (%) \| Local complications, n (%) \| Local complications, n (%) \| Local complications, n (%) \| \| Acute peripancreatic fluid collection (APFC) \| \| 14(35.90) \| 24(92.31) \| <0.001 \| \| Pancreatic pseudocyst (PP) \| \| 3(7.69) \| 2(7.69) \| >0.999 \| \| Acute necrotic collection (ANC) \| \| 2(5.13) \| 12(46.15) \| <0.001 \| \| Walled off necrosis (WON) \| \| 0(0.00) \| 2(7.69) \| 0.079 \| \| Infected necrosis \| \| 0(0.00) \| 8(30.77) \| 0.001 \| \| Systematic complication, n (%) \| Systematic complication, n (%) \| Systematic complication, n (%) \| Systematic complication, n (%) \| Systematic complication, n (%) \| \| Systemic inflammatory response syndrome (SIRS) \| \| 13(33.33) \| 22(84.62) \| <0.001 \| \| Acute kidney injury \| \| 1(2.56) \| 12(46.15) \| <0.001 \| \| Shock \| \| 0(0.00) \| 10(38.46) \| <0.001 \| \| Liver damage \| \| 1(2.56) \| 11(42.31) \| <0.001 \| \| Myocardial injury \| \| 1(2.56) \| 6(23.08) \| 0.009 \| \| Sepsis \| \| 1(2.56) \| 14(53.85) \| <0.001 \| \| Abdominal compartment syndrome (ACS) \| \| 1(2.56) \| 7(26.92) \| 0.003 \| \| Bowel obstruction \| \| 3(7.69) \| 7(26.92) \| 0.035 \| \| Outcome \| Outcome \| Outcome \| Outcome \| Outcome \| \| Organ failure, n (%) \| \| 4(10.26) \| 26 (100.0) \| <0.001 \| \| Organ failure duration (h), mean ± SD; median (IQR) \| \| 2.59 ± 8.30;0.00(0.00,0.00) \| 164.65 ± 159.85;85.00(41.50,276.00) \| <0.001 \| \| ICU, n (%) \| \| 0(0.00) \| 19(73.08) \| <0.001 \| \| ICU stay (days), mean ± SD; \| \| 0.00 ± 0.00 \| 7.15 ± 6.59 \| <0.001 \| \| Hospital stay (days), mean ± SD \| \| 8.26 ± 6.65 \| 23.04 ± 11.52 \| <0.001 \| \| Death, n (%) \| \| 0(0.00) \| 1(3.85) \| 0.217 \| ## Taxonomic features of gut microbiota in AP-ARDS patients We analyzed 745,895 reads that were clustered into 1910 ASVs. No statistically significant differences in the richness and diversity of the gut microbiota were noted between the AP-nonARDS and AP-ARDS groups. In the alpha diversity analysis, there were no significant differences in the Chao index ($p \leq 0.05$ between any two groups; Figure 1A). Compared with healthy controls, the Simpson index decreased in both the AP-nonARDS and AP-ARDS groups, but no differences were found between the AP-nonARDS and AP-ARDS groups (Figure 1B). In the rarefaction curve analysis, the curve tended to plateau as the number of reads increased, demonstrating that microbiota in the healthy control, AP-nonARDS, and AP-ARDS groups were abundant and evenly distributed (Figure 1C). PCoA for the beta diversity results clearly distinguished the three groups, but overlap did occur between the AP-nonARDS group and AP-ARDS group. This indicated a significant difference in the microbiota structure between healthy controls and patients with AP, possible similarities between the AP-nonARDS group and AP-ARDS group (Figure 1D). Figure 1: Diversity analysis of Control, AP-nonARDS and AP-ARDS group. (A) Chao index of α analysis; (B) Simpson index of α analysis. There are significant differences between the Control group and AP patients, but no significant difference between AP-nonARDS and AP-ARDS group. (C) Rarefaction curves analysis. (D). Principal coordinate analysis (PCoA). CONTROL, healthy population; AP-nonARDS, AP patients without ARDS; AP-ARDS, AP patients with ARDS. P < 0.05; ns: not significant.* The composition of the gut microbiota was significantly different among the three groups. At the phylum level, Proteobacteria and Bacteroidetes were both increased in patients with AP compared to healthy controls. Proteobacteria showed a gradually increase with disease progression (Figure 2A). At the family level, Enterobacteriaceae, Enterococcaceae, Bacteroidaceae, Clostridiales Incertae Sedis XI, and Prevotellaceae increased, while Ruminococcaceae decreased in patients with AP compared to healthy controls. In particular, the abundance of Enterobacteriaceae and Enterococcaceae increased with disease progression (Figure 2B). At the genus level, Escherichia-Shigella, Bacteroides, and Enterococcus were more abundant in patients with AP, while Bifidobacterium and Blautia were more abundant in healthy controls. Escherichia-Shigella and Enterococcus gradually increased while Bifidobacterium decreased with disease progression (Figure 2C). Compared to the AP-nonARDS group, 25 ASVs were enriched and 22 ASVs were depleted in the AP-ARDS group (Figure 2D). Figure 2E shows the top 11 different bacteria between the AP-ARDS and AP-nonARDS groups at the species level. Klebsiella pneumoniae (ASV_101, $p \leq 0.001$; ASV_71, $p \leq 0.001$), Prevotella copri (ASV_30, $p \leq 0.001$; ASV_111, $$p \leq 0.002$$), and Clostridium ramosum (ASV_150, $$p \leq 0.002$$) showed a significant increase; and Bifidobacterium longum (ASV_14, $$p \leq 0.003$$) decreased in the AP-ARDS group compared to the AP-nonARDS group. Among these microbiota, Clostridium ramosum (ASV_150) showed a gradual increase in the healthy to AP-nonARDS to AP-ARDS groups, whereas Bifidobacterium longum (ASV_40) showed a gradual decrease (Figure 2E). Figure 2: Gut microbiota composition at (A) phylum, (B) family, (C) genus levels. (D) Different amplicon sequence variants (ASVs) between AP-ARDS and AP-nonARDS group. (green= depleted in AP-ARDS group; red = enriched in AP-ARDS group; gray = no significantly difference). (E) Relative abundances of different species between AP-nonARDS and AP-ARDS groups. P < 0.05; *** P < 0.001; ns: not significant.* We performed a subgroup analysis according to ARDS severity, and identified some microbiota showing similar trends. At the phylum level, Proteobacteria increased in the non-MARDS group compared to that in the MARDS group (Figure 3A). At the family level, Enterobacteriaceae increased in the non-MARDS group (Figure 3B). At the genus level, Escherichia-Shigella was more abundant in the non-MARDS group (Figure 3C). Figure 3: Subgroup analysis of gut microbiota composition at phylum (A), family (B), and genus (C) levels. MARDS, mild ARDS; NonMARDS, moderate ARDS and severe ARDS. ## Alterations of gut microbiota in AP-ARDS patients are associated with more severe manifestations LEFSe analysis also revealed that Enterobacteriaceae and Escherichia-Shigella were dominant in AP-ARDS group while Enterococcaceae and Enterococcus were dominant in AP-nonARDS group (Figure 4A). Figure 4: Microbial Function Analysis and Clinical Correlation Analysis. (A). Linear discriminant analysis (LDA) Effect Size (LEfSe) analysis. (B). Relative abundance of aerobic bacteria in BugBase analysis. (C). Relative abundance of anaerobic bacteria in BugBase analysis. (D). Spearman correlation of clinical characteristics and different species between AP-ARDS and AP-nonARDS group. P < 0.05; *** P < 0.001, ns: not significant.* BugBase functional analysis predicted oxygen utilizing, gram staining, oxidative stress tolerance, biofilm forming, pathogenic potential, mobile element containing, and oxygen tolerance. Compared with healthy controls, anaerobic bacteria decreased in the AP-nonARDS and AP-ARDS groups (CONTROL vs. AP-nonARDS, CONTROL vs. AP-ARDS, both $p \leq 0.001$). Although there was no significant difference, anaerobic bacteria showed a decreasing trend in AP-ARDS compared with AP-nonARDS (Figure 4B). In contrast, aerobic bacteria increased in the AP-nonARDS and AP-ARDS groups (Figure 4C). Spearman correlation analysis was performed to investigate the relationship between microbiota and clinical outcomes. Two subspecies of Klebsiella pneumoniae, ASV_101 and ASV_71, were positively correlated with multiple clinical characteristics, including organ failure, bowel obstruction, sepsis, infection, and acute peripancreatic fluid collection. Prevotella copri (ASV_30) was positively correlated with the occurrence and duration of organ failure. Clostridium ramosum (ASV_150) was associated with ICU admission and length of hospital stay. As a probiotic, Bifidobacterium longum (ASV_14) negatively correlated with organ failure, Sequential Organ Failure Assessment score (SOFA score), and Acute Physiology And Chronic Health II score (APACHII score) (Figure 4D). ## The progression of AP-ARDS is closely associated with Enterobacteriaceae Considering the significant increase in Enterobacteriaceae and its potential pathogenicity, we performed further analyses of Enterobacteriaceae. Figure 5A shows the relative abundance of Enterobacteriaceae increased with disease progression. Further analysis revealed that almost all genera of the Enterobacteriaceae family were increased in the AP-ARDS group (Figure 5B). Random forest identified Escherichia-shigella as the most significant feature for distinguishing AP-ARDS from AP-nonARDS (Figure 5C). Figure 5: Enterobacteriaceae Analysis and Model Predicting. (A) The relative abundance of Enterobacteriaceae in Control, AP-nonARDS, and AP-ARDS group. (B) Major genus in Enterobacteriaceae family between AP-nonARDS, and AP-ARDS group. (C) Random forest model predicting. It screened out the Escherichia-shigella genus as the most significant feature for predicting ARDS. ## Discussion To our knowledge, this is the first study to explore the relationship between gut microbiota and AP-ARDS and reveals gut microbiota as a predictive biomarker for ARDS. The 16S rRNA sequencing analysis revealed differences of microbiota composition and function between the AP-ARDS and AP-nonARDS groups. Subgroup analysis suggested that gut microbiota composition was also related to the severity of ARDS. Before patients were diagnosed with AP-ARDS, the gut microbiota already had the characteristics of ARDS in AP patients. This indicates that the gut microbiota can be a potential biomarker for prediction and early recognition of AP-ARDS, thereby improving AP-ARDS diagnosis and treatment. In our characteristics analysis, AP patients with ARDS were more serious than that in the non-ARDS group. AP-severity-associated changes in the gut microbiota were also observed in the AP-ARDS group compared with the AP-nonARDS group. However, we also observed some microbiota changes that might be related to the occurrence and development of AP-ARDS. The enrichment of Enterobacteriaceae and Escherichia-Shigella, and the reduction of Bifidobacterium were associated with AP-ARDS. Previous studies have focused on the lung microbiota of ARDS patients and found that the composition was affected by the gut microbiota (Dickson et al., 2016; Dickson, 2018; Dickson et al., 2020). In our study, similar changes in composition were also observed in the lung microbiota. In the normal population, the most dominant phylum in the gut microbiota is Firmicutes, followed by Bacteroidetes, Actinobacteria, and Proteobacteria (Bozzi Cionci et al., 2018). In our study, the composition of healthy controls was consistent with the normal population; but in AP patients, Proteobacteria significantly increased with disease severity. Previous studies have found Proteobacteria overgrowth in patients with AP, particularly SAP (Zhu et al., 2019; Yu et al., 2020; Zhu et al., 2021). In addition, Proteobacteria in the lung microbiota are closely associated with inflammatory lung disease and positively related to alveolar TNF-α (Dickson et al., 2016). A higher abundance of Proteobacteria was a distinguishing feature of ventilator-associated pneumonia (Fromentin et al., 2021). The enrichment of Proteobacteria might be a biomarker of inflammatory status in patients. In our study, Enterobacteriaceae and Escherichia-Shigella were dominant in AP-ARDS patients. The overall levels of Enterobacteriaceae, and the individual Enterobacteriaceae genera, increased significantly in patients with ARDS. Escherichia-Shigella, a genus of Enterobacteriaceae family, is an opportunistic pathogen and more abundant in the sicker group. Random forest analysis identified Escherichia-Shigella as the most significant feature for distinguishing ARDS from non-ARDS. Multiple studies have shown that gut-associated bacteria in the lung microbiota, especially Enterobacteriaceae, are more abundant in ARDS patients. The abundance of Enterobacteriaceae in the lung microbiota was strongly associated with serum IL-6 level and the development of ARDS (Dickson et al., 2016; Dickson, 2018; Mukherjee and Hanidziar, 2018; Kyo et al., 2019). The composition of Enterobacteriaceae in the lung can help to identify ARDS patients (Dickson et al., 2020). The enrichment of gut-associated bacteria could also be a biomarker for ARDS patients (Fromentin et al., 2021). Suppression of the gut microbiota could improve the prognosis of critically ill patients (Silvestri et al., 2012). At the species level, potentially pathogenic bacteria, including Klebsiella pneumoniae, Prevotella copri, and Clostridium ramosum, increased significantly in AP-ARDS patients. Klebsiella pneumoniae, a common pathogen of the Enterobacteriaceae family, normally colonizes respiratory tract and gut (Chen et al., 2021; Wolff et al., 2021). Dickson et al. revealed that Klebsiella pneumoniae overgrowth in the lung was strongly associated with ARDS (Dickson et al., 2020). In addition, Klebsiella pneumoniae infection can influence both the gut microbiome and lung metabolome (Wu et al., 2020; Jiang et al., 2022). After inoculation of mice with Klebsiella pneumoniae, the diversity and composition of the gut microbiota changed and contributed to lung microbiota dysbiosis within several hours (Jiang et al., 2022). Therefore, Klebsiella pneumoniae in the gut may influence lung inflammation through bacterial translocation. Prevotella could increase the host sensitivity to intestinal inflammation (Iljazovic et al., 2021). Prevotella copri is the most well-known of the Prevotella genus and is positively correlated with many inflammatory diseases, such as rheumatoid arthritis and ankylosing spondylitis (Tett et al., 2021). Transplantation of Prevotella copri induces dysbiosis of inflammatory and immune functions and can induce arthritis in mice (Maeda et al., 2016; Qian et al., 2022). Although less well studied, Clostridium ramosum has been proven to be positively correlated with Covid-19 disease severity, as well as infection and bacteremia (Zuo et al., 2020). In our study, the levels of probiotics, such as Bifidobacterium and Bifidobacterium longum, decreased in patients with ARDS. Bifidobacterium longum was negatively correlated with organ failure and disease severity scores. As a beneficial bacterium, Bifidobacterium can help maintain gut barrier function, inhibit bacterial translocation, reduce lung inflammation, and therefore improve prognosis (Akshintala et al., 2019; Zhu et al., 2019; Zhu et al., 2021). In the current study, Bifidobacterium decreased in both AP patients and mice (Chen et al., 2017; Huang et al., 2017; Zhu et al., 2019; Li et al., 2020). Bifidobacterium longum can inhibit viral-induced lung inflammation and injury in mice (Groeger et al., 2020). Supplementation with Bifidobacterium longum has shown promising benefits for many diseases, such as irritable bowel syndrome, atopic dermatitis, and obesity (Schellekens et al., 2021; Fang et al., 2022; Sabaté and Iglicki, 2022). Therefore, probiotics have been used in the treatment of AP despite controversy. Previous studies found the enrichment of gut-associated bacteria in the lung is closely associated with ARDS. However, whether the changes in the gut and lung microbiota are consistent has never been studied. In our study, the variation in gut microbiota in ARDS patients is similar to those seen in lung microbiota in previous studies, which suggests that changes in the lung microbiota might be due to the translocation of the gut microbiota in ARDS patients. The gut-lung axis is a potential mechanism by which the gut microbiota influences lung inflammation. Gut microbiota can influence local immunity, systemic inflammation, and host immune suppression (Budden et al., 2017; Mukherjee and Hanidziar, 2018; Siwicka-Gieroba and Czarko-Wicha, 2020). Gut microbiota activate immune cells, which can migrate from the gut to the lung and assist in resisting systemic inflammatory disease (He et al., 2017; Mjösberg and Rao, 2018), and release metabolites and endotoxins to influence host immune response (Segain et al., 2000; Artis, 2008; Lin and Zhang, 2017). Additionally, gut microbiota dysbiosis damages the integrity of the intestinal barrier and enables bacterial translocation (Wang et al., 2022). Bacteria in the gut can translocate to the lung through the lymphatic or blood circulation systems and thus mediate lung inflammation (Mukherjee and Hanidziar, 2018). Enterobacteriaceae, Escherichia-Shigella, and several gut-associated bacteria have been detected in pancreatic fluid which suggested bacteria translocation could occur in AP patients and lead to infected pancreatic necrosis (Li et al., 2013; Hanna et al., 2014; Schmidt et al., 2014). Further studies have revealed that the composition of the lung microbiota can be easily changed, even if the immigration of gut-associated bacteria is transient (Dickson et al., 2016). Gut microbiota could be transferred to the lung by several possible mechanisms. First, intestinal mucosal permeability may be impaired owing to dysbiosis of the microbiota. In our functional analysis, there was a difference in anaerobic bacterial composition between the AP-ARDS and AP-nonARDS groups. Dysbiosis of anaerobic bacteria is correlated with intestinal epithelial integrity and promotes overgrowth of pathogenic bacteria (Hong et al., 2018; Zhou and Liao, 2021). The proliferation of pathogenic bacteria can consume fatty acids, change intestinal pH, inhibit the growth of probiotics, and damage the gut chemical barrier (Wang et al., 2022). In addition, the overgrowth of pathogens restricts the function of immune cells, such as Tregs, Th2, and B cells; promotes the production of inflammatory factors, such as IL-1β, IL-6, and TNF-α; and thus damages the gut immune barrier (Zhou and Liao, 2021; Wang et al., 2022). As a normal pathogen, Escherichis-Shigella is associated with epithelial cell injury and is strongly correlated with AP and ARDS severity (Zhu et al., 2019; Pan et al., 2021). Through reduced butyrate production and increased oxidative stress, Escherichis-Shigella could penetrate the intestinal barrier, reach the basolateral layer, and spread rapidly to adjacent cells (Fokam Tagne et al., 2018; Dong et al., 2020). A second possible mechanism may involve the lung microenvironment which is important for bacterial colonization; normally, the alveolar ecosystem is not appropriate for bacterial reproduction (Dickson et al., 2017), however, the lung barrier could be damaged in AP patients. Previous studies have shown that inflammatory factors can migrate from the gut to the lung, recruit neutrophils in the blood, and cause lung inflammation (Mjösberg and Rao, 2018; Zhou and Liao, 2021). The inflammatory cascade amplified the inflammatory response and provided a more favorable inflammatory lung microenvironment for bacterial colonization. In ARDS patients, the influx of nutrient-rich edema and establishment of stark oxygen gradients will damage the local host defenses of the lung and make it easier for bacteria to translocate from the gut to the lung (Dickson, 2018). Therefore, patients may be more sensitive to disruption of the gut microbiota. However, these hypotheses have not been fully confirmed. The translocation of bacteria from the gut to the lung has important clinical implications. Previous studies have proposed that the lung microbiota from bronchoalveolar lavage fluid (BAL) could help distinguish ARDS. However, rectal swabs are simpler to collect than BAL, therefore, gut microbiota data are easier to acquire than the lung; making gut microbiota a better prospect. In addition, our study found that the gut microbiota changed before the ARDS diagnosis. Prior to the occurrence of AP-ARDS, the gut microbiota already had characteristics relating to ARDS. Therefore, gut microbiota can be an important predictor of ARDS. Among them, Proteobacteria, Enterobacteriaceae, and Escherichia-Shigella were also found increased in lung microbiota in previous studies. These bacteria may help build prediction models for AP-ARDS that could assist clinicians in decision-making and prevent the occurrence and development of AP-ARDS. Considering the potential function of microbiota dysbiosis, restoring immune competence and disturbing microbiota is a promising therapy for AP-ARDS (Mukherjee and Hanidziar, 2018). However, the effects of probiotics on patients with AP remain controversial. Some trials have revealed that probiotic supplements may have no benefit in the clinical outcomes of AP patients (Isenmann et al., 2004; Mazaki et al., 2006; Dellinger et al., 2007; de Vries et al., 2007) and probiotic treatment may even worsen the prognosis of patients with AP. Probiotics could cause bacteremia despite the rarity and transfer of antibiotic resistance from probiotics to pathogenic bacteria may worsen infection (Salminen et al., 2002; Cannon et al., 2005; Connolly et al., 2005; Feld et al., 2008). Besselink et al. illustrated that probiotic supplementation increases the occurrence of organ failure and mortality in patients with SAP (Besselink et al., 2008). According to our study results, this poor response might be related to the overgrowth of pathogens and a disrupted intestinal mucosal barrier. For example, Klebsiella pneumoniae infection can inhibit Bifidobacterium production (Jiang et al., 2022) and the abundance of Prevotella is negatively associated with Bacteroides (Tett et al., 2021). Therefore, reducing pathogenic bacteria may promote the growth of probiotics, reduce barrier damage, and thus improve the efficacy of probiotic supplements. Targeted antibiotics are an effective strategy. Germ-free or antibiotic-treated animals are consistently protected from ARDS, and prophylactic administration of antibiotics decreases both mortality and multiple organ dysfunction syndromes, including ARDS (Dickson, 2016; Dickson, 2018). Supplementation with short-chain fatty acids (SCFAs) is another effective treatment option. Studies have found that oral supplementation with SCFAs could decrease susceptibility to bacterial infection, indicating that adjusting the gut microbiota could prevent bacterial pneumonia (Seki et al., 2021). These treatment concepts can be applied for AP patients to prevent ARDS (Siwicka-Gieroba and Czarko-Wicha, 2020). However, considering the potential harm caused by probiotics and antibiotics, targeted therapy should be provided to high-risk ARDS patients. Therefore, the prediction or early recognition of ARDS is essential. Collecting gut microbiota in the early stages of AP could help recognize and diagnose ARDS, and thus guide clinical management. Gut microbiota could help identify high risk population for developing ARDS. Early identification gives time for appropriate intervention which could help improve prognosis. However, our study has some limitations. First, the specific role of microbiota changes in the disease is unclear. Our study can only provide correlations and suggest that the microbiota might help predict ARDS. However, the mechanism by which microbiota causes pathological conditions remains unknown. Zhang et al. found that different initial sites of infection could influence lung microbiota in patients with septic ARDS. ARDS patients with initial intrapulmonary infection tend to have higher abundance of gut-associated in lung (Zhang et al., 2022). To determine the specific role of gut-lung axis, it is better to make a more nuanced classification in the future. Second, the 16S rRNA sequence analysis could not predict the real composition and function of the microbiota community because it is based on the 16S rRNA sequence library. 16S rRNA analysis cannot completely replace metagenomic analysis but can help guide further studies. Third, the detection time 16S rRNA is long now. For clinical application, quick PCR kit target to specific bacteria is still needed. In conclusion, this is the first study to report the relationship between gut microbiota and AP-ARDS. Gut microbiota showed a potential predicting ability for AP-ARDS. Dysbiosis of gut microbiota is strongly correlated with AP-ARDS. Enterobacteriaceae and Escherichia-Shigella are important prediction biomarkers for AP-ARDS. In the future, gut microbiota in early stage of patients with AP may help predict and allow early recognition of AP-ARDS, aid therapy planning, and thus improve patients’ quality of life and reduce morbidity of ARDS in AP patients. Further studies will improve our understanding of the role of microbiota in ARDS. ## 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/, PRJNA893348. ## Ethics statement The studies involving human participants were reviewed and approved by Ethics Committee of PUMCH (JS1826). The patients/participants provided their written informed consent to participate in this study. ## Author contributions XH and ZH conceived this study and drafted the manuscript. RZ and WS performed data analysis and reviewed the manuscript. ZH, RZ, WS, LG, ZY, and XS collected rectal swabs and clinical data. SZ revised the manuscript. HS and DW contributed to the study design, managed this study and revised 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. ## References 1. Akshintala V. S., Talukdar R., Singh V. K., Goggins M.. The gut microbiome in pancreatic disease. Clin. Gastroenterol. Hepatol. (2019) 17 290-295. DOI: 10.1016/j.cgh.2018.08.045 2. Artis D.. Epithelial-cell recognition of commensal bacteria and maintenance of immune homeostasis in the gut. Nat. Rev. Immunol. (2008) 8 411-420. DOI: 10.1038/nri2316 3. Banks P. A., Bollen T. L., Dervenis C., Gooszen H. G., Johnson C. D., Sarr M. G.. Classification of acute pancreatitis–2012: revision of the Atlanta classification and definitions by international consensus. Gut (2013) 62 102-111. DOI: 10.1136/gutjnl-2012-302779 4. Bellani G., Pham T., Laffey J. G.. Missed or delayed diagnosis of ARDS: a common and serious problem. Intensive Care Med. (2020) 46 1180-1183. DOI: 10.1007/s00134-020-06035-0 5. Besselink M. G., van Santvoort H. C., Boermeester M. A., Nieuwenhuijs V. B., van Goor H., Dejong C. H.. Timing and impact of infections in acute pancreatitis. Br. J. Surg. (2009) 96 267-273. DOI: 10.1002/bjs.6447 6. Besselink M. G., van Santvoort H. C., Buskens E., Boermeester M. A., van Goor H., Timmerman H. M.. Probiotic prophylaxis in predicted severe acute pancreatitis: A randomised, double-blind, placebo-controlled trial. Lancet (2008) 371 651-659. DOI: 10.1016/s0140-6736(08)60207-x 7. Bozzi Cionci N., Baffoni L., Gaggìa F., Di Gioia D.. Therapeutic microbiology: The role of bifidobacterium breve as food supplement for the Prevention/Treatment of paediatric diseases. Nutrients (2018) 10 1-27. DOI: 10.3390/nu10111723 8. Budden K. F., Gellatly S. L., Wood D. L., Cooper M. A., Morrison M., Hugenholtz P.. Emerging pathogenic links between microbiota and the gut-lung axis. Nat. Rev. Microbiol. (2017) 15 55-63. DOI: 10.1038/nrmicro.2016.142 9. Cannon J. P., Lee T. A., Bolanos J. T., Danziger L. H.. Pathogenic relevance of lactobacillus: A retrospective review of over 200 cases. Eur. J. Clin. Microbiol. Infect. Dis. (2005) 24 31-40. DOI: 10.1007/s10096-004-1253-y 10. Chen J., Huang C., Wang J., Zhou H., Lu Y., Lou L.. Dysbiosis of intestinal microbiota and decrease in paneth cell antimicrobial peptide level during acute necrotizing pancreatitis in rats. PloS One (2017) 12. DOI: 10.1371/journal.pone.0176583 11. Chen C. M., Wang M., Li X. P., Li P. L., Tian J. J., Zhang K.. Homology analysis between clinically isolated extraintestinal and enteral klebsiella pneumoniae among neonates. BMC Microbiol. (2021) 21 25. DOI: 10.1186/s12866-020-02073-2 12. Connolly E., Abrahamsson T., Björkstén B.. Safety of d(-)-lactic acid producing bacteria in the human infant. J. Pediatr. Gastroenterol. Nutr. (2005) 41 489-492. DOI: 10.1097/01.mpg.0000176179.81638.45 13. Dellinger E. P., Tellado J. M., Soto N. E., Ashley S. W., Barie P. S., Dugernier T.. Early antibiotic treatment for severe acute necrotizing pancreatitis: a randomized, double-blind, placebo-controlled study. Ann. Surg. (2007) 245 674-683. DOI: 10.1097/01.sla.0000250414.09255.84 14. de Vries A. C., Besselink M. G., Buskens E., Ridwan B. U., Schipper M., van Erpecum K. J.. Randomized controlled trials of antibiotic prophylaxis in severe acute pancreatitis: relationship between methodological quality and outcome. Pancreatology (2007) 7 531-538. DOI: 10.1159/000108971 15. Dickson R. P.. The microbiome and critical illness. Lancet Respir. Med. (2016) 4 59-72. DOI: 10.1016/s2213-2600(15)00427-0 16. Dickson R. P.. The lung microbiome and ARDS. it is time to broaden the model. Am. J. Respir. Crit. Care Med. (2018) 197 549-551. DOI: 10.1164/rccm.201710-2096ED 17. Dickson R. P., Erb-Downward J. R., Freeman C. M., McCloskey L., Falkowski N. R., Huffnagle G. B.. Bacterial topography of the healthy human lower respiratory tract. mBio (2017) 8 1-12. DOI: 10.1128/mBio.02287-16 18. Dickson R. P., Schultz M. J., van der Poll T., Schouten L. R., Falkowski N. R., Luth J. E.. Lung microbiota predict clinical outcomes in critically ill patients. Am. J. Respir. Crit. Care Med. (2020) 201 555-563. DOI: 10.1164/rccm.201907-1487OC 19. Dickson R. P., Singer B. H., Newstead M. W., Falkowski N. R., Erb-Downward J. R., Standiford T. J.. Enrichment of the lung microbiome with gut bacteria in sepsis and the acute respiratory distress syndrome. Nat. Microbiol. (2016) 1 16113. DOI: 10.1038/nmicrobiol.2016.113 20. Dong R., Bai M., Zhao J., Wang D., Ning X., Sun S.. A comparative study of the gut microbiota associated with immunoglobulin a nephropathy and membranous nephropathy. Front. Cell Infect. Microbiol. (2020) 10. DOI: 10.3389/fcimb.2020.557368 21. Fan E., Brodie D., Slutsky A. S.. Acute respiratory distress syndrome: Advances in diagnosis and treatment. Jama (2018) 319 698-710. DOI: 10.1001/jama.2017.21907 22. Fang Z., Pan T., Li L., Wang H., Zhu J., Zhang H.. Bifidobacterium longum mediated tryptophan metabolism to improve atopic dermatitis. Gut Microbes (2022) 14. DOI: 10.1080/19490976.2022.2044723 23. Feld L., Schjørring S., Hammer K., Licht T. R., Danielsen M., Krogfelt K.. Selective pressure affects transfer and establishment of a lactobacillus plantarum resistance plasmid in the gastrointestinal environment. J. Antimicrob. Chemother. (2008) 61 845-852. DOI: 10.1093/jac/dkn033 24. Fokam Tagne M. A., Noubissi P. A., Fankem G. O., Kamgang R.. Effects of oxalis barrelieri l. (Oxalidaceae) aqueous extract on diarrhea induced by shigella dysenteriae type 1 in rats. Health Sci. Rep. (2018) 1. DOI: 10.1002/hsr2.20 25. Fromentin M., Ricard J. D., Roux D.. Respiratory microbiome in mechanically ventilated patients: A narrative review. Intensive Care Med. (2021) 47 292-306. DOI: 10.1007/s00134-020-06338-2 26. Garg P. K., Madan K., Pande G. K., Khanna S., Sathyanarayan G., Bohidar N. P.. Association of extent and infection of pancreatic necrosis with organ failure and death in acute necrotizing pancreatitis. Clin. Gastroenterol. Hepatol. (2005) 3 159-166. DOI: 10.1016/s1542-3565(04)00665-2 27. Gray K. D., Simovic M. O., Chapman W. C., Blackwell T. S., Christman J. W., May A. K.. Endotoxin potentiates lung injury in cerulein-induced pancreatitis. Am. J. Surg. (2003) 186 526-530. DOI: 10.1016/j.amjsurg.2003.07.010 28. Greenberg J. A., Hsu J., Bawazeer M., Marshall J., Friedrich J. O., Nathens A.. Clinical practice guideline: Management of acute pancreatitis. Can. J. Surg. (2016) 59 128-140. DOI: 10.1503/cjs.015015 29. Groeger D., Schiavi E., Grant R., Kurnik-Łucka M., Michalovich D., Williamson R.. Intranasal bifidobacterium longum protects against viral-induced lung inflammation and injury in a murine model of lethal influenza infection. EBioMedicine (2020) 60. DOI: 10.1016/j.ebiom.2020.102981 30. Hanna E. M., Hamp T. J., McKillop I. H., Bahrani-Mougeot F., Martinie J. B., Horton J. M.. Comparison of culture and molecular techniques for microbial community characterization in infected necrotizing pancreatitis. J. Surg. Res. (2014) 191 362-369. DOI: 10.1016/j.jss.2014.05.003 31. He Y., Wen Q., Yao F., Xu D., Huang Y., Wang J.. Gut-lung axis: The microbial contributions and clinical implications. Crit. Rev. Microbiol. (2017) 43 81-95. DOI: 10.1080/1040841x.2016.1176988 32. Herridge M. S., Tansey C. M., Matté A., Tomlinson G., Diaz-Granados N., Cooper A.. Functional disability 5 years after acute respiratory distress syndrome. N Engl. J. Med. (2011) 364 1293-1304. DOI: 10.1056/NEJMoa1011802 33. Hong G., Zheng D., Zhang L., Ni R., Wang G., Fan G. C.. Administration of nicotinamide riboside prevents oxidative stress and organ injury in sepsis. Free Radic. Biol. Med. (2018) 123 125-137. DOI: 10.1016/j.freeradbiomed.2018.05.073 34. Hu X., Fan Y., Li H., Zhou R., Zhao X., Sun Y.. Impacts of cigarette smoking status on metabolomic and gut microbiota profile in Male patients with coronary artery disease: A multi-omics study. Front. Cardiovasc. Med. (2021) 8. DOI: 10.3389/fcvm.2021.766739 35. Hu X., Gong L., Zhou R., Han Z., Ji L., Zhang Y.. Variations in gut microbiome are associated with prognosis of hypertriglyceridemia-associated acute pancreatitis. Biomolecules (2021) 11 1-16. DOI: 10.3390/biom11050695 36. Huang C., Chen J., Wang J., Zhou H., Lu Y., Lou L.. Dysbiosis of intestinal microbiota and decreased antimicrobial peptide level in paneth cells during hypertriglyceridemia-related acute necrotizing pancreatitis in rats. Front. Microbiol. (2017) 8. DOI: 10.3389/fmicb.2017.00776 37. Iljazovic A., Roy U., Gálvez E. J. C., Lesker T. R., Zhao B., Gronow A.. Perturbation of the gut microbiome by prevotella spp. enhances host susceptibility to mucosal inflammation. Mucosal Immunol. (2021) 14 113-124. DOI: 10.1038/s41385-020-0296-4 38. Isenmann R., Rünzi M., Kron M., Kahl S., Kraus D., Jung N.. Prophylactic antibiotic treatment in patients with predicted severe acute pancreatitis: a placebo-controlled, double-blind trial. Gastroenterology (2004) 126 997-1004. DOI: 10.1053/j.gastro.2003.12.050 39. Jiang Q., Xu Q., Kenéz Á., Chen S., Yang G.. Klebsiella pneumoniae infection is associated with alterations in the gut microbiome and lung metabolome. Microbiol. Res. (2022) 263. DOI: 10.1016/j.micres.2022.127139 40. Johnson C. D., Abu-Hilal M.. Persistent organ failure during the first week as a marker of fatal outcome in acute pancreatitis. Gut (2004) 53 1340-1344. DOI: 10.1136/gut.2004.039883 41. Kyo M., Nishioka K., Nakaya T., Kida Y., Tanabe Y., Ohshimo S.. Unique patterns of lower respiratory tract microbiota are associated with inflammation and hospital mortality in acute respiratory distress syndrome. Respir. Res. (2019) 20 246. DOI: 10.1186/s12931-019-1203-y 42. Li X. Y., He C., Zhu Y., Lu N. H.. Role of gut microbiota on intestinal barrier function in acute pancreatitis. World J. Gastroenterol. (2020) 26 2187-2193. DOI: 10.3748/wjg.v26.i18.2187 43. Li Q., Wang C., Tang C., He Q., Li N., Li J.. Bacteremia in patients with acute pancreatitis as revealed by 16S ribosomal RNA gene-based techniques. Crit. Care Med.* (2013) 41 1938-1950. DOI: 10.1097/CCM.0b013e31828a3dba 44. Lin L., Zhang J.. Role of intestinal microbiota and metabolites on gut homeostasis and human diseases. BMC Immunol. (2017) 18. DOI: 10.1186/s12865-016-0187-3 45. Maeda Y., Kurakawa T., Umemoto E., Motooka D., Ito Y., Gotoh K.. Dysbiosis contributes to arthritis development. Arthritis Rheumatol (2016) 68 2646-2661. DOI: 10.1002/art.39783 46. Mazaki T., Ishii Y., Takayama T.. Meta-analysis of prophylactic antibiotic use in acute necrotizing pancreatitis. Br. J. Surg. (2006) 93 674-684. DOI: 10.1002/bjs.5389 47. Mjösberg J., Rao A.. Lung inflammation originating in the gut. Science (2018) 359 36-37. DOI: 10.1126/science.aar4301 48. Mukherjee S., Hanidziar D.. More of the gut in the lung: How two microbiomes meet in ARDS. Yale J. Biol. Med. (2018) 91 143-149. PMID: 29955219 49. Pan L. L., Li B. B., Pan X. H., Sun J.. Gut microbiota in pancreatic diseases: possible new therapeutic strategies. Acta Pharmacol. Sin. (2021) 42 1027-1039. DOI: 10.1038/s41401-020-00532-0 50. Pan C., Liu L., Xie J. F., Qiu H. B.. Acute respiratory distress syndrome: Challenge for diagnosis and therapy. Chin. Med. J. (Engl) (2018) 131 1220-1224. DOI: 10.4103/0366-6999.228765 51. Qian X., Zhang H. Y., Li Q. L., Ma G. J., Chen Z., Ji X. M.. Integrated microbiome, metabolome, and proteome analysis identifies a novel interplay among commensal bacteria, metabolites and candidate targets in non-small cell lung cancer. Clin. Transl. Med. (2022) 12. DOI: 10.1002/ctm2.947 52. Ranieri V. M., Rubenfeld G. D., Thompson B. T., Ferguson N. D., Caldwell E., Fan E.. Acute respiratory distress syndrome: The Berlin definition. Jama (2012) 307 2526-2533. DOI: 10.1001/jama.2012.5669 53. Sabaté J. M., Iglicki F.. Effect of bifidobacterium longum 35624 on disease severity and quality of life in patients with irritable bowel syndrome. World J. Gastroenterol. (2022) 28 732-744. DOI: 10.3748/wjg.v28.i7.732 54. Salminen M. K., Tynkkynen S., Rautelin H., Saxelin M., Vaara M., Ruutu P.. Lactobacillus bacteremia during a rapid increase in probiotic use of lactobacillus rhamnosus GG in Finland. Clin. Infect. Dis. (2002) 35 1155-1160. DOI: 10.1086/342912 55. Schellekens H., Torres-Fuentes C., van de Wouw M., Long-Smith C. M., Mitchell A., Strain C.. Bifidobacterium longum counters the effects of obesity: Partial successful translation from rodent to human. EBioMedicine (2021) 63. DOI: 10.1016/j.ebiom.2020.103176 56. Schmandt M., Glowka T. R., Kreyer S., Muders T., Muenster S., Theuerkauf N. U.. Secondary ARDS following acute pancreatitis: Is extracorporeal membrane oxygenation feasible or futile. J. Clin. Med. (2021) 10 1-11. DOI: 10.3390/jcm10051000 57. Schmidt P. N., Roug S., Hansen E. F., Knudsen J. D., Novovic S.. Spectrum of microorganisms in infected walled-off pancreatic necrosis - impact on organ failure and mortality. Pancreatology (2014) 14 444-449. DOI: 10.1016/j.pan.2014.09.001 58. Segain J. P., Raingeard de la Blétière D., Bourreille A., Leray V., Gervois N., Rosales C.. Butyrate inhibits inflammatory responses through NFkappaB inhibition: implications for crohn's disease. Gut (2000) 47 397-403. DOI: 10.1136/gut.47.3.397 59. Seki D., Mayer M., Hausmann B., Pjevac P., Giordano V., Goeral K.. Aberrant gut-microbiota-immune-brain axis development in premature neonates with brain damage. Cell Host Microbe (2021) 29 1558-1572.e1556. DOI: 10.1016/j.chom.2021.08.004 60. Shah J., Rana S. S.. Acute respiratory distress syndrome in acute pancreatitis. Indian J. Gastroenterol. (2020) 39 123-132. DOI: 10.1007/s12664-020-01016-z 61. Silvestri L., de la Cal M. A., van Saene H. K.. Selective decontamination of the digestive tract: the mechanism of action is control of gut overgrowth. Intensive Care Med. (2012) 38 1738-1750. DOI: 10.1007/s00134-012-2690-1 62. Siwicka-Gieroba D., Czarko-Wicha K.. Lung microbiome - a modern knowledge. Cent Eur. J. Immunol. (2020) 45 342-345. DOI: 10.5114/ceji.2020.101266 63. Tenner S., Baillie J., DeWitt J., Vege S. S.. American College of gastroenterology guideline: management of acute pancreatitis. Am. J. Gastroenterol. (2013) 108 1400-1415; 1416. DOI: 10.1038/ajg.2013.218 64. Tett A., Pasolli E., Masetti G., Ercolini D., Segata N.. Prevotella diversity, niches and interactions with the human host. Nat. Rev. Microbiol. (2021) 19 585-599. DOI: 10.1038/s41579-021-00559-y 65. Wang Z., Li F., Liu J., Luo Y., Guo H., Yang Q.. Intestinal microbiota - an unmissable bridge to severe acute pancreatitis-associated acute lung injury. Front. Immunol. (2022) 13. DOI: 10.3389/fimmu.2022.913178 66. Wolff N. S., Jacobs M. C., Wiersinga W. J., Hugenholtz F.. Pulmonary and intestinal microbiota dynamics during gram-negative pneumonia-derived sepsis. Intensive Care Med. Exp. (2021) 9. DOI: 10.1186/s40635-021-00398-4 67. Wu T., Xu F., Su C., Li H., Lv N., Liu Y.. Alterations in the gut microbiome and cecal metabolome during klebsiella pneumoniae-induced pneumosepsis. Front. Immunol. (2020) 11. DOI: 10.3389/fimmu.2020.01331 68. Yu S., Xiong Y., Fu Y., Chen G., Zhu H., Mo X.. Shotgun metagenomics reveals significant gut microbiome features in different grades of acute pancreatitis. Microb. Pathog. (2021) 154. DOI: 10.1016/j.micpath.2021.104849 69. Yu S., Xiong Y., Xu J., Liang X., Fu Y., Liu D.. Identification of dysfunctional gut microbiota through rectal swab in patients with different severity of acute pancreatitis. Dig Dis. Sci. (2020) 65 3223-3237. DOI: 10.1007/s10620-020-06061-4 70. Zhang P., Chen Y., Zheng W., Wu M., Wu Z., Lu Y.. Changes of lung microbiome of acute respiratory distress syndrome before and after treatment under open airway. Zhonghua Wei Zhong Bing Ji Jiu Yi Xue (2021) 33 1063-1068. DOI: 10.3760/cma.j.cn121430-20210414-00558 71. Zhang P., Liu B., Zheng W., Chen Y., Wu Z., Lu Y.. Pulmonary microbial composition in sepsis-induced acute respiratory distress syndrome. Front. Mol. Biosci. (2022) 9. DOI: 10.3389/fmolb.2022.862570 72. Zhang X. M., Zhang Z. Y., Zhang C. H., Wu J., Wang Y. X., Zhang G. X.. Intestinal microbial community differs between acute pancreatitis patients and healthy volunteers. BioMed. Environ. Sci. (2018) 31 81-86. DOI: 10.3967/bes2018.010 73. Zheng J., Lou L., Fan J., Huang C., Mei Q., Wu J.. Commensal escherichia coli aggravates acute necrotizing pancreatitis through targeting of intestinal epithelial cells. Appl. Environ. Microbiol. (2019) 85 1-15. DOI: 10.1128/aem.00059-19 74. Zhou X., Liao Y.. Gut-lung crosstalk in sepsis-induced acute lung injury. Front. Microbiol. (2021) 12. DOI: 10.3389/fmicb.2021.779620 75. Zhu Y., He C., Li X., Cai Y., Hu J., Liao Y.. Gut microbiota dysbiosis worsens the severity of acute pancreatitis in patients and mice. J. Gastroenterol. (2019) 54 347-358. DOI: 10.1007/s00535-018-1529-0 76. Zhu Y., Mei Q., Fu Y., Zeng Y.. Alteration of gut microbiota in acute pancreatitis and associated therapeutic strategies. BioMed. Pharmacother. (2021) 141. DOI: 10.1016/j.biopha.2021.111850 77. Zuo T., Zhang F., Lui G. C. Y., Yeoh Y. K., Li A. Y. L., Zhan H.. Alterations in gut microbiota of patients with COVID-19 during time of hospitalization. Gastroenterology (2020) 159 944-955.e948. DOI: 10.1053/j.gastro.2020.05.048
--- title: 'Questionnaires based on natural language processing elicit immersive ruminative thinking in ruminators: Evidence from behavioral responses and EEG data' authors: - Yulong Li - Chenxi Li - Tian Zhang - Lin Wu - Xinxin Lin - Yijun Li - Lingling Wang - Huilin Yang - Diyan Lu - Danmin Miao - Peng Fang journal: Frontiers in Neuroscience year: 2023 pmcid: PMC10025410 doi: 10.3389/fnins.2023.1118650 license: CC BY 4.0 --- # Questionnaires based on natural language processing elicit immersive ruminative thinking in ruminators: Evidence from behavioral responses and EEG data ## Abstract Rumination is closely related to mental disorders and can thus be used as a marker of their presence or a predictor of their development. The presence of masking and fabrication in psychological selection can lead to inaccurate detection of psychological disorders. Human language is considered crucial in eliciting specific conscious activities, and the use of natural language processing (NLP) in the development of questionnaires for psychological tests has the potential to elicit immersive ruminative thinking, leading to changes in neural activity. Electroencephalography (EEG) is commonly used to detect and record neural activity in the human brain and is sensitive to changes in brain activity. In this study, we used NLP to develop a questionnaire to induce ruminative thinking and then recorded the EEG signals in response to the questionnaire. The behavioral results revealed that ruminators exhibited higher arousal rates and longer reaction times, specifically in response to the ruminative items of the questionnaire. The EEG results showed no significant difference between the ruminators and the control group during the resting state; however, a significant alteration in the coherence of the entire brain of the ruminators existed while they were answering the ruminative items. No differences were found in the control participants while answering the two items. These behavioral and EEG results indicate that the questionnaire elicited immersive ruminative thinking, specifically in the ruminators. Therefore, the questionnaire designed using NLP is capable of eliciting ruminative thinking in ruminators, offering a promising approach for the early detection of mental disorders in psychological selection. ## Introduction Rumination is defined as continuous attention to negative stimuli, including the causes and consequences of negative events and the resulting negative emotions, which may lead to depression, anxiety, and other mental disorders. This phenomenon is considered a non-adaptive way of regulating emotions (Nolen-Hoeksema, 1991; Nolen-Hoeksema and Morrow, 1991; Nolen-Hoeksema et al., 1995). In the Self-Regulatory Executive Function (S-REF) model of emotional disorders, Wells and Matthews identified a persistent thinking mode that included worry and rumination and further defined it as an ineffective coping strategy. Rumination involves persistently focusing on negative thinking and feelings rather than problem solving, which is counterproductive by way of amplifying and prolonging the experience of suffering (Wells and Matthews, 1996). The relationship between rumination and depression has been widely studied (Castanheira et al., 2019; Li et al., 2020; Van Doorn et al., 2021). According to response styles theory, rumination can prolong and intensify the pain of negative or stressful events, increase despair, and aggravate depressive symptoms (Nolen-Hoeksema et al., 2008). Rumination is also associated with alcohol abuse, anxiety symptoms, generalized anxiety disorder, social anxiety disorder, obsessive-compulsive disorder, PTSD, schizophrenia, borderline personality disorder, and bulimia nervosa, among others, and other mental disorders (Watkins and Roberts, 2020). Although there are numerous scales available for assessing rumination, the most widely used remains the Ruminative Response Scale (RRS), developed by Nolen-Hoeksema in 1991 (Nolen-Hoeksema and Morrow, 1991) and compiled by Nolen Hoeksema in 1991. While RRS has been widely employed in ruminative research, the limitations of this scale are also significant, especially regarding the presence of psychological selection (Wang et al., 2020; Miao et al., 2021). Camouflage, falsification, social approval, and other issues (Dunning et al., 2005; Schwarz, 2012; Wang et al., 2020, 2021) have greatly affected the acceptance and validity of rumination-related psychological testing. Therefore, further advancements are necessary in inducing ruminative conscious activity accurately, measuring the associated brain activity objectively, and exploring the neural biomarkers of rumination to identify ruminators and predict future rumination for precise psychological selection. Previous studies have found that rumination arises as a result of negative situational memories (Sutherland and Bryant, 2007). Hence, the resurfacing of specific situational memories is considered one of the most effective and rapid means to induce rumination. In episodic memory-related studies, Wilson Mendenhall proposed “scenario immersion” and defined it as precise language that could facilitate immersive psychological imagination and enable subjects to place themselves into various imagined situations and memories (Wilson-Mendenhall et al., 2019). After experiencing different emotions, subjects form an episodic memory in their long-term memory, thus affecting the emotions they will experience in similar situations in the future. For example, when someone perceives that a car is approaching them quickly in a given scenario, their episodic memory from a previous similar scenario is activated in response. This implicitly, rapidly, and synergistically generates fearful cognitive, interoceptive, and behavioral processes in relation to the current situation (Lebois et al., 2020). Research has shown that rumination can be activated quickly when encountering situations that are similar to a past fearful event or when encountering only a small component of the original event (Wilson-Mendenhall et al., 2015). In 2022, Priyamvada Rajasethupathy confirmed the core idea of the “scenario immersion” theory with the finding that a holistic episodic memory composed of multiple sensory experiences can indeed be evoked by a single sensory cue (Yadav et al., 2022). Therefore, eliciting a specific episodic memory can be the most effective and efficient means to initiate ruminative thinking (Sutherland and Bryant, 2007). Natural language (i.e., the language used in daily life) is an important means of human communication and an essential feature that distinguishes human beings from other animals (Assale et al., 2019). A specific questionnaire based on natural language can be used to comprehensively induce the recall of the complete episodic memory (Wilson-Mendenhall et al., 2019; Ancin-Murguzur and Hausner, 2021). At present, natural language processing (NLP) technology has accomplished a series of complex functions, such as machine translation, automatic summarization, emotion analysis, and text classification (Castanheira et al., 2019). Compiling questionnaires based on natural language corpuses and NLP of rumination are the two methods that are considered effective in inducing and analyzing rumination accurately (Ferrario et al., 2020). The theory of “scenario immersion” is a useful tool for inducing rumination; therefore, we drew upon this theory and further incorporated NLP to develop a questionnaire that could elicit ruminative thinking both efficiently and effectively. Compared to cognitive neural technologies, such as functional magnetic resonance imaging (fMRI), magnetoencephalography (MEG), and functional near-infrared spectroscopy (fNIRS), wireless EEG technology is cheap, easy to transport, convenient for recording brain activity during experiments, sensitive to the alternation of brain activity, flexible across various experimental paradigms, and provides high temporal resolution signals, thereby rendering it suitable for language evaluation and the detection of brain activity during psychological selection (Deshpande et al., 2017). EEG coherence, first proposed by Robinson (Robinson, 2003), is an index of brain connectivity that is calculated by the covariance of the power spectral density at two electrodes. Coherence shows the synchronicity of neural activity and reflects brain dynamics (Markovska-Simoska et al., 2018). Neuroimaging studies have provided a great deal of data on the dysfunction and dysregulation that occurs in the brains of clinically and nonclinically depressed ruminators, including hypofunctional and hyperfunctional connectivity (Ferdek et al., 2016; Li et al., 2018; Benschop et al., 2020). Such studies were designed to examine network characteristics while ruminators were at rest or engaged in a task, thereby exposing the synchronicity of neural activity either between the brain areas (Ferdek et al., 2016) or between the brain networks (Zhang et al., 2020). Both EEG and fMRI studies have shown that altered functional connectivity in the brain networks of subjects with depression was positively related to ruminative thoughts (Benschop et al., 2020). These results indicated that the altered synchronicity within the brain might be an indicator of altered brain activity. EEG coherence may be a suitable indicator to depict the change of the brain activities in subjects with mental disorders. In this study, we developed a situational ruminative questionnaire using the natural language characteristics of ruminators to demonstrate the ability of the questionnaire to elicit immersive ruminative thinking in ruminators, as determined by their behavioral responses and characteristics of EEG signals. ## The ruminative response scale The Ruminative Response Scale (RRS) is a questionnaire composed of 22 items that is used to measure the ruminative tendency of an individual. Each item uses a 4-point Likert scale (1 = rarely, 2 = sometimes, 3 = often, 4 = almost always). The Chinese version used in this study was translated by Han Xiu and has been verified to have good reliability and validity among Chinese senior high school and college students (Han and Yang, 2009). The higher the total score, the higher the rumination level, and the highest score is 88. The demographic information of the participants is shown in Table 1. Table 1 \| Unnamed: 0 \| Frequency/mean \| Percentage/standard deviation \| \| --- \| --- \| --- \| \| Gender ( n , %) \| Gender ( n , %) \| Gender ( n , %) \| \| Women \| 65 \| 1.42 \| \| Men \| 4526 \| 98.58 \| \| Age (mean ±SD) \| 21.29 \| 2.59 \| \| Home location ( n , %) \| Home location ( n , %) \| Home location ( n , %) \| \| Urban \| 1191 \| 25.94 \| \| Village \| 3400 \| 74.06 \| \| Ethnicity ( n , %) \| Ethnicity ( n , %) \| Ethnicity ( n , %) \| \| Han \| 4109 \| 89.50 \| \| Others \| 482 \| 10.50 \| ## Natural language processing and questionnaire development We selected 607 subjects with a high tendency for rumination (577 men and 30 women) for semi-structured interviews that were conducted on an individual basis. The interviews were recorded and saved, and an intelligent conference system (version 5.0) was used to transcript the interview information for each interviewee. The demographic information of the participants is shown in Table 2. The obtained transcripts were analyzed, and the elicited material was compiled using the following six steps. [ 1] Proofreading and denoising: The text was thoroughly analyzed to check for errors and delete any extraneous characters, spaces, and so on. [ 2] Chinese participle: Chinese word segmentation was performed using Chinese LIWC software. [ 3] Stop word filtering: More uniform word segmentation text was created using the HIT edition of the “Stop Word Dictionary” to filter stop words. [ 4] Feature word extraction and coding: The TF-IDF approach was used to determine the frequency of feature words in the text (Reviewer-Lee, 2000; Jurafsky, 2009). The feature words were specifically selected by using the TF-IDF algorithm to calculate the frequency of all the words in the interview transcript of the extreme ruminators to obtain the high-frequency feature words first and then by referring to the 66 “scenario materials” words in Mendenhall's article (Wilson-Mendenhall et al., 2019). Only the content words (to construct the scenario) and the depictive words (to express the emotional experience) of the feature words were retained, and function words with no real meaning were removed. The bag-of-words model used a real-valued vector to label each transcribed word and encoded it using feature words (Ferrario et al., 2020). [ 5] Based on the semantic characteristics of the feature words, the LDA (latent Dirichlet allocation) topic model was used to extract text topics (Ghosh and Guha, 2013; Min et al., 2019, 2020), which were then used to group words under the topics. The detailed analysis pipeline of the LDA model and the document generation process are shown in Supplementary Figure 1 (Blei et al., 2003; Hao et al., 2017). Quantitative analysis was then used to identify the overarching topic of the interview material, while artificial naming was used to identify the situational topics. An example of extracting text topics is shown in Supplementary Table 1. [ 6] Creation of situational inducing materials: We recreated real-life scenarios that the interviewees had described, merged them with the six syntaxes of Mendenhall's “scenario immersion” theory, and then generated materials that elicited rumination. The neutral items were derived from paragraphs that were cut from the third edition of the Encyclopedia of China, whereby the number of words was kept similar to that of the rumination items and the chosen content was relatively boring and meaningless. Table 2 \| Unnamed: 0 \| Frequency/mean \| Percentage/standard deviation \| \| --- \| --- \| --- \| \| Gender ( n , %) \| Gender ( n , %) \| Gender ( n , %) \| \| Women \| 30 \| 4.94 \| \| Men \| 577 \| 95.06 \| \| Age (mean ±SD) \| 21.80 \| 3.03 \| \| Home location ( n , %) \| Home location ( n , %) \| Home location ( n , %) \| \| Urban \| 196 \| 32.29 \| \| Village \| 411 \| 67.71 \| \| Ethnicity ( n , %) \| Ethnicity ( n , %) \| Ethnicity ( n , %) \| \| Han \| 525 \| 86.49 \| \| Others \| 82 \| 13.51 \| ## Questionnaire evaluation The ruminative questionnaire was revised by three linguistics professors. The revisions made included connotation logic, grammar application, and character norms, among others. To verify the validity of the questionnaire regarding its ability to induce rumination, it was evaluated again by high- and low-degree ruminators. We then recruited another 1,685 subjects (all males) to complete the RRS. Of these, 78 participants were randomly selected for the final evaluation of the developed questionnaire. These participants comprised 40 high-degree ruminators (ruminators, mean age = 23.30 years, SD = 3.40 years; RRS score: 53.65 ± 8.89) and 38 low-degree ruminators (controls, mean age = 20.84 years, SD = 1.55 years; RRS score: 22.13 ± 0.34). We based the degree of rumination on the cutoff values specified in the Rosenbaum et al. ( 2018b) article: high-degree ruminators were defined as having a mean RRS score higher than 2.36 (PR > 65), while low-degree ruminators were defined as having an RRS score lower than 1.9 (PR < 27) (Rosenbaum et al., 2018b). Neutral items were also evaluated in comparison with the rumination items. The evaluation included seven dimensions: [1] repetition; [2] persistence; [3] associativity; [4] vividness; [5] uncontrolled nature; [6] assumption; [7] representativeness. The first six dimensions represent ruminative characteristics summarized from both our literature review and from interviews conducted with a large number of highly ruminative individuals. The last dimension, representativeness, reflects the extent to which an entry matched the subject's recall. These seven dimensions combine to represent the level of “scenario immersion” experienced by subjects while they were filling out the questionnaire. The demographic information of the 1,685 and 78 participants is shown in Tables 3, 4, respectively. ## EEG experiment participants We recruited subjects from the 1,685 freshmen enrolled in the 2022 cohort of Shaanxi Police College to participate in the EEG experiment. Finally, 56 voluntary participants (mean age = 22.48 years, SD = 7.73 years) selected from the high-degree ruminators (PR > 65, RRS score: 63.79 ± 6.38) were recruited as the rumination group, while 29 voluntary participants (mean age = 20.59 years, SD = 1.64 years) (PR < 27, RRS score: 22.14 ± 0.44) were recruited as the control group. The exclusion criteria were as follows: [1] patients with a history of psychiatry; [2] patients who had been hospitalized in the psychiatric department; [3] patients with current or past use of antipsychotic drugs; [4] patients with a history of neurological disorders; [5] patients who were left-handed. The study was approved by the Ethics Committee of the Air Force Medical University and was conducted in accordance with the approved guidelines (Ethics Approval Number KY20193304-1). All participants provided their informed consent prior to undergoing the formal experiment and received a small amount of compensation. ## Resting-state EEG data acquisition Resting-state EEG data were collected for all subjects before the test. First, the subjects were instructed to sit down in front of a screen in a comfortable position, with their eyes distanced ~70 cm from the stimulating screen. When the subjects were ready, the experimenter instructed them to close their eyes for more than 1 min and then to open them for more than 1 min. ## Task EEG data acquisition under material stimulation After recording resting-state EEG data, the subjects were instructed to complete the exam using the 53-item questionnaire (containing 37 rumination items and 16 neutral items). Items were displayed on the screen randomly, with a fixed 500-ms plus sign separating each stimulus from the next. All items presented the same following question: “Does the above description induce you to have repeated/continuous recall?”. The subjects selected “yes” or “no” as their response to this question using the mouse or keyboard. There were no time constraint for how long subjects were allowed to respond to each question. The item inquiry was completed as soon as the subject clicked the mouse to answer the questions. ## EEG recording and data preprocessing A 32-channel semi-dry electrode cap was used to record EEG data using a wireless multi-channel EEG acquisition device (ZhenTec NT1, ZhenTec Intelligence, China) (Yuan et al., 2021; Han et al., 2022). The sampling rate was 500 Hz. Data were referenced to CPz with a ground at FPz, and electrode placement followed the International 10-10 system. Impedance levels were set at < 20 kΩ. The common mode rejection ratio was 120 dB, the input impedance was 1 G, and the input noise was < 0.4 uVrms for all EEG channels. For each subject, more than 2 min of resting-state EEG signals were recorded under two conditions: with eyes closed and with eyes open, with each condition lasting more than 1 min. Additionally, while participants completed the test items, EEG signals were acquired. All of the EEGs were pre-processed prior to the commencement of further research using the FildTrip (Version 20221122) (Oostenveld et al., 2011) toolbox implemented in MATLAB 2018b. We first checked the quality of the data and processed the band channel using the interpolation method. Both the resting state and task signals were notched by 50 Hz to remove power-line interference. Then, the EEGs were band filtered with a 1–100 Hz zero-phase band filter. Subsequently, the signals were divided into two 1-min epochs for the eyes-closed and eyes-open conditions for resting-state EEGs. We segmented the task EEGs into each item-related signal epoch based on the time markers of eye movements where signals were acquired simultaneously (53 items and 53 task signals for each participant in each channel). We manually checked and removed any data with large interference. The data were considered invalid if the faulty segments totaled more than five. The task EEG analysis was performed on 75 subjects—the average EEG length of the ruminator. The remaining EEG data were then subjected to independent component analysis (ICA) to determine brain signals. Based on the spatial distribution and spectral power, FildTrip was used to find the independent components (ICs), including motor activity, eyeblinks, and ECG, which were then eliminated prior to further analysis. ## EEG power analysis After preprocessing, both the eyes-closed and eyes-open resting state signals were filtered into theta, delta, alpha, and beta bands. Then, we employed the “pwelch” function in MATLAB to calculate the spectral power during the resting state for all 30 channels in each of the four frequency bands: delta (0.5–4 Hz), theta (4–8 Hz), alpha (8–13 Hz), and beta (13–30 Hz). ## Coherence analysis Referring to the analysis method of a previous study by Bakker, we also adopted the coherence method to reflect the alternation in brain activity (den Bakker et al., 2018). The task EEG signals were then filtered into theta and gamma frequency bands (4–80 Hz). The bands were theta (4–8 Hz), alpha (8–13 Hz), beta (13–30 Hz), gamma1 (30–60 Hz), and gamma2 (60–80 Hz). Subsequently, we employed the “mscohere” function in MATLAB to calculate magnitude-squared coherence, which reflects how well signal x corresponds to signal y at each frequency band. The “mscohere” function estimates the magnitude-squared coherence function using Welch's overlapped averaged periodogram method (we used 512 points per window with $90\%$ overlap). The coherence value of signals x and y Cxy(f) was calculated as a function of the following: the spectral densities of signal x, which was denoted as Pxx(f); the spectral densities of y, which was denoted as Pyy(f); and the cross spectral density of x and y, which was denoted as Pxy(f): Then, the coherence values were averaged across all pairs of electrodes over the brain for further analysis. We also calculated the average coherence in each band. To minimize the effects of volume conduction, we set the coherence of neighboring electrodes to zero before calculating the average coherence in each band (Peters et al., 2013; den Bakker et al., 2018). ## Correlation between the coherence of EEGs and the arousal rate The arousal rate of each participant induced by the rumination items was calculated as the proportion of “yes” responses given to the total number of rumination items. Finally, we adopted the Pearson correlation method to investigate the relationship between brain EEG coherence in different frequency bands and the arousal rate induced by the rumination items. ## Statistical analysis Both the behavior results and EEG characteristics were analyzed statistically using SPSS 26 (IBM). To examine the behavioral difference between the two groups across the two item types in each of the seven dimensions, a two-sample t-test was utilized. To determine the power differential of the resting-state EEGs in each frequency band, we utilized a two-way repeated ANOVA using the group and channel as factors. Mauchly's test was applied to test for sphericity, while the Greenhouse–Geisser correction was used to correct the sphericity. The coherence in the five bands was then compared, both in-group and item-wise. The statistical method was also referred to in the previous study (den Bakker et al., 2018). For group-wise comparison, we first employed a two-way repeated ANOVA with the item type and band as factors to compare the averaged coherence across the brain. The difference between the rumination items and the neutral items was determined using the two-sample t-test, and Bonferroni's correction was used for multiple comparisons ($p \leq 0.01$). We then adopted a two-way ANOVA using item type and frequency as factors to assess their differences in each frequency band (theta, alpha, beta, gamma1, and gamma2). For item-wise comparison, we set the group and band as factors and performed the same statistical procedure. ## Demographic information of the participants Table 1 shows the demographic information of the 4,591 subjects who used the RRS. In this study, the total average score of the RRS was 31.57 ± 8.33 ($$n = 4$$,591), and Cronbach's α coefficient for this scale was 0.925 ($$n = 4$$,591). Table 2 shows the demographic information of the 607 high-degree ruminators (577 men and 30 women) who participated in the semi-structured interviews. Table 3 shows the demographic information of the 1,685 subjects. In this part, the total average score of the RRS was 28.64 ± 6.97 ($$n = 1$$,685), and Cronbach's α coefficient for this scale was 0.925 ($$n = 1$$,685). Table 4 shows the demographic information of the 78 chosen subjects from the 1,685 pool of subjects who completed the questionnaire. Table 5 shows the demographic information of the 75 participants who were recruited for the following EEG experiment. Table 5 \| Unnamed: 0 \| Frequency/mean \| Percentage/standard deviation \| \| --- \| --- \| --- \| \| Gender ( n , %) \| Gender ( n , %) \| Gender ( n , %) \| \| Women \| 0 \| 0 \| \| Men \| 75 \| 100 \| \| Age (mean ±SD) \| 21.75 \| 6.18 \| \| Home location ( n , %) \| Home location ( n , %) \| Home location ( n , %) \| \| Urban \| 23 \| 30.67 \| \| Village \| 52 \| 69.33 \| \| Ethnicity ( n , %) \| Ethnicity ( n , %) \| Ethnicity ( n , %) \| \| Han \| 66 \| 88.00 \| \| Others \| 9 \| 12.00 \| ## Natural language feature adaptation paradigm Using NLP techniques, we identified 17 ruminative scenario themes and constructed 37 ruminative arousal-inducing items (including 28 social dilemmas and 9 personal injuries). The validation results for the rumination questionnaire showed that the main effects of the item types were significant in all dimensions ($p \leq 0.001$). Further simple effect analysis showed that the ruminators' scores on rumination items were higher than those of the control group ($p \leq 0.05$), while the rumination group scores of the neutral items were no different from those of the control group (Figure 1). This result confirmed the effectiveness and reliability of the rumination items. Figure 1: Comparison of ruminator and control subject scores on the rumination and neutral items under the seven dimensions. p < 0.05, *p < 0.001, error line represents the standard deviation. ## The reaction time was longer in ruminators for rumination items The ANOVA of the two groups and item type revealed that the main effect of the two groups was not significant [F[1, 73]= 0.792, $$p \leq 0.377$$, ηp2 = 0.11], while the main effect of the item type was significant [F[1, 73] = 107.968, $p \leq 0.001$, ηp2 = 0.597], reflecting that the reaction time of rumination items was longer than that of the neutral items. Additionally, a significant interaction between item type and person category was found [F[1, 73] = 31.901, $p \leq 0.001$, ηp2 = 0.304]. Therefore, further simple effects analysis was conducted, and the results are as follows (Table 6). Table 6 \| Type of item \| Ruminators (n = 46) \| Controls (n = 29) \| t \| P \| \| --- \| --- \| --- \| --- \| --- \| \| Rumination \| 15.731 ± 4.624 \| 12.823 ± 3.143 \| −3.240 \| < 0.01 \| \| Neutral \| 9.663 ± 4.045 \| 11.028 ± 3.524 \| 1.494 \| 0.139 \| The results of the simple effects analysis showed that the response time of the ruminators while engaging with rumination items was longer than that of the control group (df = 73, $p \leq 0.01$), while the response time of the ruminators while engaging with neutral items did not differ from that of the control group. This result further verified the effectiveness of the ruminative questionnaire in inducing the ruminators to immerse themselves in and resonate with the different scenarios established by the questionnaires. ## The arousal rate was significantly higher in ruminators for rumination items In the questionnaire response task, whether for the rumination items or the neutral items, we set the same following question: “Does the above description induce you to have repeated/continuous recall?”. This question reflects the core definition of the concept of rumination. Therefore, we interpreted the choice of “yes” to mean that the participants were aroused by the scenario. Conversely, participants who chose “no” were not considered to be aroused. Arousal rate: the arousal rates elicited by the two item types were calculated as the proportion of the given response “yes” to the total for each kind of item (i.e., the number of “yes” responses given in rumination items/the total number of rumination items). The main effect of the two groups was found to be significant, F[1, 73] = 32.554, $p \leq 0.001$, ηp2 = 0.308, and the arousal rate of the ruminators was higher than that of the control group. The main effect of the item type was also significant, F[1, 73] = 82.300, $p \leq 0.001$, ηp2 = 0.530, and the arousal rate for the rumination items was higher than that for the neutral items. The main effect of the item type and group was found to be significant, F[1, 73] = 82.300, $p \leq 0.001$, ηp2 = 0.530, and was higher than that of the neutral items. A significant interaction of the item type with the two groups was found [F[1, 73] = 46.382, $p \leq 0.001$, ηp2 = 0.389]. Therefore, further simple effects analysis was conducted, and the results are as follows (Table 7). Table 7 \| Type of item \| Ruminators (n = 46) \| Controls (n = 29) \| t \| P \| \| --- \| --- \| --- \| --- \| --- \| \| Rumination \| 0.616 ± 0.324 \| 0.158 ± 0.187 \| −7.746 \| < 0.001 \| \| Neutral \| 0.079 ± 0.146 \| 0.082 ± 0.123 \| 0.095 \| 0.925 \| The results of the simple effects analysis showed that, for the rumination items, the arousal of the ruminators was significantly higher than that of the control group (df = 73, $p \leq 0.001$), while for the neutral items, the arousal of the ruminators did not differ from that of the control group. This finding indicates that the rumination items constructed in this study targeted the recall of the rumination group and thus triggered continuous and repeated recall. ## The brain power of normal subjects and high-degree rumination subjects were highly similar The topography of the two groups in terms of the absolute power of each frequency band is depicted in Figure 2. In both the eyes-closed and eyes-open conditions, the two-way ANOVA did not reveal any group differences. Figure 2: Topographical maps of absolute power in resting state EEGs. (A) Topographical maps of absolute power with eyes closed. (B) Topographical maps of absolute power with eyes opened. (C) Electrode locations of the EEG cap. RT, Rumination group; NC, control group. ## The coherence of the whole brain was decreased while engaging with the rumination items Table 8 shows the average EEG length of each group in different kinds of items. No significant differences were found in the length of EEG data either in groups or items. Table 8 \| Type of item \| Ruminators (n = 46) \| Controls (n = 29) \| \| --- \| --- \| --- \| \| Rumination \| 6,750 ± 1,993 \| 7,124 ± 2,146 \| \| Neutral \| 6,695 ± 2,274 \| 6,709 ± 1,963 \| As shown in Figure 3, the coherence of the whole brain was decreased in both groups while engaging with the rumination items compared to the neutral items in the test. In the rumination group, the two-way repeated ANOVA revealed a significant main effect on the item type [F[1, 90] = 9.445, $$p \leq 0.003$$]. Additionally, a no item × frequency band interaction effect was found [Greenhouse–Geisser corrected, F(1.452, 130.724) = 0.271, $$p \leq 0.691$$]. All the two-sample t-tests for post hoc analysis satisfied the test of homogeneity of variance ($p \leq 0.05$) and revealed that the coherence decreased significantly in the theta (two-tail, t = −3.373, df = 90, $$p \leq 0.001$$), alpha (t = −2.827, df = 90, $$p \leq 0.006$$), beta (t = −3.287, df = 90, $$p \leq 0.001$$), and gamma1 (t = −2.649, df = 90, $$p \leq 0.01$$) bands. In the subjects from the control group, there was no significant effect of the item type [F[1, 56] = 1.724, $$p \leq 0.195$$] and no item type × frequency band interaction effect [Greenhouse–Geisser correction, F(1.583, 88.654) = 0.015, $$p \leq 0.967$$]. The two-sample t-test showed no significant change in the coherence in the five bands while engaging with the two kinds of items in the normal subject group. Figure 3: Within the group coherence analysis, the subjects completed the rumination items and neutral items. (A) Average coherence of all frequency bands in the rumination group. (B) Average coherence for the five frequency bands in the rumination group. (C) Average coherence of all frequency bands in the controls. (D) Average coherence in the five frequency bands in the controls. p < 0.01, *p < 0.001. Red, rumination items; black, neutral items; Normal Groups means control groups. ## The coherence of the whole brain increased in the rumination group Next, we assessed the difference in coherence between the subjects from both groups while they were engaged with the two kinds of items. As shown in Figure 4, the coherence of all frequency bands increased [main effect of group F[1, 73] = 4.916, $$p \leq 0.030$$, no group × frequency band interaction effect, F(1.496, 109.172) = 0.834, $$p \leq 0.407$$, Greehouse-Geisser corrected], especially for the beta (two-tail, $t = 2.301$, df = 71.088 $$p \leq 0.017$$) and gamma2 (two-tail, $t = 2.411$, df = 72.953, $$p \leq 0.018$$) bands while the subjects were engaged with the rumination test. It is worth noting that the two groups showed a significant difference while interacting with the neutral items [main effect of group F[1, 73] = 16.844, $p \leq 0.000$, no group × frequency band interaction effect, F(1.538, 112.238) = 1.467, $$p \leq 0.235$$, Greehouse-Geisser corrected]. As Figure 4D shows, significantly increased coherence emerged in all frequency bands (two-tail: theta, $t = 4.698$, df = 72.949, $p \leq 0.000$; alpha, $t = 4.160$, df = 72.224, $p \leq 0.000$; beta, $t = 4.684$, df = 72.210, $p \leq 0.000$; gamma1, $t = 3.446$, df = 71.754, $$p \leq 0.001$$, and gamma2, $t = 4.057$, df = 72.662, $p \leq 0.000$) in ruminators engaging with the neutral items compared to the normal subjects. This phenomenon may have been due to the decreasing speed of the ruminative mood. This will be discussed later. Figure 4: Within-item coherence analysis of the two groups while they engaged with the rumination items and neutral items. (A) Average coherence of all frequency bands in the rumination items. (B) Average coherence of the five frequency bands in the rumination items. (C) Average coherence of all frequency bands in the neutral items. (D) Average coherence of the five frequency bands in the neutral items. p < 0.01, p < 0.001, p < 0.001. Red, rumination items; black, neutral items; Normal Groups means control groups. ## The arousal rate was only significantly correlated with the coherence of the gamma2 frequency band We conducted a Pearson correlation analysis to identify the relationship between the arousal rate induced by the rumination items and the EEG coherence of different frequency bands. We only found a significant correlation for the gamma2 band ($r = 0.231$, $$p \leq 0.046$$). ## Discussion In this study, we first developed an immersive rumination-inducing questionnaire that could elicit situational recall in ruminators using NLP combined with scales, interviews, and “scenario immersion” theory. The results of behavioral indicators revealed that, for the rumination items of the questionnaire, the average reaction time was longer for the ruminators, and their arousal rate was significantly higher than that of the subjects in the control group. Then, we used EEG techniques to investigate whether the immersive ruminative questionnaire could induce certain neural activities, specifically in the ruminators. The resting-state EEG results showed that there was no difference in the power of the brain between the ruminators and the control group. However, the EEG coherence analysis showed that the brain activity of the ruminators was significantly higher than that of the control group for the rumination items. Combined with the behavior result and EEG evidence, the immersive ruminative questionnaire developed with natural language features and the scenario immersion theory was successful in eliciting more immersive ruminative thinking, specifically in the ruminators. In conclusion, the questionnaire based on NLP appears to be suitable as a novel paradigm for psychological selection in the early detection of mental disorders. ## The immersive ruminative questionnaire elicited immersive ruminative thinking, specifically in the ruminators Previous studies have used various methods to induce rumination, including short statement prompts (Cooney et al., 2010; Berman et al., 2014; Milazzo et al., 2016), texts extracted from Wikipedia (Curci et al., 2015), characteristic words (Yoshimura et al., 2009; Moran et al., 2014; Apazoglou et al., 2019), music clips (Figueroa et al., 2017), self-reports (Rosenbaum et al., 2018a), emotive facial expressions (Aker et al., 2014), goal prompting tasks (Zhan et al., 2017; Mollaahmetoglu et al., 2021), videos (Bostanov et al., 2018), and other materials. However, it remains unclear whether these methods successfully induce subjects' rumination. Natural language is the first means by which humans express their thoughts and rapidly communicate accurately (Sun et al., 2013; Lee et al., 2014). Written materials pertaining to specific episodic memories represent one of the most effective forms of content to induce rumination (Haque et al., 2014; Wilson-Mendenhall et al., 2019). Rapidly developing NLP technology and “scenario immersion” theory provide us with technical support and a theoretical basis for constructing state-inducing questionnaires. In this study, we used the natural language corpus of 607 ruminators and NLP to develop a mental state-eliciting questionnaire that could elicit situational memories according to “scenario immersion” theory to activate the unique ruminative state of rumination-prone individuals both effectively and accurately. The evaluation results showed that the scores of ruminators were significantly higher than those of the control group only on the rumination items in all dimensions. On the contrary, there was no significant difference between the neutral items. Therefore, the results indicate that the ruminators agreed with the scenario description constructed by the rumination items, which could not only trigger repeated, vivid, and continuous negative immersive memories but was also highly representative of ruminative elicitation. In other words, the questionnaire we developed could elicit immersive ruminative thinking in ruminators. ## The rumination items were capable of inducing brain activity in ruminators The analysis of EEG power in the four frequency bands revealed no significant difference between the ruminators and the control group while they were at rest, indicating that there was no difference in brain activity between the ruminators and the control group in this state, which might be a contributing factor to the difficulty of diagnosing depression during its nonclinical state. EEG coherence is a measure of synchronization of the two recorded EEG signals and has been widely used to indict the dysregulation of the human brain (den Bakker et al., 2018; Minami et al., 2022; Wang et al., 2022). We examined the task EEG data from both the item type aspect and group aspect to study if the different items could elicit different brain activities in the two groups. First, we discovered that EEG coherence decreased in both groups while subjects engaged with the rumination items. The ruminators presented a significantly greater decrease in the theta, alpha, beta, and gamma frequency bands. Although a slight decrease was also observed in the control group, no statistically significant differences were identified for any of the frequency bands. EEG coherence reflects the synchronization between brain cortical regions (Markovska-Simoska et al., 2018), whereby the observed significant decrease in the coherence of the whole brain indicates that the rumination items caused more dyssynchronization of the whole brain in the ruminators, thereby demonstrating that the questionnaire was capable of inducing ruminative thinking in these subjects. We also discovered that the EEG coherence of the ruminators dramatically increased both in the beta and gamma2 bands compared to the control group while engaging with the rumination items. In the ruminators, this aberrantly elevated coherence may indicate a decreased inhibitory ability of the brain, which might be caused by a depressed state of mind (Cheng et al., 2016), excessive self-focus, and the recall of unpleasant memories that the rumination items elicited (Berman et al., 2011; Zamoscik et al., 2014). The hyper-synchronization in the beta band combined with the longer reaction time suggests that the attention ability was damaged in the ruminators (Li et al., 2017). As the EEG gamma band has been shown to be related to emotions (Li et al., 2015), the increased coherence observed in the gamma2 band provides evidence that the rumination items induced an unpleasant mood in the ruminators. Finally, we found that the coherence of the whole brain in the five frequency bands increased significantly, which can be explained by the phenomenon where the influence of the rumination items did not subside immediately in the ruminators but instead was prolonged to influence the brain activity of these subjects while they were engaged with the neutral items. This also indicates a decline in their ability to control the brain after engaging with the rumination items. The above findings showed that ruminators are more susceptible to negative scenarios, such as poor memories, bad moods, and inaccurate self-referential thinking. Over the past several years, studies have been conducted to identify the neurological bases of rumination in both clinical and nonclinical psychological diseases with the goal of developing potential biomarkers for the diagnosis and therapy of rumination-related mood disorders such as depression (Zhang et al., 2020), euthymic bipolar disease (Apazoglou et al., 2019), posttraumatic stress disorder (Philippi et al., 2020), and so on. Prior fMRI and EEG studies revealed that rumination is associated with altered brain functional connectivity, both increased (Benschop et al., 2020) and decreased (Tozzi et al., 2021). According to some studies, ruminators process self-related information excessively when exposed to external, which may be related to the overactive core subsystem in the default mode network (Lin et al., 2022), particularly the functional connectivity to the prefrontal cortex, which can be a useful neural marker to identify an individual at risk for depression (Benschop et al., 2020). Consistent with the findings of previous studies, we found that the EEG coherence of the whole brain increased significantly in the ruminators, suggesting that, through situational immersion, the questionnaire we developed can elicit excessive processing of self-related information in the high ruminators. This result indicates the validity of the questionnaire we developed. However, more research is required to determine the neurological mechanisms underlying the brain alteration that the questionnaire elicited, i.e., the prefrontal cortex activity and the relationship between different brain networks, which might be used as a neural marker of ruminative thinking. ## Limitations There are some limitations to this study. First, all the participants were men due to the nature of the police academy. Although it is widely acknowledged that women tend to ruminate more than men do, one review reported that both men and women showed strong and significant statistical correlations between depressive symptoms and ruminative thoughts (rho > 0.50; $p \leq 0.05$), suggesting that the relationship between depressive symptoms and rumination does not necessarily explain the sex differences observed in depression (Shors et al., 2017). Hence, further study is required to explore the responses of women. Moreover, since all the subjects were enrolled in the Shaanxi Police College, more subjects will be needed from the general population in the following experiment. Second, the EEG evidence of ruminative thoughts from the perspective of the entire brain was the focus of this study, while specific brain regions or brain networks should be considered to identify the neuromechanism underlying rumination. Finally, neural biomarkers of rumination may be used to predict ruminators, which would be of great significance in psychological selection. In the future, we will explore the predictive power of the rumination items and EEG data. ## 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 Committee of the Air Force Medical University. The patients/participants provided their written informed consent to participate in this study. ## Author contributions YuL contributed to the development of the questionnaire and completed the whole experiment. CL analysis the EEG data and wrote the manuscript. TZ and LWu contributed to revision of the manuscript. XL, YiL, and LWa helped to analyze the questionnaire behavior data. DL and HY contribute big work to the experiment. DM and PF organized and managed the whole experiment and 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/fnins.2023.1118650/full#supplementary-material ## References 1. Aker M., Harmer C., Landrø N. I.. More rumination and less effective emotion regulation in previously depressed women with preserved executive functions. BMC Psychiatry (2014) 14 334. DOI: 10.1186/s12888-014-0334-4 2. Ancin-Murguzur F. J., Hausner V. H.. CausalizeR: a text mining algorithm to identify causal relationships in scientific literature. PeerJ (2021) 9 e11850. DOI: 10.7717/peerj.11850 3. Apazoglou K., Küng A., Cordera P., Aubry J., Dayer A., Vuilleumier P.. Rumination related activity in brain networks mediating attentional switching in euthymic bipolar patients. Int. J. Bipolar Disord. (2019) 7 3. DOI: 10.1186/s40345-018-0137-5 4. Assale M., Dui L. G., Cina A., Seveso A., Cabitza F.. The revival of the notes field: leveraging the unstructured content in electronic health records. Front. Med. (2019) 6 e00066. DOI: 10.3389/fmed.2019.00066 5. Benschop L., Poppa T., Medani T., Shahabi H., Baeken C., Leahy R. M.. Electrophysiological scarring in remitted depressed patients: elevated EEG functional connectivity between the posterior cingulate cortex and the subgenual prefrontal cortex as a neural marker for rumination. J. Affect. Disord. (2020) 281 493-501. DOI: 10.1016/j.jad.2020.12.081 6. Berman M., Peltier G., Scott N., Evan D., Patricia J.. Depression, rumination and the default network. Soc. Cogn. Affect. Neurosci. (2011) 6 548-555. DOI: 10.1093/scan/nsq080 7. Berman M. G., Misic B., Buschkuehl M., Kross E., Deldin P. J., Peltier S.. Does resting-state connectivity reflect depressive rumination? A tale of two analyses. NeuroImage (2014) 103 267-279. DOI: 10.1016/j.neuroimage.2014.09.027 8. Blei D. M., Ng A., Jordan M. I.. Latent dirichlet allocation. J. Mach. Learn. Res. (2003) 3 993-1022. DOI: 10.1162/jmlr.2003.3.4-5.993 9. Bostanov V., Ohlrogge L., Britz R., Hautzinger M., Kotchoubey B.. Measuring mindfulness: a psychophysiological approach. Front. Hum. Neurosci. (2018) 12 e00249. DOI: 10.3389/fnhum.2018.00249 10. Castanheira L., Silva C., Cheniaux E., Telles-Correia D.. Neuroimaging correlates of depression-implications to clinical practice. Front Psychiatry (2019) 10 703. DOI: 10.3389/fpsyt.2019.00703 11. Cheng W., Rolls E. T., Qiu J., Liu W., Tang Y., Huang C.. Medial reward and lateral non-reward orbitofrontal cortex circuits change in opposite directions in depression. Brain (2016) 139 3296-3309. DOI: 10.1093/brain/aww255 12. Cooney R. E., Joormann J. F, Eugène Dennis E. L., Gotlib I. H.. Neural correlates of rumination in depression. Cogn. Affect. Behav. Neurosci. (2010) 10 470-478. DOI: 10.3758/CABN.10.4.470 13. Curci A., Soleti E., Lanciano T., Doria V., Rimé B.. Balancing emotional processing with ongoing cognitive activity: the effects of task modality on intrusions and rumination. Front. Psychol. (2015) 6 e01275. DOI: 10.3389/fpsyg.2015.01275 14. den Bakker H., Sidorov M. S., Fan Z., Lee D. J., Bird L. M., Chu C. J.. Abnormal coherence and sleep composition in children with Angelman syndrome: a retrospective EEG study. Mol. Autism (2018) 9 32. DOI: 10.1186/s13229-018-0214-8 15. Deshpande G., Rangaprakash D., Oeding L., Cichocki A., Hu X. P.. A new generation of brain-computer interfaces driven by discovery of latent EEG-fMRI linkages using tensor decomposition. Front. Neurosci. (2017) 11 e00246. DOI: 10.3389/fnins.2017.00246 16. Dunning D., Heath C., Suls J. M.. Picture Imperfect (2005) 17. Ferdek M. A., van Rijn C. M., Wyczesany M.. Depressive rumination and the emotional control circuit: an EEG localization and effective connectivity study. Cogn. Affect. Behav. Neurosci. (2016) 16 1099-1113. DOI: 10.3758/s13415-016-0456-x 18. Ferrario A., Demiray B., Yordanova K., Luo M., Martin M.. Social reminiscence in older adults' everyday conversations: automated detection using natural language processing and machine learning. J. Med. Internet Res. (2020) 22 e19133. DOI: 10.2196/19133 19. Figueroa C. A., Mocking R. J. T., van Wingen G., Martens S., Ruhé H. G., Schene A. H.. Aberrant default-mode network-hippocampus connectivity after sad memory-recall in remitted-depression. Social Cogn. Affect. Neurosci. (2017) 12 1803-1813. DOI: 10.1093/scan/nsx108 20. Ghosh D. D., Guha R.. What are we ‘tweeting' about obesity? Mapping tweets with topic modeling and Geographic Information System. Cartogr. Geogr. Inform. Sci. (2013) 40 90-102. DOI: 10.1080/15230406.2013.776210 21. Han C., Xu G., Zheng X., Tian P., Zhang K., Yan W.. Assessing the effect of the refresh rate of a device on various motion stimulation frequencies based on steady-state motion visual evoked potentials. Front. Neurosci. (2022) 15 e757679. DOI: 10.3389/fnins.2021.757679 22. Han X., Yang H.. Chinese version of Nolen-Hoeksema ruminative responses scale (RRS) used in 912 college students: reliability and validity. Chin. J. Clin. Psychol. (2009) 5 550-551 23. Hao H., Zhang K., Wang W., Gao G.. A tale of two countries: international comparison of online doctor reviews between China and the United States. Int. J. Med. Inform. (2017) 99 37-44. DOI: 10.1016/j.ijmedinf.2016.12.007 24. Haque S., Juliana E., Khan R., Hasking P.. Autobiographical memory and hierarchical search strategies in depressed and non-depressed participants. BMC Psychiatry (2014) 14 310. DOI: 10.1186/s12888-014-0310-z 25. Jurafsky D. M. J. H.. Speech and Language Processing, 2nd Edn (2009). PMID: 18754917 26. Lebois L. A. M., Wilson-Mendenhall C. D., Simmons W. K., Barrett L. F., Barsalou L. W.. Learning situated emotions. Neuropsychologia (2020) 145 106637. DOI: 10.1016/j.neuropsychologia.2018.01.008 27. Lee M. C., Chang J. W., Hsieh T. C.. A grammar-based semantic similarity algorithm for natural language sentences. Sci. World J. (2014) 2014 1-17. DOI: 10.1155/2014/437162 28. Li B. J., Karl F., Maria M., Wang H. N., Lu H. B., Hu D. W.. A brain network model for depression: from symptom understanding to disease intervention. CNS Neurosci. Therap. (2018) 24 1004-1019. DOI: 10.1111/cns.12998 29. Li E. T., Luyten P., Midgley N.. Psychological mediators of the association between childhood emotional abuse and depression: a systematic review. Front. Psychiatry (2020) 11 559213. DOI: 10.3389/fpsyt.2020.559213 30. Li H., Zhao Q., Huang F., Cao Q., Qian Q., Johnstone S. J.. Increased beta activity links to impaired emotional control in ADHD adults with high IQ. J. Atten. Disord. (2017) 23 754-764. DOI: 10.1177/1087054717739120 31. Li Y., Cao D., Wei L., Tang Y., Wang J.. Abnormal functional connectivity of EEG gamma band in patients with depression during emotional face processing. Clin. Neurophysiol. (2015) 126 2078-2089. DOI: 10.1016/j.clinph.2014.12.026 32. Lin W., He H., Guan Q.. Functional brain networks underlying rumination. Adv. Psychol. Sci. (2022) 30 1262-1269. DOI: 10.3724/SP.J.1042.2022.01262 33. Markovska-Simoska S., Pop-Jordanova N., Pop-Jordanov J.. Inter- and intra-hemispheric EEG coherence study in adults with neuropsychiatric disorders. Prilozi (2018) 39 5-19. DOI: 10.2478/prilozi-2018-0037 34. Miao B., Liu X., Zhu T.. Automatic mental health identification method based on natural gait pattern. PsyCh J. (2021) 8 434. DOI: 10.1002/pchj.434 35. Milazzo A., Ng B., Jiang H., Shirer W., Varoquaux G., Poline J. B.. Identification of mood-relevant brain connections using a continuous, subject-driven rumination paradigm. Cerebral. Cortex (2016) 26 933-942. DOI: 10.1093/cercor/bhu255 36. Min J. Y., Song S. H., Kim H. J., Min K. B.. Mining hidden knowledge about illegal compensation for occupational injury: topic model approach. JMIR Med. Inform. (2019) 7 e14763. DOI: 10.2196/14763 37. Min K., Song S., Min J.. Topic modeling of social networking service data on occupational accidents in Korea: latent dirichlet allocation analysis. J. Med. Internet Res. (2020) 22 e19222. DOI: 10.2196/19222 38. Minami S., Kato M., Ikeda S., Yoshimura M., Ueda S., Koshikawa Y.. Association between the rostral anterior cingulate cortex and anterior insula in the salience network on response to antidepressants in major depressive disorder as revealed by isolated effective coherence. Neuropsychobiology (2022) 81 475-483. DOI: 10.1159/000525338 39. Mollaahmetoglu O. M., Palmer E., Maschauer E., Nolan M. C., Stevens T., Carlyle M.. The acute effects of alcohol on state rumination in the laboratory. Psychopharmacology (2021) 238 1671-1686. DOI: 10.1007/s00213-021-05802-1 40. Moran J. M., Macrae C. N., Heatherton T. F., Wyland C. L., Kelley W. M.. Neuroanatomical evidence for distinct cognitive and affective components of self. J. Cogn. Neurosci. (2014) 18 1586-1594. DOI: 10.1162/jocn.2006.18.9.1586 41. Nolen-Hoeksema S.. Responses to depression and their effects on the duration of depressive episodes. J. Abnormal Psychol. (1965) (1991) 100 569-582. DOI: 10.1037/0021-843X.100.4.569 42. Nolen-Hoeksema S., Morrow J.. A prospective study of depression and posttraumatic stress symptoms after a natural disaster: the 1989 Loma Prieta Earthquake. J. Pers. Soc. Psychol. (1991) 61 115-121. DOI: 10.1037/0022-3514.61.1.115 43. Nolen-Hoeksema S., Wisco B. E., Lyubomirsky S.. Rethinking rumination. Perspect. Psychol. Sci. (2008) 3 400-424. DOI: 10.1111/j.1745-6924.2008.00088.x 44. Nolen-Hoeksema S., Wolfson A., Mumme D., Guskin K.. Helplessness in children of depressed and nondepressed mothers. Develop. Psychol. (1995) 31 377-387. DOI: 10.1037/0012-1649.31.3.377 45. Oostenveld R., Fries P., Maris E., Schoffelen J. M.. FieldTrip: open source software for advanced analysis of MEG, EEG, and invasive electrophysiological data. Comput. Intellig. Neurosci. (2011) 2011 156869. DOI: 10.1155/2011/156869 46. Peters J. M., Taquet M., Vega C., Jeste S. S., Fernandez I. S., Tan J.. Brain functional networks in syndromic and non-syndromic autism: a graph theoretical study of EEG connectivity. BMC Med. (2013) 11 54. DOI: 10.1186/1741-7015-11-54 47. Philippi C. L., Pessin S., Reyna L., Floyd T., Bruce S. E.. Cortical midline structures associated with rumination in women with PTSD. J. Psychiatric Res. (2020) 131 69-76. DOI: 10.1016/j.jpsychires.2020.09.001 48. Reviewer-Lee L.. Review of “Foundations of Statistical Natural Language Processing” (2000) 49. Robinson P. A.. Neurophysical theory of coherence and correlations of electroencephalographic and electrocorticographic signals. J. Theor. Biol. (2003) 222 163-175. DOI: 10.1016/S0022-5193(03)00023-7 50. Rosenbaum D., Hilsendegen P., Thomas M., Haeussinger F. B., Nuerk H., Fallgatter A. J.. Disrupted prefrontal functional connectivity during post-stress adaption in high ruminators. Sci. Rep. (2018a) 8 15588. DOI: 10.1038/s41598-018-33777-1 51. Rosenbaum D., Thomas M., Hilsendegen P., Metzger F. G., Haeussinger F. B., Nuerk H.. Stress-related dysfunction of the right inferior frontal cortex in high ruminators: an fNIRS study. NeuroImage: Clin. (2018b) 18 510-517. DOI: 10.1016/j.nicl.2018.02.022 52. Schwarz N.. Why Researchers Should Think “Real-Time”: A Cognitive Rationale: Handbook of Research Methods for Studying Daily Life (2012) 53. Shors T. J., Millon E. M., Chang H. Y. M., Olson R. L., Alderman B. L.. Do sex differences in rumination explain sex differences in depression?. J. Neurosci. Res. (2017) 95 711-718. DOI: 10.1002/jnr.23976 54. Sun W., Rumshisky A., Uzuner O.. Temporal reasoning over clinical text: the state of the art. J. Am. Med. Inform. Assoc. (2013) 20 814-819. DOI: 10.1136/amiajnl-2013-001760 55. Sutherland K., Bryant R. A.. Rumination and overgeneral autobiographical memory. Behav. Res. Ther. (2007) 45 2407-2416. DOI: 10.1016/j.brat.2007.03.018 56. Tozzi L., Zhang X., Chesnut M., Holt-Gosselin B., Ramirez C. A., Williams L. M.. Reduced functional connectivity of default mode network subsystems in depression: Meta-analytic evidence and relationship with trait rumination. NeuroImage: Clin. (2021) 30 102570. DOI: 10.1016/j.nicl.2021.102570 57. Van Doorn M., Nijhuis L. A., Egeler M. D., Daams J. G., Popma A., van Amelsvoort T.. Online indicated preventive mental health interventions for youth: a scoping review. Front. Psychiatry (2021) 12 580843. DOI: 10.3389/fpsyt.2021.580843 58. Wang X., Liu X., Wang Y., Zhu T.. How can people express their trait self-esteem through their faces in 3D space?. Front. Psychol. (2021) 12 e591682. DOI: 10.3389/fpsyg.2021.591682 59. Wang X., Wang Y., Zhou M., Li B., Zhu T.. Identifying psychological symptoms based on facial movements. Front. Psychiatry (2020) 168 S58. DOI: 10.3389/fpsyt.2020.607890 60. Wang Z., Dong F., Sun Y., Wang J., Zhang M., Xue T.. Increased resting-state alpha coherence and impaired inhibition control in young smokers. Front. Neurosci. (2022) 16 e1026835. DOI: 10.3389/fnins.2022.1026835 61. Watkins E. R., Roberts H.. Reflecting on rumination: consequences, causes, mechanisms and treatment of rumination. Behav. Res. Ther. (2020) 127 103573. DOI: 10.1016/j.brat.2020.103573 62. Wells A., Matthews G.. Modelling cognition in emotional disorder: the S-REF model. Behav. Res. Ther. (1996) 34 881-888. DOI: 10.1016/S0005-7967(96)00050-2 63. Wilson-Mendenhall C. D., Barrett L. F., Barsalou L. W.. Variety in emotional life: within-category typicality of emotional experiences is associated with neural activity in large-scale brain networks. Soc. Cogn. Affect. Neurosci. (2015) 10 62-71. DOI: 10.1093/scan/nsu037 64. Wilson-Mendenhall C. D., Henriques A., Barsalou L. W., Barrett L. F.. Primary interoceptive cortex activity during simulated experiences of the body. J. Cogn. Neurosci. (2019) 31 221-235. DOI: 10.1162/jocn_a_01346 65. Yadav N., Noble C., Niemeyer J. E., Terceros A., Victor J., Liston C.. Prefrontal feature representations drive memory recall. Nature (2022) 608 153-160. DOI: 10.1038/s41586-022-04936-2 66. Yoshimura S., Ueda K., Suzuki S., Onoda K., Okamoto Y., Yamawaki S.. Self-referential processing of negative stimuli within the ventral anterior cingulate gyrus and right amygdala. Brain Cogn. (2009) 69 218-225. DOI: 10.1016/j.bandc.2008.07.010 67. Yuan Z., Peng Y., Wang L., Song S., Chen S., Yang L.. Effect of BCI-controlled pedaling training system with multiple modalities of feedback on motor and cognitive function rehabilitation of early subacute stroke patients. IEEE Trans. Neural Syst. Rehabil. Eng. (2021) 29 2569-2577. DOI: 10.1109/TNSRE.2021.3132944 68. Zamoscik V., Huffziger S., Ebner-Priemer U., Kuehner C., Kirsch P.. Increased involvement of the parahippocampal gyri in a sad mood predicts future depressive symptoms. Soc. Cogn. Affect. Neurosci. (2014) 12 2034-2040. DOI: 10.1093/scan/nsu006 69. Zhan J., Tang F., He M., Fan J., Xiao J., Liu C.. Regulating rumination by anger: evidence for the mutual promotion and counteraction (MPMC) theory of emotionality. Front. Psychol (2017). DOI: 10.3389/fpsyg.2017.01871 70. Zhang R., Kranz G., Zou W., Deng Y., Huang X., Lin K.. Rumination network dysfunction in major depression: a brain connectome study. Prog. Neuro-Psychopharmacol. Biol. Psychiatry (2020) 98 109819. DOI: 10.1016/j.pnpbp.2019.109819
--- title: 'The role of gender inequality and health expenditure on the coverage of demand for family planning satisfied by modern contraceptives: a multilevel analysis of cross-sectional studies in 14 LAC countries' authors: - Laísa Rodrigues Moreira - Cauane Blumenberg - Beatriz Elena Caicedo Velasquez - Fernanda Ewerling - Alejandra Balandrán - Luis Paulo Vidaletti - Andrea Ramirez Varela - Franciele Hellwig - Rodolfo Gomez Ponce de Leon - Aluisio J.D. Barros - Mariangela Freitas Silveira - Fernando C. Wehrmeister journal: Lancet Regional Health - Americas year: 2023 pmcid: PMC10025422 doi: 10.1016/j.lana.2023.100435 license: CC BY 4.0 --- # The role of gender inequality and health expenditure on the coverage of demand for family planning satisfied by modern contraceptives: a multilevel analysis of cross-sectional studies in 14 LAC countries ## Body Research in contextEvidence before this studyWe searched PubMed database using the search terms (“Contextual factors”) AND (“Family planning”) AND (“Multilevel”), with no language restrictions, for results up to November 19, 2022. Using this search strategy, we have identified only 11 studies. Most of these references were studies conducted in Africa, indicating the importance of contextual factors, with only three multi-country studies. Furthermore, in the PubMed database, 130 references were available when we used the following search strategy (“contextual factors” OR “Gender inequality” OR “Health expenditure”) AND (“Family planning”), indicating an existence of few previous research and possible gaps for this area. In the Latin America and the Caribbean (LAC) region, efforts to improve reproductive health indicators were implemented over time, leading to a rapid decline in the fertility rate in the last decades. However, there is no consensus on which factors are more relevant to the demand for family planning satisfied by modern contraceptive methods (DFPSm) at the country level. Added value of this studyThis study presents the first evidence that Gender Inequality Index, a country-level measure of disadvantages affecting women (dimensions: empowerment, reproductive health, and labour market), contributes to coverage of the demand for family planning satisfied. Macro and individual-level factors should be considered when analysing family planning. Especially related to gender inequality, the lower gender inequality in the country, the higher advantages to achieving universal coverage for demand for family planning satisfied with modern methods. Implications of all the available evidenceInternational efforts to improve sexual and reproductive health indicators were implemented over time. This study provides key evidence for practice and policy, implying that including macro-level approaches focused on reducing gender disparities and considering individual-level factors is important in this field. Improving the current indicators by including contextually relevant factors to the use of modern contraception is promising. ## Summary ### Background Despite international efforts to improve reproductive health indicators, little attention is paid to the contributions of contextual factors to modern contraceptive coverage, especially in the Latin America and the Caribbean (LAC) region. This study aimed to identify the association between country-level Gender Inequality and Health Expenditure with demand for family planning satisfied by modern contraceptive methods (DFPSm) in Latin American sexually active women. ### Methods Our analyses included data from the most recent (post-2010) Demographic and Health Survey or Multiple Indicator Cluster Survey from 14 LAC countries. Descriptive analyses and multilevel logistic regressions were performed. Six individual-level factors were included. The effect of the country-level factors Gender Inequality Index (GII) and Current Health Expenditure on DFPSm was investigated. ### Findings DFPSm ranged from $41.8\%$ ($95\%$ CI: 40.2–43.5) in Haiti to $85.6\%$ ($95\%$ CI: 84.9–86.3) in Colombia, with an overall median coverage of $77.8\%$. A direct association between the odds of DFPSm and woman's education, wealth index, and the number of children was identified. Women from countries in the highest GII tertile were less likely (OR: 0.32, $95\%$ CI: 0.13–0.76) to have DFPSm than those living in countries in the lowest tertile. ### Interpretation Understanding the contribution of country-level factors to modern contraception may allow macro-level actions focused on the population's reproductive needs. In this sense, country-level gender inequalities play an important role, as well as individual factors such as wealth and education. ### Funding $\frac{10.13039}{100000865}$Bill and Melinda Gates Foundation and $\frac{10.13039}{501100012418}$Associação Brasileira de Saúde Coletiva (ABRASCO). ## Evidence before this study We searched PubMed database using the search terms (“Contextual factors”) AND (“Family planning”) AND (“Multilevel”), with no language restrictions, for results up to November 19, 2022. Using this search strategy, we have identified only 11 studies. Most of these references were studies conducted in Africa, indicating the importance of contextual factors, with only three multi-country studies. Furthermore, in the PubMed database, 130 references were available when we used the following search strategy (“contextual factors” OR “Gender inequality” OR “Health expenditure”) AND (“Family planning”), indicating an existence of few previous research and possible gaps for this area. In the Latin America and the Caribbean (LAC) region, efforts to improve reproductive health indicators were implemented over time, leading to a rapid decline in the fertility rate in the last decades. However, there is no consensus on which factors are more relevant to the demand for family planning satisfied by modern contraceptive methods (DFPSm) at the country level. ## Added value of this study This study presents the first evidence that Gender Inequality Index, a country-level measure of disadvantages affecting women (dimensions: empowerment, reproductive health, and labour market), contributes to coverage of the demand for family planning satisfied. Macro and individual-level factors should be considered when analysing family planning. Especially related to gender inequality, the lower gender inequality in the country, the higher advantages to achieving universal coverage for demand for family planning satisfied with modern methods. ## Implications of all the available evidence International efforts to improve sexual and reproductive health indicators were implemented over time. This study provides key evidence for practice and policy, implying that including macro-level approaches focused on reducing gender disparities and considering individual-level factors is important in this field. Improving the current indicators by including contextually relevant factors to the use of modern contraception is promising. ## Introduction In the last decades, the total fertility rate rapidly declined in the Latin America and Caribbean (LAC) region, from 5.9 births per woman in 1960 to 1.9 births per woman in 2020, producing demographic, social, and economic changes in this region, affecting the age structure and life expectancy.1,2 In the LAC region, there are evident characteristics such as high adolescent fertility rates (60 births per 1000 women ages 15–19 in 2020), high urban population percentages ($81\%$ of the total population in 2021), high total unemployment percentages ($9.7\%$ of the total labour force in 2021–national estimate), and a total life expectancy at birth of 76 years in 2020.2 The demand for family planning satisfied by modern contraceptive methods (indicator that includes women of reproductive age in need of contraception in its denominator) varies between LAC countries. Brazil ($93.7\%$), Ecuador ($89.8\%$), Cuba ($89.5\%$), Costa Rica ($86.8\%$), Colombia ($83.2\%$), Dominican Republic ($82.9\%$), and Mexico ($81.5\%$) are countries with higher coverage of demand for family planning satisfied by modern contraceptive methods, while Bolivia ($43.4\%$) and Haiti ($44.1\%$) are identified as countries with low levels of coverage.3, 4, 5 In this sense, it is important to understand the processes producing these changes and the possible factors influencing the LAC indicators. Countries in the LAC region present low national fertility rates, which vary according to population subgroups. Contraceptive use also presents inequalities, especially regarding the use of long-acting reversible contraceptives (LARCs).3,6 The subgroups of sexually active women from the rural area, adolescents 15–17 years old, from lower wealth quintiles, indigenous ethnicity, and with no education presented a lower prevalence of LARC contraception compared to their peers in the same country.6 The limited offer of a mix of modern contraceptive methods, or even the lack of availability of modern contraception and reduced access to health care, contribute to long-lasting inequalities in some localities.7,8 In this sense, the Sustainable Development Goals offer directions to improve family planning actions, especially targeting access and availability of contraceptive methods for all.9 These directions mainly target individual-level factors that are strongly associated with modern contraceptive methods utilization, including the level of education, age at first sex, marital status, and mass media exposure.10, 11, 12 More recently, country-level factors have also been explored as potential factors that could affect the use of modern contraceptive methods due to their possible implications on women's sexual and reproductive health, such as influences on unintended pregnancy estimates, community knowledge level of modern contraceptives and attitudes towards family planning.10,13, 14, 15 Recent analyses show that modern contraception is affected by contextual-level factors, including a convenient location of health facilities, exposure to family planning messages, living in localities with low maternal mortality and high antenatal care coverage, and aspects related to the quality of family planning care.13,16, 17, 18 However, there is no consensus on which contextual factors influence family planning coverage between LAC countries. Worldwide, Gender Inequality Index (GII) was pointed out as an important contextual contributor to different health-related outcomes.19 Furthermore, GII includes empowerment as one of its dimensions. Studies have demonstrated the effects of different domains of women's empowerment on family planning.20,21 However, the effects of gender inequality as a contextual-level factor on family planning are not clear. In addition, Health Expenditure and Gross Domestic Product (GDP) were mentioned as relevant contextual factors that allow worldwide comparisons for various health indicators using different analytical approaches.22, 23, 24 Evidence has also indicated that country-level expenditures on reproductive health and family planning contributed to the use of contraception.25, 26, 27 However, there is a lack of contextual data available for reproductive health and family planning expenditure in different countries, especially for those from the LAC region. In this sense, the investigation of the health expenditure seems to be a suitable approximation to the country's investments.22 Investigations about the use of contraception and its associated factors are available in the scientific literature, but multilevel analyses assessing simultaneously the influence of country-level factors and individual factors on demand for family planning satisfied by modern contraceptive methods are scarce, particularly in the LAC region.10, 11, 12,14,15,28 Ignoring the intertwined effects of these factors could mask inter-country, regional, and inter-regional reproductive health differences worldwide. This study aimed to investigate individual and country-level factors' roles in the demand for family planning satisfied by modern contraceptive methods (DFPSm). In particular, this research seeks to examine whether country-level GII, and Health Expenditure affect DFPSm above and beyond women's individual-level characteristics. ## Methods We investigated data of 109,149 sexually active women of reproductive age (15–49 years old), irrespective of marital status, from 14 LAC countries: Belize, Colombia, Costa Rica, Cuba, Dominican Republic, El Salvador, Guatemala, Guyana, Haiti, Honduras, Mexico, Paraguay, Suriname, and Trinidad and Tobago. For each country, the most recent (post-2010) Multiple Indicator Cluster Surveys (MICS) or Demographic and Health Surveys (DHS) were included in the analyses. DHS and MICS are publicly available population-based cross-sectional standardised surveys that allow the comparison of indicators between countries.29, 30, 31 Both surveys aimed to investigate child, maternal, and reproductive health data using design peculiarities and multistage sampling strategies for participants' selection, with more detailed information available in the Supplementary Material (File 1) and elsewhere.29, 30, 31 Data has a natural hierarchy structure with 109,149 women nested within the 14 countries. Other studies using DHS and MICS surveys for epidemiological or public health research were also identified in the LAC context.32, 33, 34, 35, 36, 37 ## Sources of data Data about the DHS and MICS cross-sectional surveys were obtained from: <https://dhsprogram.com/data/available-datasets.cfm> and <https://mics.unicef.org/surveys>, respectively. Data for GDP per capita were available at the World Bank website data <https://data.worldbank.org/indicator/NY.GDP.PCAP.CD?view=chart>. The CHE%GDP data were accessed in the Global Health Expenditure Database <https://apps.who.int/nha/database/Select/Indicators/en> with data available up to 2019. We used country-level data corresponding to the survey year included in the analysis. For the Gender Inequality Index (GII), data were available at Human Development Data Center <http://hdr.undp.org/en/data>. ## Outcome definition DFPSm was defined as among women of reproductive age (15–49 years old), sexually active at the moment of the interview (married or in a union; or women who had sexual intercourse in the last 30 days), and in need of contraception, those who are using modern contraceptive methods. Modern contraceptive methods included male and female sterilisation, subdermal implants, intrauterine devices, oral contraceptives, male and female condoms, emergency contraceptive pills, injectables, vaginal rings, and patches.38 Women were considered in need of family planning if they were fecund and did not intend to become pregnant within the next two years or were unsure about when or whether they wanted to become pregnant.30 In addition, pregnant women whose pregnancy was mistimed or unwanted were also defined as needing contraceptive use. This indicator better captures the success and gaps in family planning programs. It illustrates a strong commitment to the rights of individuals and couples to determine the number and timing of their children.39 Also, this indicator is part of those monitored by the Sustainable development goals. ## Independent factors Independent factors were grouped into individual- and country-level factors. The choice of the individual-level independent factors to be included was motivated by the scientific literature in this area.10, 11, 12, 13, 14, 15,28 Individual-level factors included: 1) marital status; 2) current woman's age; 3) woman's schooling; 4) wealth index (constructed based on a principal component analysis including household characteristics and ownership of selected assets)40; 5) area of residence; and 6) number of children. Regarding the country-level independent factors, in this study, we decided to include more simple and general known factors to be cautious, avoiding a possible black box effect and any spurious relationships. Country-level factors were: Current Health Expenditure (CHE) and Gender Inequality Index (GII). The CHE was measured by multiplying the CHE as a percentage (%) of the Gross Domestic Product (GDP) and the GDP per capita (current US$).2,41 We included the CHE in absolute terms. In this case, for the interpretation of the CHE measure we consider 1000 dollars, representing how much an increase of 1000 USD would increase the outcome. GII: this variable relates to reproductive health (maternal mortality ratio and adolescent fertility), empowerment (share of parliament seat and secondary/higher education attainment) and labour market (participation in the workforce) dimensions.42 The GII varies from 0 to 1, where 0 (the best scenario) indicates that women and men fare equally, and 1 (the worst scenario) means that men or women fare poorly compared to each other in all dimensions. More detailed information on independent factors definition and classification is available in Table 1.Table 1Independent factors (individual- and country-level factors), definition and classification. Independent factorsDefinitionClassificationIndividual-levelMarital statusMarital statusUnmarried sexually activeMarried/in a unionWoman's ageCurrent woman's age in years15–1920–3435–49Woman's educationWoman's schooling levelNonePrimary/elementary schoolSecondaryHigherWealth indexWealth index constructed based on a principal component analysis including household characteristics and ownership of selected assets,40 dividing the score into quintiles, the first quintile represented the poorest $20\%$Poorest2nd3rd4thWealthiestArea of residenceArea of residenceUrbanRuralNumber of childrenNumber of children currently alive for each woman0123 or moreCountry-levelCHEThe current health expenditure (CHE) was measured by multiplying the CHE as a percentage (%) of the Gross Domestic Product (GDP), adjusted for purchasing power parity. We included the CHE in absolute terms. For interpretation purposes, in the CHE we consider not one, but 1000 dollars. The CHE%GDP data were accessed in the Global Health Expenditure Database <https://apps.who.int/nha/database/Select/Indicators/en>. Data for GDP per capita were available at the World Bank website data <https://data.worldbank.org/indicator/NY.GDP.PCAP.CD?view=chart>.CHE in dollar value. GIIGII: this variable relates to reproductive health, empowerment and labour market dimensions. The GII varies from 0 to 1, where 0 (the best scenario) indicates that women and men fare equally, and 1 (the worst scenario) means that men or women fare poorly compared to each other in all dimensions. In this study, we divided the GII in tertiles, where (1st: GII = [0.291, 0.433], 2nd: GII = [0.449, 0.477], and 3rd: GII = [0.479, 0.776]), and the 1st tertile representing the most equitable group of countries and the 3rd the most unequal countries. For the Gender Inequality Index (GII) data were available at Human Development Data Center <http://hdr.undp.org/en/data>Least2ndHighest ## Statistical analyses Descriptive analyses were performed to present the distribution of individual and contextual factors. DFPSm coverage and $95\%$ confidence intervals ($95\%$ CI) for each country were described, and the median coverage for all LAC countries included. We also presented relative frequencies of the individual-level factors and described the DFPSm coverage and $95\%$ CI according to these characteristics. Crude and adjusted multilevel logistic regression with two levels were conducted: individuals (level 1) and country (level 2). Odds ratios (OR) and the corresponding $95\%$ CIs were calculated, and the intraclass correlation coefficient (ICC) was estimated. The ICC varies from 0 to 1 and quantifies the proportion of the observed variability of the outcome that is explained by the country effect.43 In other words, the ICC in this study represents the percentage of the DFPSm variability explained by the difference between countries (level 2). We performed the null model, a multilevel model with no predictor, allowing us to assess whether there was significant between-country variation in the DFPSm pattern. The unadjusted model, a single-level univariate model, was also estimated. In addition, we performed three other models. Model 1 was fitted only with individual-level factors. Model 2 was fitted only with country-level factors, and Model 3 was fitted with individual- and country-level factors. We used the Akaike information criterion (AIC) and Bayesian information criterion (BIC) to assess the model's goodness of fit. Smaller values of AIC and BIC represent better fitted models. Statistical analyses were conducted using Stata 17.0 software (StataCorp LLC, College Station, TX, USA) and accounted for survey sample weights. ## Role of the funding source This work was supported by the Bill and Melinda Gates Foundation [OPP1199234] and [INV-010051]. The Associação Brasileira de Saúde Coletiva (ABRASCO) also acts as a sponsor of the study. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. ## Results We included data of 109,149 sexually active women (unweighted number) from 14 LAC countries (Table 2). Women included in the analyses were mainly married/in a union ($85.0\%$), aged 20–34 years ($52.2\%$), from secondary or higher educational levels ($66.6\%$), and residents in urban areas ($64.0\%$). Around $65\%$ of the women had two or more children (Table 3).Table 2Overall description of the 14 Latin American and Caribbean countries, survey characteristics, and the sample included in this study. CountrySurvey yearSurvey typeIncome levelFemale population aged 15–49CHE (thousands)GIINumber of womenaBelize2015MICSUpper-middle99,23728.30.4232549Colombia2015DHSUpper-middle12,952,32846.50.43321,551Costa Rica2018MICSUpper-middle1,309,29991.10.2914286Cuba2019MICSUpper-middle2,574,011101.20.3046448Dominican Republic2014MICSUpper-middle2,677,15038.00.47717,040El Salvador2014MICSLower-middle1,764,00127.60.4007127Guatemala2014DHSLower-middle4,167,45922.30.51111,719Guyana2014MICSLower-middle199,83421.30.4792825Haiti2016DHSLow2,874,3916.30.7766521Honduras2011DHSLower-middle2,196,08718.20.46911,766Mexico2015MICSUpper-middle33,378,76155.00.3477287Paraguay2016MICSUpper-middle1,766,32635.60.4524485Suriname2018MICSUpper-middle146,65354.80.4493862Trinidad and Tobago2011MICSHigh364,20689.90.3591683CHE = (GDP ∗ CHE%GDP), where: CHE–Current Health Expenditure; GDP–Gross Domestic Product per capita (current US$); CHE % GDP–CHE as percentage (%) of GDP.GII–Gender Inequality Index.aUnweighted number of sexually active women analysed in each survey. Table 3Individual-level factors and coverage ($95\%$ Confidence Interval, CI) of demand for family planning satisfied by modern contraceptive methods (DFPSm).Individual-level factorsDistribution (%)DFPSm % ($95\%$ CI)Marital status Unmarried sexually active15.063.1 (55.1–71.2) Married/in a union85.071.3 (63.8–78.9)Woman's age (years) 15–197.655.6 (45.9–65.4) 20–3452.269.7 (62.8–76.7) 35–4940.273.6 (65.8–81.4)Woman's education None5.163.8 (55.6–72.0) Primary/elementary school28.369.9 (61.9–77.8) Secondary46.570.4 (62.9–78.0) Higher20.171.9 (65.4–78.4)Wealth index Poorest17.962.8 (53.5–72.2) 2nd19.668.6 (60.4–76.7) 3rd20.771.2 (64.1–78.3) 4th21.372.7 (65.3–80.2) Wealthiest20.574.3 (66.8–81.8)Area of residence Urban64.071.8 (64.6–79.0) Rural36.067.9 (60.1–75.8)Number of children 012.556.3 (47.4–65.2) 122.368.6 (61.6–75.5) 227.974.0 (67.4–80.5) 3 or more37.373.9 (65.5–82.3) All countries, except for Haiti (low-income) and Trinidad and Tobago (high-income), were from lower- or upper-middle-income levels (Table 2). Mexico has the largest female population aged 15–49 [33,378,761], followed by Colombia [12,952,328], while Belize [99,237] was the country with the smallest female population. Regarding the contextual factor's description, Cuba [101,160], Costa Rica [91,090], and Trinidad and Tobago [89,879] presented the highest CHE, while Haiti had the lowest value [6,323] (Table 2). Regarding the GII, Haiti shows the worst scenario in terms of gender inequality, showing that men or women fare poorly compared to each other in all dimensions (GII: 0.776), while Costa Rica presented the best scenario (GII: 0.291), indicating that in this country women and men fare more equally. The coverage of DFPSm in the 14 Latin American and Caribbean countries included in the analyses ranged from $41.8\%$ ($95\%$ CI: 40.2–43.5) in Haiti to $85.6\%$ ($95\%$ CI: 84.9–86.3) in Colombia, with an overall median coverage of $77.8\%$ (Fig. 1). Trinidad and Tobago, the only high-income country, presented coverage almost 17 percentage points below the median value. Although there were no coverage differences between Colombia and Cuba, these countries have shown a significantly higher coverage compared to the other ones—14.9 and 14.4 percentage points higher, respectively, compared to the median coverage of other countries. On the other hand, Haiti has presented significantly lower DFPSm coverage than all other LAC countries analysed (Fig. 1). Among the individual-level subgroups, married women or in a union, aged 35–49 years, with higher education, the wealthiest, living in urban areas, with more than one child presented the highest coverages of DFPSm (Table 3).Fig. 1Coverage of demand for family planning satisfied by modern contraceptive methods (DFPSm) (%) in 14 Latin American and Caribbean countries. The orange color indicates below median value (77.8), and the blue color represents above median value (77.8). The null model ICC showed a value of $11.8\%$, indicating that $11.8\%$ of the variation in DFPSm was attributable to differences across the countries (Table 4). Table 4 also showed the unadjusted effects of the individual and country-level variables on DFPSm derived from the multilevel logistic regression model. According to the results, married and educated women, older than 20 years old, in the highest level of wealth, from urban areas, and with more than one child are significantly more likely to present DFPSm. Table 4Results from multilevel logistic regression analysis investigating the association between individual- and country-level factors and demand for family planning satisfied by modern contraceptive methods (DFPSm) among women in Latin American and Caribbean countries. VariablesNull modelUnadjusted model OR ($95\%$ CI)Model 1 OR ($95\%$ CI)Model 2 OR ($95\%$ CI)Model 3 OR ($95\%$ CI)Individual-levelMarital status Unmarried sexually active111 Married/in a union1.51 (1.28–1.77)1.04 (0.85–1.28)1.05 (0.85–1.28)Woman's age (years) 15–19111 20–341.95 (1.68–2.26)1.27 (1.07–1.50)1.26 (1.07–1.50) 35–492.39 (1.98–2.89)1.27 (1.07–1.51)1.26 (1.06–1.50)Woman's education None111 Primary/elementary school1.35 (1.12–1.63)1.32 (1.12–1.55)1.31 (1.11–1.54) Secondary1.39 (1.06–1.81)1.43 (1.16–1.76)1.42 (1.16–1.74) Higher1.50 (1.13–1.99)1.53 (1.28–1.82)1.51 (1.27–1.81)Wealth index Poorest111 2nd1.32 (1.21–1.45)1.33 (1.23–1.43)1.33 (1.23–1.43) 3rd1.51 (1.26–1.81)1.51 (1.31–1.74)1.51 (1.31–1.75) 4th1.65 (1.25–2.17)1.65 (1.32–2.06)1.65 (1.32–2.07) Wealthiest1.80 (1.27–2.55)1.80 (1.35–2.39)1.81 (1.36–2.41)Area of residence Urban111 Rural0.82 (0.68–0.99)0.95 (0.89–1.03)0.96 (0.89–1.03)Number of children 0111 11.79 (1.51–2.11)1.78 (1.43–2.23)1.79 (1.43–2.24) 22.38 (1.88–3.02)2.35 (1.72–3.21)2.35 (1.72–3.22) 3 or more2.37 (1.79–3.14)2.65 (1.82–3.85)2.66 (1.82–3.88)Country-levelCHE1.01 (1.00–1.02)1.00 (0.99–1.01)1.00 (0.99–1.02)GII Least111 2nd0.83 (0.43–1.62)0.83 (0.37–1.88)0.83 (0.37–1.86) Highest0.31 (0.17–0.59)0.32 (0.13–0.76)0.32 (0.14–0.74)ICC (%)11.812.16.66.6Model goodness of fit Log pseudolikelihood−5677−5525−5672−5521 AIC11,35711,07611,35411,067 BIC11,37611,20111,40211,192Na109,103109,103109,103109,103Null model: a multilevel model with no predictor. Unadjusted model: a single-level univariate model. Model 1: a single-level multivariate adjusted model. Model 2: a multilevel-model with only country-level predictors. Model 3: the fully-adjusted multi-level model. CHE = (GDP ∗ CHE%GDP), where: CHE–Current Health Expenditure; GDP–Gross Domestic Product per capita (current US$); CHE % GDP–CHE as percentage (%) of GDP.GII–Gender Inequality Index; OR–Odds ratios; CI–Confidence Interval. Bold letter indicate $p \leq 0.05.$aUnweighted sample size. About the country-level crude effects, we found that women living in high gender inequality countries significantly have an odd of DFPSm $70\%$ lower compared to those living in countries with low gender inequality. CHE level of countries was not associated with DFPSm (Table 4). Adjusted results from Model 1, which only includes individual-level variables, showed that the significant effect of woman's age and education, wealth index, and the number of children remained after adjustment. Women aged 20–34 and those 35–49 had $27\%$ higher odds of having DFPSm. Women with three or more children had 2.65 ($95\%$ CI: 1.82–3.85) times higher odds of DFPSm than women without children. The association of marital status and area of residence with DFPSm disappeared after adjustment. Model 2 in Table 4 shows the effect of the country-level variables where, again, the only variable associated with DFPSm was the GII, in which women with the highest GII tertile were less likely (OR: 0.32, $95\%$ CI: 0.13–0.76) to have demand for family planning satisfied by modern contraceptive methods than those living in countries from the lowest GII tertile. Results from Model 3, which include both individual and country-level predictors, show that the association of residing in countries with high gender inequalities was virtually unaltered by adjustment. After adjustment for individual-level variables, the odds of DFPSm for women from countries belonging to the highest tertile of GII remained lower when compared to those women from countries in the lowest tertile. According to the AIC criterion, Model 3 fits better than the others. Finally, the results of the ICC of the adjusted models showed that individual and country-level variables contributed to explain the differences between countries. In the final model, the ICC was $6.6\%$, indicating that around $7\%$ of the variation in DFPSm could still be attributable to differences across the countries. ## Discussion To the best of our knowledge, this is the first multi-country research investigating the association between Gender Inequality and Health Expenditure with DFPSm in sexually active women from LAC countries. We found that the country-level factor GII, beyond individual-level factors, plays a relevant role in explaining the variations in the coverage of DFPSm in sexually active women from LAC countries. The coverage of DFPSm varied greatly among the LAC countries, with a median value of $77.8\%$, ranging from Haiti with the lowest median coverage of DFPSm to Colombia with the highest value. DFPSm was directly associated with woman's education, wealth index, and the number of children. In the LAC region, many countries provide contraceptive methods free of charge in public facilities, despite the still present inequalities.7 Furthermore, family planning policies are directed to protect and promote women's rights, guarantee gender equality, and other important issues related to sexual and reproductive health.44 More detailed information on this topic and the respective search strategies are available in the Supplementary Material (Files 2, 3, and 4). Among sexually active women from LAC countries, the coverage of DFPSm varied from $43.4\%$ in Bolivia to $89.5\%$ in Cuba, with coverage inequalities in which the poorest, youngest, less educated women and those living in rural areas presented the lowest DFPSm.3 Initiatives that may contribute to reducing inequalities are present in the LAC region, such as Conditional and Unconditional Cash Transfer Programs. However, findings about the direct impacts of these programs on the use of modern contraception are inconsistent.45 Improving individual-level factors may contribute to family planning advances beyond health and social changes. The importance of country-level socioeconomic determinants is also mentioned in the scientific literature, despite the lack of consensus on which factors are more relevant for different health-related outcomes in each region and country.19,46,47 This study added that GII, a measure of disadvantages affecting women, which has empowerment, reproductive health, and labour market as its dimensions, plays a relevant role in the coverage of DFPSm.48 Regarding empowerment, a study found an association between different empowerment dimensions and the use of contraception.15 Empowerment measures were developed, validated, and expanded. It is currently possible to apply these measures to different low-and middle-income countries.49,50 Women's empowerment is frequently related to family planning and the possible reproductive health benefits for women.20,21 In addition, one possible hypothesis is that participation in the labour market and reproductive autonomy mediate the relationship between gender inequality and DFPSm. Furthermore, higher scores of GII were related to other health outcomes, such as lower life expectancy and healthy life expectancy, as well as increased years of life lost, morbidity and years lived with disability.19 CHE is a complex measure included in this analysis. So, it is important to be cautious when interpreting its country-level effects. The result that only GII was significant does not mean that health investments are irrelevant to family planning. Besides, sexual and reproductive health investments may be shared between maternal health and family planning. Although this study focused on GII and CHE, many other contextual factors are available on databases around the world, such as Gross National Income (GNI), poverty headcount ratio at national poverty lines, population density, and density of health centres, which are directly related in the other indicators or that there is a lack of data for many countries. In view of comparison purposes, investigating countries from other regions may also be important to understand the country-level variations in DFPSm worldwide. Sexual acceptability of contraception and its broad range of related macro, relational, and individual factors may contribute to contraceptive use, implying that family planning includes a range of factors that are not always captured in common models.51 One of the strengths of this study is including information for 14 LAC countries. In addition, the demand for family planning satisfied measure shows advantages over the measure of the prevalence of contraceptive use since it includes in its denominator only women who are in need of contraception.3,30,52 Many contraceptive methods are available for couples, and modern methods are more effective and less prone to failure for all age subgroups.38,53 Furthermore, modern contraceptive methods are key to preventing unplanned pregnancies, which may impact women, children, and families lives. Unplanned pregnancies are associated with adverse outcomes such as less education access for adolescents, unsafe abortion, late antenatal care, and its possible consequences on women's and child's health and increased health system expenditures.54,55 Although DFPSm presents the advantage of including women who need contraception in its denominator, there are limitations to this indicator. Problems are mainly related to the lack of information on the coital frequency, especially among married women (the subgroup on which most studies about modern contraception focus), the lack of information on the contextual scenario, and the type of contraception women have more access to in each country.3,30,52 This study did not include a factor about family planning availability, despite its possible influences on family planning outcomes. One limitation identified during the conduction of this study was the lack of contextual data availability on expenditure on reproductive health and family planning for LAC countries. In June 2022, when the final version of the data analysis was performed, only two LAC countries included in this analysis (Guyana and Haiti) presented data on current health expenditure on reproductive health and specifically on contraceptive management (family planning) available in the Global Health Expenditure Database (https://apps.who.int/nha/database/Select/Indicators/en). Regarding the description of the CHE results, the data order may vary according to the unit of measurement used. Furthermore, efforts are needed to improve the current indicators by performing more contextualised modern contraception comparisons worldwide. One of the limitations of performing this type of analysis is that not all countries have data on family planning at the country level, making it difficult to carry out this type of study. Furthermore, this study included data from a pre-COVID-19 pandemic scenario. Currently, there is evidence pointing out that changes in family planning occurred, such as specific disruptions in contraceptive use, resulting in unintended pregnancies.56,57 The findings of this study need to be interpreted considering this new reality, given its possible implications, contributing to planning different strategies. Strengthening the databases with standardised country-level data on access to family planning, health care facilities, and family planning policies, among others, may facilitate future comparisons and possible generalisations for similar realities. ## Conclusion The country-level factor GII, beyond individual-level factors, plays a relevant role in explaining the variations in the coverage of DFPSm in sexually active women from LAC countries. Less gender inequality at the country level may play a relevant role in the coverage of DFPSm. In addition, individual factors such as woman's age, education, wealth index, and the number of children also contribute to modern contraceptive coverage. In this sense, planning actions, including macro-level approaches focusing on reducing gender disparities and considering individual-level factors, may be essential to guarantee reproductive health to the population of women in need of contraception. ## Contributors LRM conceived the idea of the present article, analysed and interpreted the data, and wrote the manuscript. CB, FE, and LPVR contributed to the conceptualisation, methodology, formal analysis, and substantial manuscript revision. BECV, AB, ARV, FH, RGPL, AJDB, and MFS substantially revised the manuscript and contributed to data interpretation. FCW conceived the idea of the present article, supervised the project, and participated in all subsequent steps. All authors commented on the draft manuscript and approved its final version. The authors do not report any conflicting interests. ## Data sharing statement The data are anonymised and geographically scrambled to ensure confidentiality and are publicly available through the agencies’ websites. ## Declaration of interests The authors declare no conflict of interest. RGPL, who is a staff member of the Pan American Health Organization, hold sole responsibility for the views expressed in their texts, which may not necessarily reflect the opinion or policy of the Pan American Health Organization. ## Supplementary data Supplementary Material ## References 1. Guzmán J., Rodríguez J., Martínez J., Contreras J., González D.. The demography of Latin America and the Caribbean since 1950. Popul (2006) 61 519-620 2. World Bank Open Data. Free and open access to global development data. (2022) 3. Ewerling F., Victora C.G., Raj A., Coll C.V.N., Hellwig F., Barros A.J.D.. Demand for family planning satisfied with modern methods among sexually active women in low- and middle-income countries: who is lagging behind?. Reprod Health (2018) 15 42. DOI: 10.1186/s12978-018-0483-x 4. Hellwig F., Barros A.J.. Learning from success cases: ecological analysis of pathways to universal access to reproductive health care in low- and middle-income countries. Gates Open Res (2022) 6 59. DOI: 10.12688/gatesopenres.13570.1 5. Haakenstad A., Angelino O., Irvine C.M.S.. Measuring contraceptive method mix, prevalence, and demand satisfied by age and marital status in 204 countries and territories, 1970–2019: a systematic analysis for the Global Burden of Disease Study 2019. Lancet (2022) 400 295-327. DOI: 10.1016/S0140-6736(22)00936-9 6. Ponce de Leon R.G., Ewerling F., Serruya S.J.. Contraceptive use in Latin America and the Caribbean with a focus on long-acting reversible contraceptives: prevalence and inequalities in 23 countries. Lancet Glob Health (2019) 7 e227-e235. DOI: 10.1016/S2214-109X(18)30481-9 7. Fagan T., Dutta A., Rosen J., Olivetti A., Klein K.. Family planning in the context of Latin America's universal health coverage agenda. Glob Health Sci Pract (2017) 5 382-398. DOI: 10.9745/GHSP-D-17-00057 8. Ali M., Miller K., Gómez Ponce de Leon R.F.. Family planning and Zika virus: need for renewed and cohesive efforts to ensure availability of intrauterine contraception in Latin America and the Caribbean. Eur J Contracept Reprod Health Care (2017) 22 102-106. DOI: 10.1080/13625187.2017.1288902 9. 9United NationsUnited nations general assembly resolution A/Res/70/1: transforming our world: the 2030 agenda for sustainable development. Resolution adopted by the general assembly on 25 september 20152015. (2015) 10. Ahinkorah B.O.. Predictors of modern contraceptive use among adolescent girls and young women in Sub-Saharan Africa: a mixed effects multilevel analysis of data from 29 demographic and health surveys. Contracept Reprod Med (2020) 5 32. DOI: 10.1186/s40834-020-00138-1 11. Ahinkorah B.O., Budu E., Aboagye R.G.. Factors associated with modern contraceptive use among women with no fertility intention in Sub-Saharan Africa: evidence from cross-sectional surveys of 29 countries. Contracept Reprod Med (2021) 6 1-13. DOI: 10.1186/s40834-021-00165-6 12. Alo O.D., Daini B.O., Omisile O.K., Ubah E.J., Adelusi O.E., Idoko-Asuelimhen O.. Factors influencing the use of modern contraceptive in Nigeria: a multilevel logistic analysis using linked data from performance monitoring and accountability 2020. BMC Womens Health (2020) 20 1-9. DOI: 10.1186/s12905-020-01059-6 13. Borges A.L.V., Duarte L.S., Lay A.A.R., Fujimori E.. Individual and context correlates of the oral pill and condom use among Brazilian female adolescents. BMC Womens Health (2021) 21 307. DOI: 10.1186/s12905-021-01447-6 14. Zegeye B., Ahinkorah B.O., Idriss-Wheeler D., Olorunsaiye C.Z., Adjei N.K., Yaya S.. Modern contraceptive utilization and its associated factors among married women in Senegal: a multilevel analysis. BMC Public Health (2021) 21 231. DOI: 10.1186/s12889-021-10252-7 15. Yaya S., Uthman O.A., Ekholuenetale M., Bishwajit G.. Women empowerment as an enabling factor of contraceptive use in sub-Saharan Africa: a multilevel analysis of cross-sectional surveys of 32 countries. Reprod Health (2018) 15 214. DOI: 10.1186/s12978-018-0658-5 16. Kamuyango A., Yu T., Ao C.-K.. The public-sector family planning program impact scores in association with long-acting reversible contraceptive use among young women in 22 Sub-Saharan African countries: a pooled multi-level analysis. Contraception (2022) 108 44-49. DOI: 10.1016/j.contraception.2021.12.013 17. Nyarko S.H., Kabir R.. Spatial variations and socioeconomic determinants of modern contraceptive use in Ghana: a Bayesian multilevel analysis. PLoS One (2020) 15 18. Nurjaeni N., Sawangdee Y., Pattaravanich U., Holumyong C., Chamratrithirong A.. The role of structural and process quality of family planning care in modern contraceptive use in Indonesia: a multilevel analysis. BMC Public Health (2021) 21 1790. DOI: 10.1186/s12889-021-11858-7 19. Veas C., Crispi F., Cuadrado C.. Association between gender inequality and population-level health outcomes: panel data analysis of organization for economic co-operation and development (OECD) countries. eClinicalMedicine (2021) 39. DOI: 10.1016/j.eclinm.2021.101051 20. James-Hawkins L., Peters C., VanderEnde K., Bardin L., Yount K.M.. Women's agency and its relationship to current contraceptive use in lower- and middle-income countries: a systematic review of the literature. Glob Public Health (2018) 13 843-858. DOI: 10.1080/17441692.2016.1239270 21. Prata N., Fraser A., Huchko M.J.. Women's empowerment and family planning: a review of the literature. J Biosoc Sci (2017) 49 713-743. DOI: 10.1017/S0021932016000663 22. Gerdtham U.-G., Jönsson B.. Handbook of health economics (2000) 11-53. DOI: 10.1016/S1574-0064(00)80160-2 23. Musgrove P., Zeramdini R., Carrin G.. Basic patterns in national health expenditure. Bull World Health Organ (2002) 80 134-142. PMID: 11953792 24. Ekholuenetale M., Barrow A.. Inequalities in out-of-pocket health expenditure among women of reproductive age: after-effects of national health insurance scheme initiation in Ghana. J Egypt Public Health Assoc (2021) 96 6. DOI: 10.1186/s42506-020-00064-9 25. Azim M.R., Sultana N., Tulon K.M.N.I., Islam M.S., Hamid S.A.. Analyzing family planning expenditure in Bangladesh. J Glob Health Econ Policy (2022) 2. DOI: 10.52872/001c.34245 26. Weinberger M., Bellows N., Stover J.. Estimating private sector out-of-pocket expenditures on family planning commodities in low-and-middle-income countries. BMJ Glob Health (2021) 6 e004635. DOI: 10.1136/bmjgh-2020-004635 27. Rannan-Eliya R.P., Berman P., Eltigani E.E., Somanathan A., Sumathiratne V.. Expenditures for reproductive health services in Egypt and Sri Lanka. (2000) 28. Komasawa M., Yuasa M., Shirayama Y., Sato M., Komasawa Y., Alouri M.. Demand for family planning satisfied with modern methods and its associated factors among married women of reproductive age in rural Jordan: a cross-sectional study. PLoS One (2020) 15 e0230421. DOI: 10.1371/journal.pone.0230421 29. Corsi D.J., Neuman M., Finlay J.E., Subramanian S.. Demographic and health surveys: a profile. Int J Epidemiol (2012) 41 1602-1613. DOI: 10.1093/ije/dys184 30. Bradley S.E.K., Croft T.N., Fishel J.D., Wetoff C.F.. (2012) 31. Murray C., Newby H.. Data resource profile: united nations children's fund (UNICEF). Int J Epidemiol (2012) 41 1595-1601. DOI: 10.1093/ije/dys185 32. Costa J.C., Mujica O.J., Gatica-Domínguez G.. Inequalities in the health, nutrition, and wellbeing of Afrodescendant women and children: a cross-sectional analysis of ten Latin American and Caribbean countries. Lancet Reg Health Am (2022) 15. DOI: 10.1016/j.lana.2022.100345 33. Mesenburg M.A., Restrepo-Mendez M.C., Amigo H.. Ethnic group inequalities in coverage with reproductive, maternal and child health interventions: cross-sectional analyses of national surveys in 16 Latin American and Caribbean countries. Lancet Glob Health (2018) 6 e902-e913. DOI: 10.1016/S2214-109X(18)30300-0 34. Flores-Quispe M.D.P., Restrepo-Méndez M.C., Maia M.F.S., Ferreira L.Z., Wehrmeister F.C.. Trends in socioeconomic inequalities in stunting prevalence in Latin America and the Caribbean countries: differences between quintiles and deciles. Int J Equity Health (2019) 18 1-10. DOI: 10.1186/s12939-019-1046-7 35. Restrepo-Méndez M.C., Barros A.J.D., Requejo J.. Progress in reducing inequalities in reproductive, maternal, newborn,’and child health in Latin America and the Caribbean: an unfinished agenda. Rev Panam Salud Publica (2015) 38 9-16. PMID: 26506316 36. Gatica-Domínguez G., Mesenburg M.A., Barros A.J.D., Victora C.G.. Ethnic inequalities in child stunting and feeding practices: results from surveys in thirteen countries from Latin America. Int J Equity Health (2020) 19 1-13. DOI: 10.1186/s12939-020-01165-9 37. Colomé-Hidalgo M., Campos J.D., de Miguel Á.G.. Exploring wealth-related inequalities in maternal and child health coverage in Latin America and the Caribbean. BMC Public Health (2021) 21 115. DOI: 10.1186/s12889-020-10127-3 38. Hubacher D., Trussell J.. A definition of modern contraceptive methods. Contraception (2015) 92 420-421. DOI: 10.1016/j.contraception.2015.08.008 39. Demand satisfied: understanding the sustainable development goal indicator for family planning.. (2022) 40. Filmer D., Pritchett L.H.. Estimating wealth effects without expenditure data—or tears: an application to educational enrollments in states of India. Demography (2001) 38 115-132. DOI: 10.1353/dem.2001.0003 41. 41World Health Organization (WHO)Global health expenditure database2014World Health OrganizationAvailable from:https://apps.who.int/nha/database/Select/Indicators/en. (2014) 42. 42United Nations Development Programme (UNDP)Gender inequality index (GII)2022United Nations Development ProgrammeAvailable from:http://hdr.undp.org/en/data. (2022) 43. Sommet N., Morselli D.. Keep calm and learn multilevel logistic modeling: a simplified three-step procedure using Stata, R, mplus, and SPSS. Int Rev Soc Psychol (2017) 30 203-218. DOI: 10.5334/irsp.90 44. 44ECLAC. United NationsGender equality plans in Latin America and the Caribbean: road maps for development2017United NationsSantiagoAvailable from:https://www.cepal.org/sites/default/files/events/files/gender_equality_plans_in_latin_america_and_the_caribbean._road_maps_for_development.pdf. (2017) 45. Khan M.E., Hazra A., Kant A., Ali M.. Conditional and unconditional cash transfers to improve use of contraception in low and middle income countries: a systematic review. Stud Fam Plann (2016) 47 371-383. DOI: 10.1111/sifp.12004 46. Ataey A., Jafarvand E., Adham D., Moradi-Asl E.. 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. DOI: 10.3961/jpmph.19.100 47. Swift R.. The relationship between health and GDP in OECD countries in the very long run. Health Econ (2011) 20 306-322. DOI: 10.1002/hec.1590 48. Gaye A., Klugman J., Kovacevic M., Twigg S., Zambrano E.. Measuring key disparities in human development: the gender inequality index. United Nations Dev Program (2010) 46 1-37 49. Ewerling F., Lynch J.W., Victora C.G., Van Eerdewijk A., Tyszler M., Barros A.J.D.. The SWPER index for women's empowerment in Africa: development and validation of an index based on survey data. Lancet Glob Health (2017) 5 e916-e923. DOI: 10.1016/S2214-109X(17)30292-9 50. Ewerling F., Raj A., Victora C.G., Hellwig F., Coll C.V., Barros A.J.. SWPER Global: a survey-based women's empowerment index expanded from Africa to all low- and middle-income countries. J Glob Health (2020) 10. DOI: 10.7189/jogh.10.020434 51. Higgins J.A., Smith N.K.. The sexual acceptability of contraception: reviewing the literature and building a new concept. J Sex Res (2016) 53 417-456. DOI: 10.1080/00224499.2015.1134425 52. Barros A.J.D., Boerma T., Hosseinpoor A.R., Restrepo-Méndez M.C., Wong K.L.M., Victora C.G.. Estimating family planning coverage from contraceptive prevalence using national household surveys. Glob Health Action (2015) 8. DOI: 10.3402/gha.v8.29735 53. Bradley S.E.K., Polis C.B., Bankole A., Croft T.. Global contraceptive failure rates: who is most at risk?. Stud Fam Plann (2019) 50 3-24. DOI: 10.1111/sifp.12085 54. Gipson J.D., Koenig M.A., Hindin M.J.. The effects of unintended pregnancy on infant, child, and parental health: a review of the literature. Stud Fam Plann (2008) 39 18-38. DOI: 10.1111/j.1728-4465.2008.00148.x 55. Le H.H., Connolly M.P., Bahamondes L., Cecatti J.G., Yu J., Hu H.X.. The burden of unintended pregnancies in Brazil: a social and public health system cost analysis. Int J Womens Health (2014) 6 663-670. DOI: 10.2147/IJWH.S61543 56. Dema E., Gibbs J., Clifton S.. Initial impacts of the COVID-19 pandemic on sexual and reproductive health service use and unmet need in Britain: findings from a quasi-representative survey (Natsal-COVID). Lancet Public Health (2022) 7 e36-e47. DOI: 10.1016/S2468-2667(21)00253-X 57. Impact of COVID-19 on family planning: what we know one year into the pandemic. (2021)
--- title: Maternal nano-titanium dioxide inhalation alters fetoplacental outcomes in a sexually dimorphic manner authors: - Julie A. Griffith - Allison Dunn - Evan DeVallance - Kallie J. Schafner - Kevin J. Engles - Thomas P. Batchelor - William T. Goldsmith - Kimberley Wix - Salik Hussain - Elizabeth C. Bowdridge - Timothy R. Nurkiewicz journal: Frontiers in Toxicology year: 2023 pmcid: PMC10025460 doi: 10.3389/ftox.2023.1096173 license: CC BY 4.0 --- # Maternal nano-titanium dioxide inhalation alters fetoplacental outcomes in a sexually dimorphic manner ## Abstract The placenta plays a critical role in nutrient-waste exchange between the maternal and fetal circulations, thus functioning as an interface that profoundly impacts fetal growth and development. The placenta has long been considered an asexual organ, but, due to its embryonic origin it shares the same sex as the fetus. Exposures to toxicant such as diesel exhaust, have been shown to result in sexually dimorphic outcomes like decreased placental mass in exposed females. Therefore, we hypothesize that maternal nano-TiO2 inhalation exposure during gestation alters placental hemodynamics in a sexually dimorphic manner. Pregnant Sprague-Dawley rats were exposed from gestational day 10–19 to nano-TiO2 aerosols (12.17 ± 1.69 mg/m3) or filtered air (sham-control). Dams were euthanized on GD20, and fetal tissue was collected based on fetal sex: whole placentas, placental junctional zone (JZ), and placental labyrinth zone (LZ). Fetal mass, placental mass, and placental zone percent areas were assessed for sex-based differences. Exposed fetal females were significantly smaller compared to their exposed male counterparts (2.65 ± 0.03 g vs 2.78 ± 0.04 g). Nano-TiO2 exposed fetal females had a significantly decreased percent junctional zone area compared to the sham-control females (24.37 ± $1.30\%$ vs 30.39 ± $1.54\%$). The percent labyrinth zone area was significantly increased for nano-TiO2 females compared to sham-control females (75.63 ± $1.30\%$ vs 69.61 ± $1.54\%$). Placental flow and hemodynamics were assessed with a variety of vasoactive substances. It was found that nano-TiO2 exposed fetal females only had a significant decrease in outflow pressure in the presence of the thromboxane (TXA2) mimetic, U46619, compared to sham-control fetal females (3.97 ± 1.30 mm Hg vs 9.10 ± 1.07 mm Hg) and nano-TiO2 fetal males (9.96 ± 0.66 mm Hg). Maternal nano-TiO2 inhalation exposure has a greater effect on fetal female mass, placental zone mass and area, and adversely impacts placental vasoreactivity. This may influence the female growth and development later in life, future studies need to further study the impact of maternal nano-TiO2 inhalation exposure on zone specific mechanisms. ## 1 Introduction Adverse intrauterine environments have been shown to influence fetal development in a sex-dependent way, as was classically shown in work examining the Dutch famine at the end of WWII (Gabory et al., 2013). Undernutrition in mid to late gestation caused a sex-based difference, with increased placental thickness in female fetuses compared to males (Roseboom et al., 2011). It was speculated this may be the female placenta’s attempt to compensate for reduced growth by deeper spiral artery invasion (Roseboom et al., 2011). Diseases, such as preeclampsia (PE), can also result in hindered fetal development in a sexually dimorphic manner. It has been suggested that the normal growth of males in PE pregnancies is due to an adaptation in peripheral microvascular tone by the maternal circulation to maintain fetal-placental blood flow, despite disrupted hemodynamics and placental insufficiency (Stark et al., 2006). Female fetuses from a PE pregnancy do not demonstrate altered microvascular function, due to a lack of compensatory peripheral vascular responses, have reduced uteroplacental blood flow, and thusly decreased placental hemodynamics that ultimately decrease fetal female growth and development (Stark et al., 2009). In addition to nutritional or vascular derived disease states affecting fetal growth and development, environmental exposures during gestation may also result in compromised fetal health. Exposure to certain toxicants can modify placental and fetal growth in a sexually dimorphic manner (Miller et al., 2020). A rat model of inhaled ozone found that fetal females from this study demonstrated adaptive mechanisms to increase nutrient availability to support fetal development, while males did not (Miller et al., 2020). In pregnant mice exposed to diesel exhaust, female offspring in the exposed group had decreased placental mass and crown-to-rump length (Behlen et al., 2021). Exposed fetal females demonstrated increased placental decidua area, lacunae areas, and lipid metabolism signaling (Behlen et al., 2021). While not an inhalation exposure, arsenic exposure via drinking water in humans also produces a sexually dimorphic effect, with female placentas having increased levels of the aquaglyceroporin transporter (Winterbottom et al., 2017). This transporter may lead to increased movement of arsenic across the female placenta and elicit the expression of a subset of genes that are female-specific in response to arsenic exposure (Winterbottom et al., 2017). Maternal inhalation of nano-titanium dioxide, a nanomaterial used in building materials and water/air filters (Bowdridge et al., 2019), during gestation has caused decreased female pup mass (Griffith et al., 2022) and decreased male: female sex ratios in early and mid-gestation (Garner et al., 2022b). This exposure paradigm our laboratory has utilized for nano-titanium dioxide (nano-TiO2) inhalation exposure is a model for pregnant women working in an occupational setting would experience. This encompasses the time periods when women may not realize they are pregnant through late gestation (Griffith et al., 2022). These studies indicate that maternal environmental exposures affect offspring in a sexually dimorphic manner that appears to be paradigm specific. Adaptations in vascular reactivity, to estrogen stimulatory compounds such as prostacyclin (Sobrino et al., 2010), may be part of the sexual dimorphic responses seen in detrimental in utero environments such as improper nutrition, disease, or toxicant exposure. Prostacyclin (PGI2) and thromboxane (TXA2), potent vasodilator and vasoconstrictor, respectively, are vital in establishing vascular resistance systemically, and are especially critical to the uterine microcirculation and placental vasculature. In normotensive human pregnancies, umbilical arteries and chorionic plate arteries had decreased PGI2 induced-vasodilatory capability (Chaudhuri et al., 1993). A study using human placenta chorionic plate vessels determined the TXA2 mimetic, U46619, increased perfusion pressure in normotensive fetal placental circulation in vitro (Read et al., 1999). A rat gestational hypoxia model demonstrated that vasoactivity can change in both the uterine circulation and umbilical vein (Aljunaidy et al., 2016). Our lab has demonstrated that maternal nano-titanium dioxide (nano-TiO2) inhalation alters the uterine microcirculation and results in increased sensitivity to the vasoconstrictive actions of U46619 (Griffith et al., 2022). Maternal inhalation of nano-TiO2 also reduced vasodilation in response to the stable PGI2 analog, carbaprostacyclin (Griffith et al., 2022). In conjunction with this, we have also found that maternal nano-TiO2 inhalation during gestation results in litter decreased fetal mass (Bowdridge et al., 2019), increased placental mass (Bowdridge et al., 2019), and decreased male: female ratio in early and mid-gestational exposures (Garner et al., 2022b). Further studies expanded on the fetal mass and determined that maternal nano-TiO2 inhalation during gestation specifically decreased female pup mass (Griffith et al., 2022). This has led us to hypothesize herein that maternal nano-TiO2 inhalation exposure during gestation alters placental hemodynamics and therefore influences fetal health outcomes in a sexually dimorphic manner. ## 2.1 Animal model Timed pregnant Sprague-Dawley (SD; delivered on GD 5–10) rats were purchased from Hilltop Laboratories (Scottdale, PA) and single-housed in an American Association for Accreditation of Laboratory Animal Care (AAALAC) approved facility at West Virginia University (WVU) Health Sciences Center. Rats were housed in a maintained environment: temperature (20–26°C), relative humidity (30–$70\%$), and light-dark cycle (12:12 h). Rats were acclimated for 48–72 h, then randomly assigned to either sham-control ($$n = 13$$) or nano-TiO2 ($$n = 14$$) exposure groups. Rat cages were lined with standard bedding (0.25-inch corncob) and had ad libitum access to standard chow (2918X; Envigo, Indianapolis, IN) and water throughout the acclimation and exposure periods. On GD 20, rats were weighed and then anesthetized with isoflurane gas ($5\%$ induction, 2–$3.5\%$ maintenance), placed on a warm heating pad, and maintained at a rectal temperature of 37°C. Rats were euthanized via thoracotomy and heart removal and then distribution of fetuses within the uterine horns and fetal sex was recorded. Fetal tissue was weighed and grouped according to fetal sex: whole placentas, placental junctional zone, and placental labyrinth zone. All procedures were approved by the WVU Institutional Animal Care and Use Committee. ## 2.2 Nanomaterial Nano-TiO2 powder was obtained from Evonik (P25 Aeroxide TiO2, Parsippany, NJ) and is composed of a mixture of anatase ($80\%$) and rutile ($20\%$) TiO2. Particle characteristics have previously been determined, including primary particle size (21 nm), specific surface area (48.08 m2/g), and Zeta potential (-56.6 mV) (Yi et al., 2013; Stapleton et al., 2018). ## 2.3 Inhalation exposure and aerosol characterization A high-pressure acoustical generator (HPAG, IEStechno, Morgantown, WV) created nano-TiO2 aerosols. Output from the generator was fed into a Venturi pump (JS-60M, Vaccon, Medway, MA) to further de-agglomerate particles. The nano-TiO2 aerosol mix enters a whole-body exposure chamber and a personal DataRAM (pDR-1500; Thermo Environmental Instruments Inc., Franklin, MA) samples the air to determine aerosol mass concentration in real-time. Software feedback loops automatically adjust the acoustic energy needed to maintain a stable mass aerosol concentration throughout the exposure. Gravimetric aerosol sampling measurements were conducted with Teflon filters concurrently with the DataRAM measurements to obtain calibration factors. Gravimetric measurements were taken during each exposure to calculate the mass concentration measurement. Real-time aerosol size distributions were measured in the exposure chamber at a target mass concentration of 12 mg/m3 via: 1) a high-resolution electrical low-pressure impactor (ELPI+; Dekati, Tampere, Finland); 2) a scanning particle mobility sizer (SMPS 3938; TSI Inc., St. Paul, MN); 3) an aerodynamic particle sizer (APS 3321; TSI Inc., St. Paul, MN); and 4) a micro-orifice uniform deposit impactor (MOUDI 115R, MSP Corp, Shoreview, MN). Bedding material was soaked to maintain proper humidity (20–$70\%$). Similar temperature and humidity conditions were maintained in exposure chambers utilized only for sham-control animals, which were exposed to HEPA-filtered air only. Inhalation exposures were performed for 6 non-consecutive days from GD 10–19 to prevent pregnancy loss. A target concentration of 12 mg/m3 was used for late gestation inhalation exposure (Stapleton et al., 2013; Stapleton et al., 2018). For estimation of lung deposition (dose) with nano-TiO2 aerosols, equation D = F•V•C•T was used where F is the deposition fraction ($10\%$), V is the minute ventilation (208.3 cc), C is mass concentration (mg/m3), and T equals the exposure duration (minutes) (Nurkiewicz et al., 2008; Stapleton et al., 2013). The exposure paradigm (12 mg/m3, 6 h/exposure, 6 days) produced a calculated cumulative lung deposition of 525 ± 16 µg (Bowdridge et al., 2022; Griffith et al., 2022) with the last exposure occurring on GD19 24-h prior tissue collection. The calculations represent total lung deposition and do not account for lung clearance (MPPD Software v 2.11, Arlington, VA). ## 2.4 Pressure myography with isolated placentas Once dams were euthanized and pups per horn count was recorded, the uterus was surgically excised and placed into a dissection dish containing physiological salt solution (PSS, in mmol/L: 129.8 NaCl, 5.4 KCl, 0.5 NaH2PO4, 0.83 MgSO4, 19.0 NaHCO3, 1.8 CaCl2, 5.5 glucose). The uterus was incised longitudinally, and amnionic sacs were opened to allow for quick identification of fetal sex. Fetal sex and position within the horn was recorded, then the first male and female nearest the cervix were removed with the placenta still attached. The placenta/pup units were placed into dissection dishes with PSS maintained at 4°C and were utilized for placental hemodynamic assessment (Garner et al., 2022a). The umbilical artery and vein were separated from the umbilical cord. Once the amnionic sac and vitelline vessels were removed, the umbilical vessels were cut as close to the pup as possible (Garner et al., 2022a). The placenta was then closed at the site of implantation with 6–0 silk sutures (AD Surgical, Sunnyvale, CA) and placentas were transferred to an isolated vessel chamber (Living Systems Instrumentation, Burlington, VT) containing 10 ml of oxygenated ($21\%$ O2/$5\%$ CO2) 37°C PSS. The umbilical artery was attached to the inflow glass pipette tip and the umbilical vein was attached to the outflow pipette tip (Garner et al., 2022a) using 6–0 silk sutures. Placentas were then pressurized from 0 to 20 mm Hg in 5 mm Hg increments. Outflow pressure and flow rate were assessed following addition of vasoactive drugs to assess vascular hemodynamics. Endothelium-dependent responses were assessed by acetylcholine (ACh, 1 × 10−4 M), application of s-nitroso-N-acetyl-dl-penicillamine (SNAP, 1 × 10−4 M) assessed endothelium-independent responses, addition of phenylephrine (PE, 1 × 10−4 M) assessed α1-adrenergic vasoconstriction, addition of carbaprostacyclin (1 × 10−10 M) assessed cyclooxygenase metabolite vasodilation, and application of U46619 (1 × 10−4 M) assessed cyclooxygenase metabolite vasoconstriction. Drugs were added to the bath individually to assess outflow pressure and flow rate response. Washes were done between each drug to ensure clearance. Once all drug response were assessed PSS was removed and replaced with Ca2+-free PSS to assess passive maximum outflow pressure and flow rate. ## 2.5 Placental histology Male and female placentas were collected from sham-control ($$n = 5$$ per sex) and nano-TiO2 ($$n = 5$$ per sex) exposed dams. Placentas were perfused with $4\%$ paraformaldehyde and fixed ex situ with $4\%$ paraformaldehyde at 4°C overnight. Placental tissue was then cleared with phosphate buffer solution (PBS) and transferred PBS overnight. Tissue was then rapidly frozen via isopentane and liquid nitrogen and stored at -80°C. Using a cryostat at -20°C, placentas were sectioned at 10 µm thickness. Sections from the center of the placenta, which provided the largest cross-sectional area, were placed on subbed slides, and stained with hematoxylin and eosin (H&E) following provided protocol instructions (Vector Laboratories, Burlingame, CA). Tissue sections were incubated with hematoxylin for 1 min and eosin for 5 min. Slides were imaged at 10x and analyzed using ImageJ (Xu et al., 2020). Total placental area and percent total area of the junctional and labyrinth zones were determined using an average of measures from three sections per pup. ## 2.6 Immunohistochemical staining of placentas Male and female placentas from sham-control ($$n = 5$$ per sex) and nano-TiO2 ($$n = 5$$ per sex) dams were collected on GD20. Placentas were perfused and then placed into a $4\%$ paraformaldehyde fixative overnight at 4°C and then transferred to PBS for the following night (Bowdridge et al., 2019). Tissue was then flash frozen and stored at -80°C until sectioned (Bowdridge et al., 2019). Placentas were sectioned at 10 µm through the center of the placenta. Three sections per pup were analyzed. Sections were washed 4 × 5 min with 0.1 M PBS to remove cryoprotectant. Sections were then incubated for 10 min with $1\%$ H2O2 and washed 4 × 5 min with 0.1 M PBS. Sections were incubated for 1 h at room temperature with 0.1 M PBS, $0.4\%$ Triton-X100 (Sigma-Aldrich, St. Louis, MO, United States of America) and $20\%$ normal goat serum (NGS; Jackson ImmunoResearch Laboratories, Inc., West Grove, PA, United States of America). Sections were then incubated with mouse monoclonal anti-Pan cytokeratin antibody (1:250, F3418; Sigma-Aldrich) (Nteeba et al., 2020) for 24 h at 4°C. Slides were washed 5 × 5 min, and then incubated in mouse monoclonal anti-rat CD163 (1:500, MCA342R; Bio-Rad Laboratories, Hercules, CA, United States of America) (Rosario et al., 2009) for 24 h at four°C. The final day, slides were washed with 0.1 M PBS 3 × 5 min. Then incubated with Alexa555 goat anti-mouse IgG1 (1:200; A21127; Thermo Fisher Scientific Inc., Waltham, MA, United States of America) for 1 h. Slides were washed 4 × 5 min and covered with a coverslip using ProLong Diamond Antifade Mountant with DAPI (Thermo Fisher). Slides were stored in the dark at 4°C until analysis. ## 2.7 statistics Dam characteristics, such as age, mass, and litter size, were assessed by unpaired t-test with Welch’s correction. The remainder of the dam characteristics were assessed by two-way analysis of variance (ANOVA). Fetal mass characteristics and total area of placental zones were analyzed via a two-way mixed-effects ANOVA. A two-way mixed effects model was used for assessing point-to-point differences in dose response curves to vascular agonists and increased pressure curves. If statistical significance occurred, then a Tukey post hoc test was used for all ANOVA analysis. All data are reported as mean ± SEM, unless otherwise stated. Significance was set at p ≤ 0.05. ## 3.1 Nanoparticle aerosol characteristics Average real-time aerosol mass concentration over the course of exposures was 12.17 mg/m3 with a standard deviation of 1.69 (Figure 1A). SMPS and APS measured the aerosol mobility diameter, which had a count median diameter (CMD) of 118 nm and a geometric standard deviation (GSD) of 2.10 (Figure 1B). ELPI assessment of the aerosol aerodynamic diameter showed, a mass median aerodynamic diameter (MMAD) of 164 nm with a geometric standard deviation of 1.89 (Figure 1C). A Nano Micro-Orifice Uniform Deposit Impactor (MOUDI 115R, MSP Corp, Shoreview, MN) was utilized to measure mass size distribution, which had a mass median aerodynamic diameter (MMAD) of 0.92 µm and a GSD of 2.47 (Figure 1D). The morphology of the nano-TiO2 agglomerates has been previously characterized extensively with electron microscopy (Abukabda et al., 2019; Bowdridge et al., 2019). FIGURE 1: Nano-TiO2 aerosol real-time characterizations. Aerosol characterization was monitored and verified during the exposure periods. Red lines indicate size distribution curves for a log normal fit of the size data. (A) Program software-controlled aerosol mass concentration over the 6-h exposure paradigm. Real-time nano-TiO2 aerosol mass concentration (black line) was maintained near the desired 12 mg/m3, the target average concentration (red line). (B) Aerosol aerodynamic diameter was assessed by scanning mobility particle sizer (SMPS; light gray) and an aerodynamic particle sizer (APS; dark gray). The particle diameter, determined by SMPS and APS, had a CMD of 118 nm and a geometric standard deviation of 2.10. (C) Aerosol aerodynamic diameter was also assessed by a high-resolution electrical low-pressure impactor (ELPI) which had a CMD of 164 nm and geometric standard deviation of 1.89. (D) A nano micro-orifice uniform deposit impactor (MOUDI) was used to evaluate aerosol mass size distribution and indicated a mass median aerodynamic diameter (MMAD) of 0.918 µm and a geometric standard deviation of 2.47. ## 3.2 Pregnant rat and litter characteristics Dams displayed no significant differences in age, litter size, or fetal sex on GD20 between groups (Table 1). There was a significant decrease in dam mass on GD20 in nano-TiO2 exposed dams ($$n = 14$$) compared to sham-control ($$n = 13$$; Table 1). TABLE 1 \| Exposure \| N \| Dam age (d) \| Dam mass (g) \| Litter size (pup number) \| Fetal horn distribution (pup number) \| Fetal horn distribution (pup number).1 \| Sex distribution (pup number) \| Sex distribution (pup number).1 \| Resorptions distribution (number of sites) \| Resorptions distribution (number of sites).1 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Exposure \| N \| Dam age (d) \| Dam mass (g) \| Litter size (pup number) \| Left \| Right \| Male \| Female \| Left \| Right \| \| Sham-control \| 13 \| 70.2 ± 2.2 \| 358.4 ± 9.3 \| 13.2 ± 0.6 \| 6.7 ± 0.4 \| 6.5 ± 0.6 \| 6.5 ± 0.3 \| 6.2 ± 0.5 \| 0.9 ± 0.09 \| 0.55 ± 0.21 \| \| Nano-TiO2 \| 14 \| 71.9 ± 1.6 \| 330.8 ± 4.7 * \| 12.6 ± 0.6 \| 7.0 ± 0.5 \| 5.6 ± 0.6 \| 6.0 ± 0.4 \| 5.4 ± 0.2 \| 0.08 ± 0.08 \| 0.23 ± 0.17 \| Fetal pup and placental zone mass after nano-TiO2 inhalation exposure during gestation were assessed according to fetal sex to determine any sexually dimorphic outcomes (Figure 2). Maternal nano-TiO2 inhalation exposure significantly decreased fetal female wet mass (2.65 ± 0.03 g) compared to the nano-TiO2 fetal males (2.78 ± 0.04 g; Figure 2A). The nano-TiO2 exposed fetal female mass decreased compared to the sham-control fetal female mass (2.74 ± 0.03 g). The placenta is composed of two core regions, the junctional zone (JZ) and the labyrinth zone (LZ). The JZ (Figure 2B) presented sex-based differences in mass and an effect of exposure seen only in the females. Sham-control fetal females (0.23 ± 0.01 g) had significantly larger JZ compared to sham-control fetal males (0.20 ± 0.01 g). Nano-TiO2 exposed fetal female JZ wet mass (0.18 ± 0.01 g) was significantly decreased compared to sham-control fetal females (0.23 ± 0.01 g) and nano-TiO2 exposed fetal males (0.20 ± 0.04 g). The wet LZ mass (Figure 2C) of nano-TiO2 fetal females (0.30 ± 0.01 g) was significantly smaller compared to nano-TiO2 fetal males (0.34 ± 0.01 g). This indicates that maternal nano-TiO2 inhalation exposure during gestation has a greater effect on fetal female pup and placenta mass. FIGURE 2: Fetal Pup and Placental Zone Mass. Fetal and placental mass were recorded based on fetal sex on GD 20 (N = 13, sham-control fetal male and female, and N = 14, nano-TiO2 fetal male and female). (A) Wet fetal mass. (B) Wet junctional zone mass. (C) Wet labyrinth zone mass. (D) Dry fetal mass. (E) Dry junctional zone mass. (F) Dry labyrinth zone mass. a, p ≤ 0.05 vs sham-control fetal male. b, p ≤ 0.05 vs sham-control fetal female. c, p ≤ 0.05 vs nano-TiO2 fetal male. Dry mass was also measured to discern if wet mass differences were driven by water content. Dry fetal mass (Figure 2D) was significantly decreased in nano-TiO2 fetal females (0.33 ± 0.01 g) compared to the nano-TiO2 male counterparts (0.35 ± 0.01 g). The dry sham-control fetal males (0.36 ± 0.01 g) tended to be larger than the dry fetal sham-control females (0.34 ± 0.01 g). Dry JZ mass (Figure 2E) was also significantly decreased in nano-TiO2 fetal females (0.028 ± 0.001 g) compared to nano-TiO2 fetal males (0.033 ± 0.001 g). There were no significant differences in dry LZ mass across treatment groups (Figure 2F). This provides evidence that nano-TiO2 exposure during gestation causes structural mass changes, not based on water content. ## 3.3 Sexually dimorphic placental hemodynamics Placental outflow pressure for sham-control fetal males ($$n = 5$$–6), sham-control fetal females ($$n = 4$$–6), nano-TiO2 fetal males ($$n = 5$$–7), and nano-TiO2 fetal females ($$n = 5$$–7) was measured to assess vascular resistance within the perfused tissue. Placentas were incubated in the presence of PSS, ACh, carbaprostacyclin, and thromboxane and outflow pressure was assessed (Figures 3, 4). Responses were also determined following exposure to SNAP, PE, and Ca2+-free PSS (Supplementary Figure S1). FIGURE 3: Placental Outflow Hemodynamics. Outflow pressure readings were recorded in conjunction with increased inflow input. (A) Outflow pressure in the presence of physiological saline solution (PSS). (B) Maximum outflow response in the presence of PSS. (C) Outflow pressure in the presence of the endothelium-dependent vasodilator, ACh. (D) Maximum outflow response in the presence of ACh. Sham-control fetal male, n = 5-6, nano-TiO2 fetal male, n = 5-7, sham-control fetal female, n = 4-5, and nano-TiO2 fetal female, n = 5-6. FIGURE 4: Placental Cyclooxygenase Metabolites Outflow Hemodynamics. Outflow pressure readings were recorded with increased inflow input. (A) Placental outflow pressure in the presence of carbaprostacyclin, a stable PGI2 agonist. (B) Maximum outflow response to carbaprostacyclin. (C) Outflow pressure with U46619, the TXA2 mimetic. (D) Maximum outflow pressure response in the presence of U46619. Sham-control fetal male, n = 5-6, nano-TiO2 fetal male, n = 5-7, sham-control fetal female, n = 4-5, and nano-TiO2 fetal female, n = 5–6. a, p ≤ 0.05 vs sham-control fetal male. b, p ≤ 0.05 vs sham-control fetal female. c, p ≤ 0.05 vs nano-TiO2 fetal male. Placental responses to physiological saline solution (PSS) were assessed to test baseline outflow pressure (Figures 3A, B), which was not significantly different amongst treatments. Placental outflow pressure response to the endothelium-dependent vasodilator, ACh (Figures 3C, D) were not significantly different amongst treatments. Responses to carbaprostacyclin, a cyclooxygenase vasodilator (Figures 4A, B), were also not significantly different between groups. The thromboxane mimetic, U46619 (Figures 4C, D), resulted in sham-control fetal females to have increased outflow compared to sham-control fetal males at 15 mm Hg inflow pressure (9.10 ± 1.07 mm Hg vs 5.11 ± 1.02 mm Hg). Placentas of nano-TiO2 exposed fetal females had significantly decreased outflow pressure (3.97 ± 1.30 mm Hg) compared to sham-control fetal females (9.10 ± 1.07 mm Hg) and nano-TiO2 exposed fetal male placentas (9.96 ± 0.66 mm Hg). Nano-TiO2 exposed fetal male placentas also had significantly increased outflow compared to sham-control fetal males (9.96 ± 0.66 vs 5.11 ± 1.02 mm Hg, respectively). There were no significant differences between groups for outflow pressures when incubated with PE, SNAP, or Ca2+-free PSS (Supplementary Figure S1A–C). This indicates that maternal nano-TiO2 inhalation exposure during gestation results in modified placental hemodynamics that are specific to thromboxane. Placental flow rates were also assessed for each group to determine hemodynamic responses to inflow pressure and vasoactive drugs. There were no significant differences for placenta flow rates across groups when incubated with PSS, ACh, PGI2 analog, carbaprostacyclin, and the TXA2 mimetic, U46619 (Figures 5A–D and Figures 6A–D). Placenta flow rates did not demonstrate significant differences between groups when incubated with PE, SNAP, or Ca2+-free PSS (Supplementary Figure S2A–C). FIGURE 5: Placental flow rate hemodynamics. Flow rate was calculated and recorded throughout experiment as inflow pressure was increased stepwise manner. (A) Flow rate in PSS bathe. (B) Maximum flow rate response in PSS. (C) Flow rate of placentas in the presence of the ACh, an endothelium-dependent vasodilator. (D) Maximum flow rate response across increased pressure. Sham-control fetal male, n = 5-6, nano-TiO2 fetal male, n = 5-7, sham-control fetal female, n = 4-5, and nano-TiO2 fetal female, n = 5-6. FIGURE 6: Placental Cyclooxygenase Metabolites flow rate hemodynamics. Flow rate was calculated and recorded during inflow pressure increases in a stepwise manner. (A) Placenta flow rate in the presence of carbaprostacyclin, a stable PGI2 agonist, across increased inflow pressure. (B) Maximum flow rate in response to carbaprostacyclin. (C) Placenta flow rate with U46619, a TXA2 mimetic, added to the bathe. (D) Maximum flow rate response to U46619. Sham-control fetal male, n = 5-6, nano-TiO2 fetal male, n = 5-7, sham-control fetal female, n = 4-5, and nano-TiO2 fetal female, n = 5-6. ## 3.4 Placental histology & immunohistochemistry Placentas were collected for each group (sham-control males and females ($$n = 5$$/sex); nano-TiO2 males and females ($$n = 5$$/sex)) to assess differences in placental JZ and LZ area and anatomy between fetal sex and exposure paradigm. A representative image of the placenta histology is shown in Figure 7A. Total placenta area was assessed, in which there was a significant increase in area size for sham-control female compared to sham-control male (84,130 ± 3834 AU vs 69,956 ± 3660 AU; Figure 7B). Nano-TiO2 female area was also significantly increased compared to nano-TiO2 male total placenta area (89,697 ± 6141 AU vs 76,558 ± 4272 AU; Figure 7B). The percent JZ area was significantly decreased for nano-TiO2 males (26.15 ± $1.59\%$) compared to sham-control males (30.93 ± $1.37\%$; Figure 7C). Nano-TiO2 females also had a significantly decreased JZ area compared to sham-control females (24.37 ± $1.30\%$ vs 30.39 ± $1.54\%$; Figure 7C). There was no significant difference between fetal sex within their exposure group (Figure 7C). Total LZ area is highlighted in Figure 7D. Nano-TiO2 males had a significant increase in LZ area compared to sham-control males (73.85 ± $1.59\%$ vs 69.07 ± $1.37\%$). Nano-TiO2 females (75.63 ± $1.30\%$) had a significant increase of LZ area compared to sham-control females (69.61 ± $1.54\%$; Figure 7D). There was not a significant difference for fetal sex within their exposures. This indicates that not only does maternal nano-TiO2 inhalation exposure during gestation change placental mass, but results in placental area changes as well. FIGURE 7: Placental Histology. Total placental zone areas were assessed after H&E staining to determine anatomical and structural differences after maternal nano-TiO2 inhalation exposure. (A) Representative images of placenta histology are depicted for each group. (B) Total area of the placenta. (C) Percent total area of the junctional zone (JZ). (D) Percent total area of the labyrinth zone (LZ). Sham-control males (n = 5), sham-control females (n = 5), nano-TiO2 males (n = 5), and nano-TiO2 females (n = 5). a, p ≤ 0.05 vs sham-control fetal male. b, p ≤ 0.05 vs sham-control fetal female. c, p ≤ 0.05 vs nano-TiO2 fetal male. Additionally, a subset of placentas was used to quantify Hofbauer cell (CD163; macrophages specific to gestation) and trophoblast cell (anti-Pan cytokeration; placental lineage cells), along with their co-localization. A representative image is provided in Figure 8A. Hofbauer cell pixel intensity for the total placenta is depicted in Figure 8B. There was a significant increase in fluorescent intensity for nano-TiO2 females compared to sham-control females (93.41 ± 3.05 AU vs 52.80 ± 6.67 AU; Figure 8B). Trophoblast cell fluorescent intensity was significantly decreased in sham-control female (57.12 ± 3.84 AU) compared to sham-control males (80.17 ± 7.90 AU) and compared to nano-TiO2 females (81.18 ± 6.53 AU; Figure 8C) in the total placenta. Colocalization of Hofbauer and trophoblast cells is depicted in Figure 8D. Co-localization was significantly increased in nano-TiO2 females compared to sham-control females (23,387 ± 3,172 AU vs 11,293 ± 1,896 AU). Maternal nano-TiO2 inhalation exposure during gestation results in modified cellular composition of the placentas. FIGURE 8: Whole Placental Immunohistochemistry. Total placental fluorescence intensity and colocalization of anti-CD163 and anti-pan cytokeratin was assessed after maternal nano-TiO2 inhalation exposure. (A) Representative placental images for each group. (B) Total placental Hofbauer cell pixel intensity. (C) Total placental pan-cytokeratin pixel intensity. (D) Total placental colocalization of Hofbauer: Pan-cytokeratin. Sham-control males (n = 5), sham-control females (n = 5), nano-TiO2 males (n = 5), and nano-TiO2 females (n = 5). a, p ≤ 0.05 vs sham-control fetal male. b, p ≤ 0.05 vs sham-control fetal female. Immunohistochemistry staining was also evaluated based on JZ and LZ fluorescent intensity and colocalization, which is shown in Figure 9. Hofbauer cells significantly decrease in nano-TiO2 females compared to nano-TiO2 males (66.47 ± 5.08 AU vs 103.1 ± 5.75 AU). Nano-TiO2 males tended to have increased CD163 intensity compared to sham-control males (78.65 ± 5.91 AU; $$p \leq 0.06$$; Figure 9A). Pan-cytokeratin intensity for JZ was significantly increased in sham-control males compared to females (67.24 ± 5.21 AU vs 44.43 ± 2.59 AU; Figure 9B) and nano-TiO2 female (67.38 ± 9.43) tended to have increased compared to sham-control females ($$p \leq 0.07$$). Colocalization of CD163 and pan-cytokeratin was assessed in the JZ, in which there was no significant difference (Figure 9C). FIGURE 9: Placental Zone Immunohistochemistry. Placental fluorescence intensity and colocalization of anti-CD163 and anti-pan cytokeratin in each placenta zone. (A) Junctional zone (JZ) pixel intensity of anti-CD163 (Hofbauer cell). (B) JZ pixel intensity for pan-cytokeratin. (C) JZ colocalization for CD163: pan-cytokeratin. (D) Labyrinth zone (LZ) pixel intensity for Hofbauer cells. (E) LZ pixel intensity for pan-cytokeratin. (F) LZ colocalization for CD163: pan-cytokeratin. Sham-control males (n = 5), sham-control females (n = 5), nano-TiO2 males (n = 5), and nano-TiO2 females (n = 5). a, p ≤ 0.05 vs sham-control fetal male. b, p ≤ 0.05 vs sham-control fetal female. c, p ≤ 0.05 vs nano-TiO2 fetal male. Within the LZ, there was a significant decrease for CD163 intensity of sham-control females compared to males (62.00 ± 12.80 AU vs 94.74 ± 15.00 AU; Figure 9D). Nano-TiO2 females (116.1 ± 7.869) had significantly increased CD163 intensity compared to sham-control females (62.00 ± 12.80) and nano-TiO2 males (75.61 ± 6.58 AU). LZ pan-cytokeratin fluorescence was significantly decreased for sham-control female compared to males (64.65 ± 7.47 AU vs 91.55 ± 10.23 AU; Figure 9E) and nano-TiO2 females (95.19 ± 7.19 AU) tended to have increased staining compared to sham-control females ($$p \leq 0.08$$). Within the LZ, there was a significant decrease of colocalization for sham-control female (3,415 ± 819.8 AU) compared to sham-control male (6,513 ± 719.8 AU) and nano-TiO2 female (18,706 ± 2,802 AU; Figure 9F). There was also a significant increase for colocalization for nano-TiO2 female compared to nano-TiO2 male (18,706 ± 2,802 vs 11,631 ± 11,577). The placenta zones of the exposed fetuses are also impacted with changes in their cellular composition, and thus may change their functionality. ## 4 Discussion The primary aim of this project was to determine if maternal nano-TiO2 inhalation exposure during gestation alters placental vascular reactivity and fetal growth in a sexually dimorphic manner. Herein, we demonstrated that maternal nano-TiO2 gestational inhalation exposure produces placental dysfunction in a sex-dependent manner. While female fetuses have the greatest impact, with decreased placental size and area, fetal growth, and placental hemodynamic capabilities, these are likely adaptations to preserve fetal life. Our laboratory has also previously demonstrated decreased male to female ratio in early and mid-gestation inhalation exposures (Garner et al., 2022b). Males are more susceptible to fetal loss due to external maternal stress during gestation (Kraemer, 2000). Additionally, maternal disease (like Diabetes Mellitus) may affect male fetal congenital development and perinatal outcomes (Evers et al., 2009; García-Patterson et al., 2011). It appears that females are more adaptable in hostile environments to ensure they survive gestation, but these adaptations may be to their detriment later in life. In utero perturbations can result in fetal intrauterine growth restriction (IUGR), which is a risk factor for many adult diseases such as cardiovascular disease (CVD), diabetes, dyslipidemia, hypertension, metabolic syndrome, or renal diseases later in life (Menendez-Castro et al., 2018). Insults that result in a hostile gestational environments that cause diseases later in life is part of the Barker hypothesis, widely referred to as the developmental origins of health and disease (DOHaD) (Barker, 1990; Barker and Martyn, 1992). In this study, we observed a significant decrease in fetal mass for the nano-TiO2 exposed fetal females compared to nano-TiO2 males (Figure 2A) which was anticipated as this has been previously shown (Griffith et al., 2022). Modification of blood flow or nutrient exchange to the fetus can have different impacts on the progression of fetal growth between sexes. In a gestational guinea pig model, it was found that early-onset hypoxia caused both male and female mass to decrease, but late-onset hypoxia caused only female mass to decrease compared to sex-matched controls (Thompson et al., 2020). Hypoxic models are important as they indicate modifications in blood flow and vascular resistance changes to increased oxygen delivery to critical organs (Heinonen et al., 2016). Intrauterine growth restriction (IUGR), as seen in our study and guinea pig hypoxia study (Thompson et al., 2020), has a strong association with impaired fetal blood flow (Laurin et al., 1987), thus leaving fetuses to attempt to adapt to this hostile gestational environment. Fetal growth can be impacted by toxicant exposures, in a sexually dimorphic manner and these perturbations can be exasperated by direct toxic effects on the placenta. Toxicant exposures can affect total placental mass, placenta zone mass and area of placental zones. Herein, placental zone mass (Figures 2B, C) and placenta zone areas (Figures 7C, D) changes occurred after nano-TiO2 exposure during gestation. Placental perturbations were most pronounced in the nano-TiO2 exposed females, which had decreased JZ and LZ mass, decreased JZ area, and increased LZ area. Decreased JZ mass and area could lead to modifications in hormone production and increased LZ area results in modifications to placental nutrient-waste exchange capabilities in a sex-dependent manner (Gårdebjer et al., 2014). Studies of diet restriction, in mice have demonstrated decreased fetal mass (Belkacemi et al., 2009; Coan et al., 2010; Connor et al., 2020), decreased JZ and LZ mass (Belkacemi et al., 2009), and decreased JZ volume (Coan et al., 2010) or area (Schulz et al., 2012; Connor et al., 2020). Undernutrition in mice has also been reported to increase LZ area or volume (Coan et al., 2010; Schulz et al., 2012) and make up a larger proportion of the placenta. These studies came to similar conclusions, that while the overall placenta mass was decreased, the increased LZ area is an attempt to compensate for nutrient restriction and preserve fetal growth (Coan et al., 2010; Schulz et al., 2012). Changes in the mass and area of the JZ and LZ are important to fetal development, however the cellular composition of these zones is equally as important. These changes could be due to cell proliferation or hypertrophy and can result in functional changes in the fetoplacental unit. A hostile in utero environment during gestation may cause the placenta to go through modifications in zone size, mass, area, and volume, as discussed above, but it may also result in cellular composition changes preserve fetal life. Indeed, we observed changes in the cellular composition of placentas in our exposure model. Further, these changes occur in a sexually dimorphic manner in the whole placenta (Figures 8B, C) and in the placenta zones (Figures 9A–E, H). A reduced uteroplacental perfusion pressure (RUPP) mouse model for PE during pregnancy found that RUPP surgery on E14.5 resulted in altered proliferation and differentiation LZ trophoblast makers (Natale et al., 2018). At E16.5, there was an increase in trophoblast and endothelial proliferation markers within the LZ of RUPP placentas (Natale et al., 2018). The JZ of RUPP placentas had a constant trophoblast giant cell (TGC) population, shrinking spongiotrophoblast relative to placenta size, unlike controls which decreased their TGC population over gestation (Natale et al., 2018). It was proposed in this study that the altered trophoblast proliferation was due to hypoxic conditions, which has been demonstrated to trigger trophoblast proliferation in vitro (Caniggia et al., 2000; Natale et al., 2018). This is important because increased trophoblast proliferation, may indicate increased fetal blood space area (Natale et al., 2018), to help increase nutrient-waste exchange. The Hofbauer cell marker, CD163, has been shown to preferentially localize near fetal vessels and trophoblasts and are found within the placenta throughout the majority of gestation (Swieboda et al., 2020). Hofbauer cells are still not fully understood, but their function has been shown to be perturbed in diseases like chronic villitis or villitis of unknown etiology (VUE), in which proliferation of Hofbauer cells is seen (Reyes and Golos, 2018). In these disease states, the Hofbauer cells exhibit more inflammatory phenotypes, which is actually thought to cause more placental damage (Reyes and Golos, 2018). In this model, it is likely the trophoblast cells are increasing within the LZ, much like the Hofbauer cells, and both are functioning to compensate for the nutrient restriction and aid to preserve fetal life and growth. Our lab has previously reported that gestational nano-TiO2 inhalation exposure alters maternal uterine radial arteriole vascular reactivity (Bowdridge et al., 2019; Garner et al., 2022b; Griffith et al., 2022). Therefore, it is likely that the changes in fetal female JZ and LZ placenta mass and area are changing in response to the toxicant exposure and the upstream alterations that are occurring on the maternal side of the vasculature to preserve fetal life and growth. Adaptations within the placental structure and cellular composition are not the only way that fetal life preservation can be achieved in perturbed uterine environments. The maternal side of the vasculature, such as the uterine radial arterioles, have been shown to have reduced vasoreactivity in the presence of vasoactive compounds like prostacyclin or thromboxane (Griffith et al., 2022). Additionally, the placentas from nano-TiO2 exposed dams have increased reactive oxygen species production rate (Bowdridge et al., 2022). Increased reactive oxygen species, like H2O2, has been shown to increase TXB2, the stable TXA2 metabolite, production in hypertensive rat mesenteric arterioles (Gao and Lee, 2001). This study demonstrates that nano-TiO2 exposed placentas also have modified placenta outflow in the presence of thromboxane agonist, U46619 (Figures 4C, D). Exposed fetal female placentas have decreased outflow pressure in the presence of U46619 compared to sham-control females and nano-TiO2 fetal males. Understanding the importance of this requires understanding the fetoplacental unit and the flow of nutrient-waste exchange by the umbilical cord (Figure 10). The umbilical vein (outflow from the placenta) carries the nutrient-rich oxygenated blood to the fetus while the umbilical artery (inflow toward the placenta) carries deoxygenated, nutrient-deprived blood to the placenta (Wang and Zhao, 2010). With this in mind, ultrasound and doppler flow measurements visualize and record measurements of umbilical artery, in which greater flow from the fetus to the placenta reflects a healthier fetus, that utilizes more nutrients for increased metabolic processes, growth, and development (Wang and Zhao, 2010). In conjunction with this, reduced umbilical vein blood flow is associated with low fetal birthweight (Wang and Zhao, 2010). In this model, the exposed female placentas demonstrate a decreased outflow pressure in the presence of a highly vasoconstrictive compound, U46619. The decreased outflow pressure reflects elevated placenta resistance in the presence of U46619. It is possible that these placentas have adapted to decrease their responsiveness to TXA2 to preserve fetal growth and ultimately prevent fetal death. Females are smaller at GD 20 (Griffith et al., 2022) (Figure 2A) and this same pattern of diminished growth persists up to 8-week (data not published). Additionally, adult females (∼10–11 weeks of age) exposed to nano-TiO2 in utero will also have smaller pups at GD 20 (Bowdridge et al., 2022) and have decreased plasma estrogen levels, which could be due to the decreased JZ area and mass seen in this study. In studies of mouse maternal nano-TiO2 inhalation exposure during gestation we found that fetal hearts had decreased cardiac output and increased LV mass (Kunovac et al., 2019). Young adult mice from this same exposure paradigm demonstrated decreased systolic radial displacement (Hathaway et al., 2017), decreased ejection fraction and fractional shortening (Kunovac et al., 2019). This indicates that maternal exposure to nano-TiO2 inhalation during gestation may cause fetal cardiac dysfunction that reaches into adulthood. As such, the fetal females from exposed dams are impacted to a greater extent, as shown by decreased fetal mass, decreased JZ area and mass, increased LZ area, and decreased outflow pressure in the presence of TXA2 mimetic. These adaptations may be compensatory mechanisms in impacted females that support gestational survival, growth, and reproduction later in life. FIGURE 10: Sex Based Placental Outcomes Summary. Maternal nano-TiO2 exposure during gestation results in sexually dimorphic outcomes in the placental and fetal tissue at GD 20. Females undergo more placental changes including decreases in mass (JZ and LZ) and area (JZ) when compared to males. These resulting changes may be responsible for the decreased fetal mass seen in females compared to their male counterparts at GD 20, and these changes have been shown to result in health deficits for the females into adulthood (Bowdridge et al., 2022). In conclusion, this project sought to determine if maternal nano-TiO2 inhalation exposure during gestation alters fetal growth, placental size, trophoblast invasion, and placental vasoactivity in a sexually dimorphic manner. Our exposure paradigm provides evidence that maternal nano-TiO2 inhalation exposure had a greater, and lasting, impact on fetal females regarding mass, placental zone mass, zone area, as well as placental vasoactivity. The modifications reported may be physiological adaptations for the fetal females to guarantee survival after maternal nano-TiO2 inhalation exposure, but at what cost? Future studies should investigate the impact of maternal nano-TiO2 inhalation exposure on zone specific mechanisms: H2O2 levels, production of TXA2 and PGI2, and this interaction in a sexually dimorphic manner. This would clarify if H2O2 and redox metabolites are driving these changes seen based on fetal sex. ## 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 animal study was reviewed and approved by West Virginia University Institutional Animal Care and Use Committee. ## Author contributions Study design, JG, EB, and TN. Data collection: JG, AD, ED, KS, KW, TB, and EB. Data analysis and interpretation, JG, AD and EB. Animal exposures, JG, AD, KE, TB, WG, KW, and EB. Manuscript draft, JG. Critical revisions and final decision to submit, 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. ## Supplementary material The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/ftox.2023.1096173/full#supplementary-material ## References 1. Abukabda A. B., Bowdridge E. C., McBride C. R., Batchelor T. P., Goldsmith W. T., Garner K. L.. Maternal titanium dioxide nanomaterial inhalation exposure compromises placental hemodynamics. Toxicol. Appl. Pharmacol. (2019) 367 51-61. DOI: 10.1016/j.taap.2019.01.024 2. Aljunaidy M. M., Morton J. S., Cooke C.-L., Davidge S. T.. Maternal vascular responses to hypoxia in a rat model of intrauterine growth restriction. Am. J. Physiol.-Regul. Integr. Comp. Physiol. (2016) 311 R1068-R1075. DOI: 10.1152/ajpregu.00119.2016 3. Barker D. J., Martyn C. N.. The maternal and fetal origins of cardiovascular disease. J. Epidemiol. Community Health (1992) 46 8-11. DOI: 10.1136/jech.46.1.8 4. Barker D. J.. The fetal and infant origins of adult disease. BMJ (1990) 301 1111. DOI: 10.1136/bmj.301.6761.1111 5. Behlen J. C., Lau C. H., Li Y., Dhagat P., Stanley J. A., Rodrigues Hoffman A.. Gestational exposure to ultrafine particles reveals sex- and dose-specific changes in offspring birth outcomes, placental morphology, and gene networks. Toxicol. Sci. (2021) 184 204-213. DOI: 10.1093/toxsci/kfab118 6. Belkacemi L., Chen C. H., Ross M. G., Desai M.. Increased placental apoptosis in maternal food restricted gestations: Role of the fas pathway. Placenta (2009) 30 739-751. DOI: 10.1016/j.placenta.2009.06.003 7. Bowdridge E. C., Abukabda A. B., Engles K. J., McBride C. R., Batchelor T. P., Goldsmith W. T.. Maternal engineered nanomaterial inhalation during gestation disrupts vascular kisspeptin reactivity. Toxicol. Sci. Off. J. Soc. Toxicol. (2019) 169 524-533. DOI: 10.1093/toxsci/kfz064 8. Bowdridge E. C., DeVallance E., Garner K. L., Griffith J. A., Schafner K., Seaman M.. Nano-titanium dioxide inhalation exposure during gestation drives redox dysregulation and vascular dysfunction across generations. Part. Fibre Toxicol. (2022) 19 18. DOI: 10.1186/s12989-022-00457-y 9. Caniggia I., Mostachfi H., Winter J., Gassmann M., Lye S. J., Kuliszewski M.. Hypoxia-inducible factor-1 mediates the biological effects of oxygen on human trophoblast differentiation through TGFβ. J. Clin. Invest. (2000) 105 577-587. DOI: 10.1172/JCI8316 10. Chaudhuri G., Cuevas J., Buga G. M., Ignarro L. J.. NO is more important than PGI2 in maintaining low vascular tone in feto-placental vessels. Am. J. Physiol. (1993) 265 H2036-H2043. DOI: 10.1152/ajpheart.1993.265.6.H2036 11. Coan P. M., Vaughan O. R., Sekita Y., Finn S. L., Burton G. J., Constancia M.. Adaptations in placental phenotype support fetal growth during undernutrition of pregnant mice. J. Physiol. (2010) 588 527-538. DOI: 10.1113/jphysiol.2009.181214 12. Connor K. L., Kibschull M., Matysiak-Zablocki E., Nguyen T. T.-T. N., Matthews S. G., Lye S. J.. Maternal malnutrition impacts placental morphology and transporter expression: An origin for poor offspring growth. J. Nutr. Biochem. (2020) 78 108329. DOI: 10.1016/j.jnutbio.2019.108329 13. Evers I. M., de Valk H. W., Visser G. H. A.. Male predominance of congenital malformations in infants of women with type 1 diabetes. Diabetes Care (2009) 32 1194-1195. DOI: 10.2337/dc09-0367 14. Gabory A., Roseboom T. J., Moore T., Moore L. G., Junien C.. Placental contribution to the origins of sexual dimorphism in health and diseases: Sex chromosomes and epigenetics. Biol. Sex. Differ. (2013) 4 5. DOI: 10.1186/2042-6410-4-5 15. Gao Y. J., Lee R. M. K. W.. Hydrogen peroxide induces a greater contraction in mesenteric arteries of spontaneously hypertensive rats through thromboxane A2 production. Br. J. Pharmacol. (2001) 134 1639-1646. DOI: 10.1038/sj.bjp.0704420 16. García-Patterson A., Aulinas A., Sojo L., Ginovart G., Adelantado J. M., de Leiva A.. Poorer perinatal outcome in male newborns of women with pregestational diabetes mellitus. Diabet. Med. J. Br. Diabet. Assoc. (2011) 28 436-439. DOI: 10.1111/j.1464-5491.2011.03227.x 17. Gårdebjer E. M., Cuffe J. S. M., Pantaleon M., Wlodek M. E., Moritz K. M.. Periconceptional alcohol consumption causes fetal growth restriction and increases glycogen accumulation in the late gestation rat placenta. Placenta (2014) 35 50-57. DOI: 10.1016/j.placenta.2013.10.008 18. Garner K. L., Bowdridge E. C., DeVallance E., Griffith J. A., Kelley E. E., Nurkiewicz T. R.. Using the isolated rat placenta to assess fetoplacental hemodynamics. Front. Toxicol. (2022a) 4 814071. DOI: 10.3389/ftox.2022.814071 19. Garner K. L., Bowdridge E. C., Griffith J. A., DeVallance E., Seman M. G., Engels K. J.. Maternal nanomaterial inhalation exposure: Critical gestational period in the uterine microcirculation is angiotensin II dependent. Cardiovasc. Toxicol. (2022b) 22 167-180. DOI: 10.1007/s12012-021-09712-8 20. Griffith J. A., Garner K. L., Bowdridge E. C., DeVallance E., Schafner K. J., Engles K. J.. Nanomaterial inhalation during pregnancy alters systemic vascular function in a cyclooxygenase-dependent manner. Toxicol. Sci. Off. J. Soc. Toxicol. Kfac055. (2022) 188 219-233. DOI: 10.1093/toxsci/kfac055 21. Hathaway Q. A., Nichols C. E., Shepherd D. L., Stapleton P. A., McLaughlin S. L., Stricker J. C.. Maternal-engineered nanomaterial exposure disrupts progeny cardiac function and bioenergetics. Am. J. Physiol.-Heart Circ. Physiol. (2017) 312 H446-H458. DOI: 10.1152/ajpheart.00634.2016 22. Heinonen I. H. A., Boushel R., Kalliokoski K. K.. The circulatory and metabolic responses to hypoxia in humans – with special reference to adipose tissue physiology and obesity. Front. Endocrinol. (2016) 7 116. DOI: 10.3389/fendo.2016.00116 23. Kraemer S.. The fragile male. BMJ (2000) 321 1609-1612. DOI: 10.1136/bmj.321.7276.1609 24. Kunovac A., Hathaway Q. A., Pinti M. V., Goldsmith W. T., Durr A. J., Fink G. K.. ROS promote epigenetic remodeling and cardiac dysfunction in offspring following maternal engineered nanomaterial (ENM) exposure. Part. Fibre Toxicol. (2019) 16 24. DOI: 10.1186/s12989-019-0310-8 25. Laurin J., Lingman G., Marsál K., Persson P. H.. Fetal blood flow in pregnancies complicated by intrauterine growth retardation. Obstet. Gynecol. (1987) 69 895-902. PMID: 3554065 26. Menendez-Castro C., Rascher W., Hartner A.. Intrauterine growth restriction - impact on cardiovascular diseases later in life. Mol. Cell. Pediatr. (2018) 5 4. DOI: 10.1186/s40348-018-0082-5 27. Miller C. N., Dye J. A., Henriquez A. R., Stewart E. J., Lavrich K. S., Carswell G. K.. Ozone-induced fetal growth restriction in rats is associated with sexually dimorphic placental and fetal metabolic adaptation. Mol. Metab. (2020) 42 101094. DOI: 10.1016/j.molmet.2020.101094 28. Natale B. V., Mehta P., Vu P., Schweitzer C., Gustin K., Kotadia R.. Reduced Uteroplacental Perfusion Pressure (RUPP) causes altered trophoblast differentiation and pericyte reduction in the mouse placenta labyrinth. Sci. Rep. (2018) 8 17162. DOI: 10.1038/s41598-018-35606-x 29. Nteeba J., Varberg K. M., Scott R. L., Simon M. E., Iqbal K., Soares M. J.. Poorly controlled diabetes mellitus alters placental structure, efficiency, and plasticity. BMJ Open Diabetes Res. Care (2020) 8 e001243. DOI: 10.1136/bmjdrc-2020-001243 30. Nurkiewicz T. R., Porter D. W., Hubbs A. F., Cumpston J. L., Chen B. T., Frazer D. G.. Nanoparticle inhalation augments particle-dependent systemic microvascular dysfunction. Part. Fibre Toxicol. (2008) 5 1. DOI: 10.1186/1743-8977-5-1 31. Read M. A., Leitch I. M., Giles W. B., Bisits A. M., Boura A. L., Walters W. A.. U46619-mediated vasoconstriction of the fetal placental vasculature. J. Hypertens. (1999) 17 389-396. DOI: 10.1097/00004872-199917030-00012 32. Reyes L., Golos T. G.. Hofbauer cells: Their role in healthy and complicated pregnancy. Front. Immunol. (2018) 9 2628. DOI: 10.3389/fimmu.2018.02628 33. Rosario G. X., Ain R., Konno T., Soares M. J.. Intrauterine fate of invasive trophoblast cells. Placenta (2009) 30 457-463. DOI: 10.1016/j.placenta.2009.02.008 34. Roseboom T. J., Painter R. C., de Rooij S. R., van Abeelen A. F. M., Veenendaal M. V. E., Osmond C.. Effects of famine on placental size and efficiency. Placenta (2011) 32 395-399. DOI: 10.1016/j.placenta.2011.03.001 35. Schulz L. C., Schlitt J. M., Caesar G., Pennington K. A.. Leptin and the placental response to maternal food restriction during early pregnancy in mice. Biol. Reprod. (2012) 87 120. DOI: 10.1095/biolreprod.112.103218 36. Sobrino A., Oviedo P. J., Novella S., Laguna-Fernandez A., Bueno C., García-Pérez M. A.. Estradiol selectively stimulates endothelial prostacyclin production through estrogen receptor-{alpha}. J. Mol. Endocrinol. (2010) 44 237-246. DOI: 10.1677/JME-09-0112 37. Stapleton P. A., Minarchick V. C., Yi J., Engels K., McBride C. R., Nurkiewicz T. R.. Maternal engineered nanomaterial exposure and fetal microvascular function: Does the barker hypothesis apply?. Am. J. Obstet. Gynecol. (2013) 209 227.e1-227.11. DOI: 10.1016/j.ajog.2013.04.036 38. Stapleton P. A., McBride C. R., Yi J., Abukabda A. B., Nurkiewicz T. R.. Estrous cycle-dependent modulation of. Reprod. Toxicol. Elmsford N. (2018) 78 20-28. DOI: 10.1016/j.reprotox.2018.03.001 39. Stark M. J., Dierkx L., Clifton V. L., Wright I. M. R.. Alterations in the maternal peripheral microvascular response in pregnancies complicated by preeclampsia and the impact of fetal sex. J. Soc. Gynecol. Investig. (2006) 13 573-578. DOI: 10.1016/j.jsgi.2006.06.006 40. Stark M. J., Clifton V. L., Wright I. M. R.. Neonates born to mothers with preeclampsia exhibit sex-specific alterations in microvascular function. Pediatr. Res. (2009) 65 292-295. DOI: 10.1203/PDR.0b013e318193edf1 41. Swieboda D., Johnson E. L., Beaver J., Haddad L., Enninga E. A. L., Hathcock M.. Baby’s first macrophage: Temporal regulation of hofbauer cell phenotype influences ligand-mediated innate immune responses across gestation. J. Immunol. (2020) 204 2380-2391. DOI: 10.4049/jimmunol.1901185 42. Thompson L. P., Turan S., Aberdeen G. W.. Sex differences and the effects of intrauterine hypoxia on growth and. Am. J. Physiol.-Regul. Integr. Comp. Physiol. (2020) 319 R243-R254. DOI: 10.1152/ajpregu.00249.2019 43. Wang Y., Zhao S.. Placental blood circulation, vascular biology of the placenta (2010) 44. Winterbottom E. F., Koestler D. C., Fei D. L., Wika E., Capobianco A. J., Marsit C. J.. The aquaglyceroporin AQP9 contributes to the sex-specific effects of. Environ. Health (2017) 16 59. DOI: 10.1186/s12940-017-0267-8 45. Xu B., Chen X., Ding Y., Chen C., Liu T., Zhang H.. Abnormal angiogenesis of placenta in progranulin-deficient mice. Mol. Med. Rep. (2020) 22 3482-3492. DOI: 10.3892/mmr.2020.11438 46. Yi J., Chen B. T., Schwegler-Berry D., Frazer D., Castranova V., McBride C.. Whole-body nanoparticle aerosol inhalation exposures. J. Vis. Exp. (2013) e50263. DOI: 10.3791/50263
--- title: Conditioning on parental mating types can reduce necessary assumptions for Mendelian randomization authors: - Keisuke Ejima - Nianjun Liu - Luis Miguel Mestre - Gustavo de los Campos - David B. Allison journal: Frontiers in Genetics year: 2023 pmcid: PMC10025466 doi: 10.3389/fgene.2023.1014014 license: CC BY 4.0 --- # Conditioning on parental mating types can reduce necessary assumptions for Mendelian randomization ## Abstract Mendelian randomization (MR) has become a common tool used in epidemiological studies. However, when confounding variables are correlated with the instrumental variable (in this case, a genetic/variant/marker), the estimation can remain biased even with MR. We propose conditioning on parental mating types (a function of parental genotypes) in MR to eliminate the need for one set of assumptions, thereby plausibly reducing such bias. We illustrate a situation in which the instrumental variable and confounding variables are correlated using two unlinked diallelic genetic loci: one, an instrumental variable and the other, a confounding variable. Assortative mating or population admixture can create an association between the two unlinked loci, which can violate one of the necessary assumptions for MR. We simulated datasets involving assortative mating and population admixture and analyzed them using three different methods: 1) conventional MR, 2) MR conditioning on parental genotypes, and 3) MR conditioning on parental mating types. We demonstrated that conventional MR leads to type I error rate inflation and biased estimates for cases with assortative mating or population admixtures. In the presence of non-additive effects, MR with an adjustment for parental genotypes only partially reduced the type I error rate inflation and bias. In contrast, conditioning on parental mating types in MR eliminated the type I error inflation and bias under these circumstances. Conditioning on parental mating types is a useful strategy to reduce the burden of assumptions and the potential bias in MR when the correlation between the instrument variable and confounders is due to assortative mating or population stratification but not linkage. ## Introduction Randomized experiments, often called randomized controlled trials, are the gold standard for drawing causal inferences. In randomized experiments, observational units (e.g., subjects) are randomly assigned to different levels of the variable being used to assess the causal effect, e.g., the treatment. The randomization process eliminates the influence of potential confounding variables on the exposure variable (e.g., treatment or control). Therefore, we can conclude that the observed difference in outcomes between groups in randomized controlled trials is purely caused by the treatment (barring stochastic variations). However, randomized experiments are not always ethical, feasible, or practical (Sanson-Fisher et al., 2007). Observational studies do not always yield unbiased estimates of effects because of their lack of random assignment. Of the multiple limitations that these studies have, herein, we will only consider the bias due to confounding. To mitigate confounding, researchers often include potential confounders in analyses as covariates in regression models or stratify analyses by confounders. Figure 1A depicts a general causal model with an exposure variable (X), an outcome (Y), a confounder (U), and a genetic marker (G), where U is associated with both X and Y and G determines X. The variables X, Y, and U are assumed to be continuous. Causal effects and associations are represented by directional and bidirectional arrows, respectively. If U is observable and is included in the model, the estimate of the effect of X on Y will be unbiased, provided the estimation method does not induce a bias. However, the confounder (U) is not always measurable or known. If U is a set of confounders of the relationship between X and Y and is not appropriately accounted for in the analysis, the estimator of the regression coefficient of Y on X will be biased. FIGURE 1: Causal models for Mendelian randomization. Directional and bidirectional arrows correspond to causal and associational relationships, respectively. β s are regression coefficients. Variables in rectangles and ovals correspond to measurable or unmeasurable variables, respectively. (A) Generalized model for Mendelian randomization (MR) with three assumptions. (B) Explicit causal model separating confounding variables, U , into the variable consisting of unlinked heritable variants in the nuclear genome, H , and all other confounding variables, U¯ . (C) Explicit causal model for the father–mother–offspring trio. The parental mating type, P , is the combination of parental genotypes ( Gm and Gf ), which takes one of the six possible values. The dotted line connecting Xm and Xf implies assortative mating of these variables. Mendelian randomization (MR) was proposed to address the issue of unmeasured confounders in observational studies (Smith and Ebrahim, 2003; Boutwell and Adams, 2020; Sanderson et al., 2022). MR uses genotypic (G) data from loci that affect the exposure variable (X), do not have a direct effect on the outcome, and are uncorrelated with potential confounders. The most commonly used process of estimation is as follows: 1) X is regressed on G to obtain the predicted value of X, X^; 2) Y is regressed on X^, and then, the estimated coefficient is an unbiased estimator of the effect of X on Y under some assumptions. As a simple and robust approach for causal inference, MR has become common in epidemiological studies during the last few decades. However, MR rests on three assumptions (Emdin et al., 2017): “1) the genetic variant is associated with the risk factor; 2) the genetic variant is not associated with confounders; and 3) the genetic variant influences the outcome only through the risk factor.” In Figure 1A, these assumptions correspond to the following: 1) G and X are associated, 2) there is no association between G and U, and 3) there is no direct effect of G on Y, not through X. If any of the aforementioned three assumptions are violated, the estimated effect is not guaranteed to be unbiased. Unfortunately, the violation of assumptions, especially the violation of assumption [2], is quite plausible: the genotype (G) can be associated with confounders (U). Even without a direct effect of G on U (or vice versa), assortative mating and population stratification can yield associations between them, which violate assumption [2]. Furthermore, it is hard to verify this assumption because U includes unmeasurable variables: “The second and third assumptions, however, cannot be empirically proven and require both judgment by the investigators and the performance of various sensitivity analyses” (Emdin et al., 2017). This paper proposes conditioning on parental mating types (defined as a combination of genotypes of parents at a locus used as an instrumental variable (Allison, 1997)) in MR to eliminate the bias in conventional MR, when there is correlation between the instrumental variable and confounding variables. This means that our approach obviates the need for one of the three necessary assumptions in MR. This paper consists of two parts. First, we demonstrate that the estimation using conventional MR (without conditioning on parental mating types) could lead to biased estimates, when there is a correlation between the instrumental variable and confounding variables due to assortative mating or population stratification. Second, we propose the use of parental mating types in conventional MR and to assess the utility of this approach. ## Mechanisms violating assumptions for MR: Assortative mating and population stratification First, we define the variables, parameters, and error terms used in the simulation and analyses (summarized in Table 1), with explicit mathematical expressions and causal mechanisms. There are six variables: X, Y, U, G, H, and P. X, Y, and U¯ are an exposure variable, an outcome variable, and a confounder, respectively, and all are quantitative traits (thus, continuous variables), such as weight and height. G and H are genotypes defined by SNPs; thus, they are one of the three statuses: AA,Aa,aa for G and BB,Bb,bb for H, respectively; fA* and fD* are functions to calculate the additive and dominance effect of a genotype, respectively (fA counts the number of A [or B] alleles for the genotype; fD is 1 for a heterozygote and 0 for a homozygote). P is the parental mating type, a combination of genotypes of parents at a locus used as an instrumental variable, and one of the six statuses: AA/AA,AA/Aa,AA/aa,Aa/Aa,Aa/aa,aa/aa. We introduce five indicator functions to compute the genetic effect of parental mating types, IAA/AAP,IAA/AaP,IAA/aaP,IAa/AaP,IAa/aaP, where the function is 1 if P is the same as the subscript of the function and otherwise, 0. The effect of a variable M on a variable N (i.e., the difference in N due to a single unit increase in M) is represented by βMN. It should be noted that the additive effect and the dominance effect of a genotype M on a variable N are represented as βMAN (i.e., difference in N by substituting allele A [or B] for allele a [or b]) and βMDN (i.e., deviance from the average of genotypic values of the two homozygotes), respectively. Furthermore, there are five coefficients to represent the effect of parental mating type P on a variable M using the parental mating type aa/aa as a reference group. Thus, for example, βAA/AAM is the unit increase in M for the parental mating type AA/AA compared to the increase in the parental mating type aa/aa. Estimated regression coefficients are distinguished from the causal effect using the following: β^MN is the regression coefficient estimated by regressing N on M. TABLE 1 \| Parameter \| Description \| \| --- \| --- \| \| X \| Exposure variable \| \| Y \| Outcome \| \| G \| Genotype of a locus with effects on X G∈AA,Aa,aa \| \| H \| Confounder consisting of a genotype of a locus with effects on X and Y H∈BB,Bb,bb \| \| U¯ \| All other (non-genetic) confounders (with effects on X and Y ) \| \| P \| Parental mating type on G P∈AA/AA,AA/Aa,AA/aa,Aa/Aa,Aa/aa,aa/aa \| Figure 1B is a causal model in which the confounding variable set, U, is separated into two sets of variables: one set includes a confounding variable consisting of a genotype on a single biallelic locus (H) with two alleles B and b, and the second set U¯ consists of all other confounders. We note that in our scenario, H and the genotype on the biallelic locus, used as an instrumental variable (G), are unlinked. However, G and H could be correlated (i.e., non-linkage disequilibrium), which would violate assumption [2]. The following describes two situations, assortative mating and population stratification, which can cause such a non-linkage disequilibrium. ## Situation 1: Assortative mating In human and other animal populations, the choice of a mate does not plausibly occur at random. One may be more likely to mate with another who has specific phenotypes, resulting in non-random or assortative mating (Anonymous, 1903). For example, assortative mating for body mass index (BMI) or body fatness (i.e., individuals with a high BMI or body fatness are more likely to mate with one another, as are individuals with low BMI or body fatness) is widely observed (Allison et al., 1996; Silventoinen et al., 2003; Jackson et al., 2007). We modeled assortative mating as being dependent on the exposure variable X. Mothers and fathers are separately sorted by X, and they are paired according to the order. For this purpose, each parent’s genotype, exposure, outcome, and confounders are explicitly modeled. Variables are given with one of the two subscripts, m or f, for either the mother or the father (variables without these subscripts are for an offspring). The model is summarized in Figure 1C. Briefly, the correlation between G and H is explained as follows: Assortative mating on X (i.e., Xm and Xf) induces associations between Gm and Hf and Gf and Hm, which result in an association between G and H, thus violating the MR assumption [2]. ## Situation 2: Population stratification Population stratification occurs and can create genotype–phenotype associations in the absence of linkage or a causal effect of the specific genotype on the specific phenotype, when a population consists of multiple subpopulations (Freedman et al., 2004) and some subpopulations have different allele frequencies and phenotypic distributions. By using the framework given in Figure 1C without assortative mating, we assume two different subpopulations. Therefore, within each subpopulation, three assumptions are held for conventional MR. The difference between the two populations is that they have different allele frequencies. If data from the two subpopulations were analyzed as a single population without accounting for the population substructure, they would yield a spurious association between G and H (because all parental loci [Gm, Hm, Gf, and Hf] are associated), which violates the MR assumption [2]. ## Correcting the bias in MR: Conditioning on parental mating types Assuming the aforementioned two situations, the conventional MR estimation procedure can lead to biased estimates because MR assumption [2] is violated. To eliminate the bias, we propose conditioning on the parental mating type P, which is a combination of parental genotypes used for the instrumental variable in MR. The rationale for using P is that both Gm and Gf are located on open (i.e., d connected) paths between genotypes G and H in both situations 1 and 2, and conditioning on P blocks the path. We also follow the approach of using parental genotypes instead of mating types, as proposed by Hartwig et al. [ 2018], which is another reference method. In the following, we show details of three methods: conventional MR, MR conditioning on parental genotypes [a method proposed by Hartwig et al. [ 2018]], and MR conditioning on parental mating types (which we propose in this study). It should be noted that unmeasurable variables (variables in ovals, given in Figure 1C) do not appear in any of the analyses. ## Conventional MR 1) Conventional MR uses the following model: X=β1+βGAXfAG+βGDXfDG+εX, where εX is an error term. Therefore, the auxiliary regression of X on fAG and fDG is performed to obtain the estimated value of X (= X^): X^=β^1+β^GAXfAG+β^GDXfDG.2) Then, the regression of Y on X^ is conducted by assuming the following model with an error term εY: Y=β2+βXYX^+εY. ## MR conditioning on parental genotypes To correct for the bias in MR, Hartwig et al. [ 2018] proposed conditioning on parental genotypes. The analysis proceeds as follows:1) MR conditioning on parental genotypes uses the following model: X=β1+βGAXfAG+βGDXfDG+βGmAXfAGm+βGmDXfDGm+βGfAXfAGf+βGfDXfDGf+εX. Therefore, the auxiliary regression of X on fAG and fDG conditioning on fAGm, fDGm, fAGf, and fDGf are performed to obtain the estimated value of X (= X^): X^=β^1+β^GAXfAG+β^GDXfDG+β^GmAXfAGm+β^GmDXfDGm+β^GfAXfAGf+β^GfDXfDGf.2) The regression of Y on X^ is conducted assuming the following model: Y=β2+βXYX^+βGmAYfAGm+βGmDYfDGm+βGfAYfAGf+βGfDYfDGf+εY. ## MR conditioning on parental mating types Hartwig et al. [ 2018] assumed an additive model and, thus, used a parental genotype as an instrumental variable. However, if the effect of the parental genotype on an offspring’s phenotype is non-additive, using a parental mating type, i.e., a combination of parental genotypes taking one of the six possible values (Figure 1C), is more appropriate. The corresponding analysis proceeds as follows:1) MR conditioning on parental mating types uses the following model: X=β1+βGAXfAG+βGDXfDG+βAA/AAXIAA/AAP+βAA/AaXIAA/AaP+βAA/aaX IAA/aaP,+βAa/AaXIAa/AaP+βAa/aaX IAa/aaP+εX. Therefore, the auxiliary regression of X on G1 conditioning on the parental mating type P is performed to obtain the estimated value of X (= X^): X^=β^1+β^GAXfAG+β^GDXfDG+β^AA/AAXIAA/AAP+β^AA/AaXIAA/AaP+β^AA/aaX IAA/aaP,+β^Aa/AaXIAa/AaP+β^Aa/aaX IAa/aaP. 2) The regression of Y on X^ is conducted by assuming the following model: Y=β2+βXYX^+β^AA/AAXIAA/AAP+β^AA/AaXIAA/AaP+β^AA/aaX IAA/aaP,+β^Aa/AaXIAa/AaP+β^Aa/aaX IAa/aaP+εY. ## Simulations To demonstrate the potential bias when conventional MR is used due to the violation of the MR assumption [2] and the utility of using parental mating types to eliminate the bias, we performed simulations considering assortative mating and population stratification. For the simulation of each situation, we created data for 1,000 trio (father–mother–offspring) families (500 trios each for the population for situation 2) for a single simulation and performed three different analyses on each dataset. We repeated the process 1,000 times for each parameter setting. The type I error rate (when βXY=0) is defined as the proportion of simulations in which the estimated association between X and Y is statistically significant (false-positive finding). The bias in the estimated coefficient Eβ^XY−βXY is also assessed when βXY>0. The coefficient βXY was set as 1.0 for bias assessment. The sensitivity of the type I error rate and the bias on the magnitude of the violation of MR assumption [2] were assessed by varying the parameters. The significance level was set as 0.05. The process for generating data and analyses are described in the next section. ## Simulation 1: Assortative mating The following is a step-by-step protocol and parameter setting for the simulation:1) Allele frequencies of A and B are $10\%$ for each: ProbA=ProbB=0.1,Proba=Probb=0.9. Each parent’s genotypes (Gm, Hm, Gf, and Hf) are determined assuming the Hardy–Weinberg equilibrium (Hardy, 1908). It should be noted that G and H are independent.2) The confounding variables for parents, U¯m and U¯f, are determined, which follow a bivariate normal distribution: N0,0.1.3) The exposure variables of parents, Xm and Xf, are determined by their genotype and confounding variable: Xm=β1+βGAXfAGm+βGDXfDGm+βU¯XU¯m+εX, where εX∼N0,0.1. β1 is interpreted as the genotypic effect of the genotype aa on X. Xf is determined in the same way as Xm.4) The outcome of parents, Ym and Yf, are determined by their exposure, genotype, and confounding variable: Ym=β2+βXYXm+βHAYfAHm+βHDYfDHm+βU¯YU¯m+εY, where εY∼N0,0.1. β2 is interpreted as the genotypic effect of the genotype bb on Y, when both X and U¯ are zero. Yf is determined in the same way as Ym.5) Proportion p is selected from paternal and maternal populations. In the selected population, both parents are sorted separately by the exposure Xm or Xf and are paired according to the order of Xm and Xf. Unselected parents (1- p) are randomly coupled regardless of the values of X and Y.6) The genotype of the offspring, G and H, are determined by randomly selecting an allele from each parent.7) The exposure, X, and the outcome, Y, of the offspring are determined by following the same process as for the parents (see 3 and 4). The sensitivity of the type I error rate and bias was assessed by changing p from 0.0 to 0.8. All effects from U¯ to X and Y are assumed to be 1. For genetic effects, we assumed that there is no additive effect (βGAX=βHAX=βHAY=0), but there is a strong dominance effect (βGDX=βHDX=βHDY=1) of G and H on any associated variables. ## Simulation 2: Population stratification The simulation setting for simulation 2 is similar to simulation 1 save for a couple of differences: 1) no assortative mating and 2) we assume two populations (i.e., subpopulation 1 and subpopulation 2) with different allele frequencies. Allele frequencies of A and B for subpopulation 1 are $10\%$ each. Otherwise, all simulation settings, including parameter settings, are the same as those in simulation 1. The source of the violation of MR assumption [2] is different allele frequencies. To demonstrate the sensitivity of the type I error rate and bias on the magnitude of the violation of MR assumption [2], allele frequencies of A and B for subpopulation 2 were varied from $10\%$ to $90\%$. All simulations and analyses were performed using statistical computing software R (version 3.6.1). ## Results The type I error rate for simulation 1 is shown in Figure 2A. Type I error inflation was observed for conventional MR and MR conditioning on parental genotypes, and it increased as the proportion involved in assortative mating increased. Type I error inflation was not observed for MR conditioning on parental mating types. Type I error inflation was mitigated by conditioning on parental genotypes to some extent, which still remained. The type I error rate for simulation 2 is shown in Figure 2B. Type I error inflation was observed for both conventional MR and MR conditioning on parental genotypes but not for MR conditioning on parental mating types. As shown in simulation 1, conditioning on parental genotypes reduced but did not eliminate type I error rate inflation. Interestingly, we observed a large type I error inflation when allele frequencies for the subpopulation were intermediate (0.5). This is because we assumed that homozygous genotypes (i.e., AA,aa and BB,bb) have the same effect on phenotypes (X and Y). FIGURE 2: Type I error rate and bias of estimated coefficients for three different types of MR. Open squares, open circles, and open triangles correspond to conventional MR, MR conditioning on parental mating types, and MR conditioning on parental genotypes, respectively. For simulation 1, the proportion of the population involved in assortative mating was changed from 0 to 0.8. For simulation 2, allele frequencies of A and B for subpopulation 2 were varied from 10% to 90%. (A, B) Type I error rates for simulations 1 and 2. Gray dotted lines are significance levels (= 0.05). (C, D) Bias in the estimated regression coefficient of an offspring’s outcome on exposure ( β^XY−βXY ) for simulations 1 and 2. The bias of the estimated coefficient is shown in Figures 2C, D. We observed similar results for the bias in estimation as in type I error rates. When type I error rate inflation was observed, a statistically significant bias was also observed, and magnitudes of type I error rate inflation and absolute bias were positively associated. ## Discussion MR has become a common approach for causal inference in epidemiology, as genetic data become more accessible owing to fast and efficient DNA sequencing technology and as journals and funding bodies encourage data sharing (Levey et al., 2009; Bloom et al., 2014; Loder and Groves, 2015). However, as for most epidemiological approaches, MR has essential assumptions we need to check before performing analysis. Among them, the assumption of no association between genetic variants used in MR and confounders [MR assumption [2]] could be violated or is difficult to check in practice. First, we demonstrated that MR produces inflation in type I error rates and a biased estimation in realistic settings where the assumption is violated. We introduced two plausible situations: assortative mating and population stratification. The sensitivity of type I error rates and estimation bias was assessed by changing parameters relevant to the violation of the MR assumption. As expected, we observed type I error inflation and estimation bias in these realistic settings when conventional MR was used, and such inflations and biases worsened as violations became more severe. They were mitigated by conditioning on parental genotypes to some extent; however, type I error inflation remained. Second, we proposed the use of parental mating types for a valid association inference for these two situations. We successfully confirmed that conditioning on parental mating types solves the problem in both situations. We noted that we are not the first to propose the idea of considering parental genetic information in an epidemiological study. The idea was originally proposed in testing for linkages in the presence of associations (Allison, 1997). Redden et al. suggested using parental mating types in the inference of genotype–phenotype associations (Redden and Allison, 2006). Later, Liu et al. [ 2015] extended the idea to testing causal effects of a fetal drive. In this work, they showed the relationship between this idea and MR. In MR, the genetic variant needs to be a causal variant. However, it may be difficult to verify this assumption in practice, if not impossible. Conditioning on parental mating types is one way to identify causal genetic variants, thus relaxing assumptions, specifically assumption [2] of MR (resulting in the strengthening of MR). In the context of MR, Hartwig et al. proposed using parental genotypes in the case of assortative mating, which violates MR assumption [3] (Hartwig et al., 2018). They proposed two methods to integrate parental genotypes in MR analyses. The first method is to adjust conventional MR by parental allele scores, which we used in this study. The second method is to use parental non-transmitted allele scores and the offspring allele score as instrumental variables of parental and offspring exposure variables. They demonstrated that both methods provide unbiased estimates of the exposure–outcome association and avoid type I error inflation even under strong assortative mating conditions. The difference between the study by Hartwig et al. and ours is that we assumed that the locus influencing the outcome (H) also influences the exposure (X). Therefore, their model is considered a special case of ours. Although Hartwig et al. [ 2018] concluded that only cross-trait assortative mating (between X and Y) yields a bias, we found that same-trait assortative mating (between X s or between Y s) can also yield a bias due to the heritable confounding variable (H). Furthermore, we found that conditioning of parental genotypes is not enough to control the bias if effects of alleles on phenotypes are non-additive. In our previous work (Liu et al., 2015), we indicated that random mating is not assumed with conditioning on parental mating types. We also explained that it is necessary to condition on parental mating types to achieve randomization, which is the basis for causal inference. Further insights into the rationale for this or other ways of expressing fundamental ideas can be found in the study by Pearl et al. [ 2016]. We list a few limitations of our approach. One apparent limitation is the data availability. Most genetic epidemiological research studies do not have (or is not designed to collect) parental genetic data (i.e., mother–father–offspring). However, because family trio data collection is considered to be a powerful tool for identifying rare diseases, even outside the context of MR, and owing to technological advancements in gene sequencing, the collection of family trio data may become more common (Infante-Rivard et al., 2009). In a recent study, Young et al. proposed imputing parental genotypes to reduce biases in GWA studies (Young et al., 2022). The imputation strategy presented in this study provides an opportunity to implement methods we proposed here for MR in situations where parental genotypes are not directly available. In this work, we propose that conditioning on parental genetic mating types can reduce assumptions needed for MR. We illustrate this key principle using a simulation study involving one locus with dominance effects. However, the approach we propose is general and does not require dominance effects. Indeed, our approach will also work under an additive model because the additive model is a special case of the more general model we use for conditioning. However, if the mode of action of the locus is strictly additive, conditioning on a parental allele dosage may be enough to reduce the bias. Therefore, in future studies, we plan to assess the superiority of conditioning on parental mating types relative to conditioning on allele dosages. Furthermore, we plan to assess the principle we proposed in a broader range of realistic circumstances. We are, particularly, interested in investigating two situations. The first is to evaluate the performance of the proposed approach in a multi-locus context for models involving epistatic interactions, which seem common (Zhu et al., 2015). The second situation is one where there is a selection bias on the exposure, X. Since X is a collider of G,H,and U¯, if a subpopulation was sampled according to X (people with X higher than the threshold, for example), spurious correlations among G,H,and U¯ might occur. In this case, conditioning on parental genetic mating types can account for the correlation between G and H but not for the correlation between G and U¯ because U¯ is not a heritable variable. However, regardless of the limitations suggested previously, conditioning on parental mating types in MR can strengthen assumptions and help avoid type I error inflation and bias, when a heritable confounding variable is associated with the instrumental variable in MR. ## 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 in: Zenodo (doi: 10.5281/zenodo.6972710). ## Author contributions DA conceived the research idea. KE, LM, GC, and NL implemented the simulation and performed the analyses. DA and GC oversaw data analyses. All authors were involved in the writing of the manuscript and gave final approval of submitted and published versions. ## Conflict of Interest DA and his institutions (Indiana University and the Indiana University Foundation) have received consulting fees, donations, grants, and contracts or promises for the same, from numerous not-for-profit, for-profit (including food, pharmaceutical, litigation, dietary supplement, and other entities), and government organizations with interests in health, causal inference, genetics, and statistics; however, none of these could reasonably be taken to represent a conflict of interest with this statistical methodology paper. 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. ## References 1. Allison D. B., Neale M. C., Kezis M. I., Alfonso V. C., Heshka S., Heymsfield S. B.. Assortative mating for relative weight: Genetic implications. Behav. Genet. (1996) 26 103-111. DOI: 10.1007/BF02359888 2. Allison D. B.. Transmission-disequilibrium tests for quantitative traits. Am. J. Hum. Genet. (1997) 60 676-690. PMID: 9042929 3. Assortative mating in man: A cooperative study. Biometrika (1903) 2 481-498. DOI: 10.1093/biomet/2.4.481 4. Bloom T., Ganley E., Winker M.. Data access for the open access literature: PLOS's data policy. PLoS Biol. (2014) 12 e1001797. DOI: 10.1371/journal.pbio.1001797 5. Boutwell B. B., Adams C. D.. A research note on mendelian randomization and causal inference in criminology: Promises and considerations. J. Exp. Criminol. (2020) 18 171-182. DOI: 10.1007/s11292-020-09436-9 6. Emdin C. A., Khera A. V., Kathiresan S.. Mendelian randomization. JAMA (2017) 318 1925-1926. DOI: 10.1001/jama.2017.17219 7. Freedman M. L., Reich D., Penney K. L., McDonald G. J., Mignault A. A., Patterson N.. Assessing the impact of population stratification on genetic association studies. Nat. Genet. (2004) 36 388-393. DOI: 10.1038/ng1333 8. Hardy G. H.. Mendelian propositions in a mixed population. Science (1908) 28 49-50. DOI: 10.1126/science.28.706.49 9. Hartwig F. P., Davies N. M., Davey Smith G.. Bias in Mendelian randomization due to assortative mating. Genet. Epidemiol. (2018) 42 608-620. DOI: 10.1002/gepi.22138 10. Infante-Rivard C., Mirea L., Bull S. B.. Combining case-control and case-trio data from the same population in genetic association analyses: Overview of approaches and illustration with a candidate gene study. Am. J. Epidemiol. (2009) 170 657-664. DOI: 10.1093/aje/kwp180 11. Jackson D. M., Stewart J., Djafarian K., Speakman J. R.. Assortative mating for obesity. Am. J. Clin. Nutr. (2007) 86 316-323. DOI: 10.1093/ajcn/86.2.316 12. Levey A., Stevens L., Schmid C., Zhang Y., Castro A., Feldman H.. A new equation to estimate glomerular filtration rate. Ann. Intern Med. (2009) 150 604-612. DOI: 10.7326/0003-4819-150-9-200905050-00006 13. Liu N., Archer E., Srinivasasainagendra V., Allison D. B.. A statistical framework for testing the causal effects of fetal drive. Front. Genet. (2015) 5 464. DOI: 10.3389/fgene.2014.00464 14. Loder E., Groves T.. The BMJ requires data sharing on request for all trials. BMJ (2015) 350 h2373. DOI: 10.1136/bmj.h2373 15. Pearl J., Glymour M., Jewell N. P.. Causal inference in statistics: A primer (2016) 16. Redden D. T., Allison D. B.. The effect of assortative mating upon genetic association studies: Spurious associations and population substructure in the absence of admixture. Behav. Genet. (2006) 36 678-686. DOI: 10.1007/s10519-006-9060-0 17. Sanderson E., Glymour M. M., Holmes M. V., Kang H., Morrison J., Munafò M. R.. Mendelian randomization. Nat. Rev. Methods Prim. (2022) 2 6-21. DOI: 10.1038/s43586-021-00092-5 18. Sanson-Fisher R. W., Bonevski B., Green L. W., D’Este C.. Limitations of the randomized controlled trial in evaluating population-based health interventions. Am. J. Prev. Med. (2007) 33 155-161. DOI: 10.1016/j.amepre.2007.04.007 19. Silventoinen K., Kaprio J., Lahelma E., Viken R. J., Rose R. J.. Assortative mating by body height and BMI: Finnish twins and their spouses. Am. J. Hum. Biol. (2003) 15 620-627. DOI: 10.1002/ajhb.10183 20. Smith G. D., Ebrahim S.. Mendelian randomization': Can genetic epidemiology contribute to understanding environmental determinants of disease?. Int. J. Epidemiol. (2003) 32 1-22. DOI: 10.1093/ije/dyg070 21. Young A. I., Nehzati S. M., Benonisdottir S., Okbay A., Jayashankar H., Lee C.. Mendelian imputation of parental genotypes improves estimates of direct genetic effects. Nat. Genet. (2022) 54 897-905. DOI: 10.1038/s41588-022-01085-0 22. Zhu Z., Bakshi A., Vinkhuyzen A. A., Hemani G., Lee S. H., Nolte I. M.. Dominance genetic variation contributes little to the missing heritability for human complex traits. Am. J. Hum. Genet. (2015) 96 377-385. DOI: 10.1016/j.ajhg.2015.01.001
--- title: Multigenerational mistimed feeding drives circadian reprogramming with an impaired unfolded protein response authors: - Kai Huang - Tao Zhang - Wenjun Zhang - Yue Gu - Pan Yu - Lanqing Sun - Zhiwei Liu - Tao Wang - Ying Xu journal: Frontiers in Endocrinology year: 2023 pmcid: PMC10025471 doi: 10.3389/fendo.2023.1157165 license: CC BY 4.0 --- # Multigenerational mistimed feeding drives circadian reprogramming with an impaired unfolded protein response ## Abstract Mistimed food intake in relation to the day/night cycle disrupts the synchrony of circadian rhythms in peripheral tissues and increases the risk of metabolic diseases. However, the health effects over generations have seldom been explored. Here, we established a 10-generation mouse model that was continuously fed with daytime-restricted feeding (DRF). We performed RNA-seq analysis of mouse liver samples obtained every 4 h over a 24 h period from F2, F5 and F10 generations exposed to DRF. Multigenerational DRF programs the diurnal rhythmic transcriptome through a gain or loss of diurnal rhythmicity over generations. Gene ontology (GO) analysis of the differential rhythmic transcriptome revealed that adaptation to persistent DRF is accompanied by impaired endoplasmic reticulum (ER) stress. Consistently, a substantially higher level of folding-deficient proinsulin was observed in F10 liver tissues than in F2 and F5 liver tissues following tail vein injection. Subsequently, tunicamycin induced more hepatocyte death in F10 samples than in F2 and F5 samples. These data demonstrate that mistimed food intake could produce cumulative effects over generations on ER stress sensitivity in mice. ## Introduction Lifestyle has become more irregular in industrialized nations, including night shift work, which negatively affects our overall health and wellbeing. Irregular food intake itself may drive downstream dysregulation of circadian rhythms in peripheral tissues such as the liver and is believed to contribute to the development of numerous diseases [1]. Furthermore, skipping breakfast or postponing dinner can lead to an increased risk of obesity (2–4). Most animals consume their food during their active phase of a daily cycle, which is the light phase for diurnal species and the dark phase for nocturnal species. Daytime-restricted feeding (DRF) programs glucose intolerance and impairs insulin secretion in nocturnal rat offspring [5]. Metabolic organs receive food at the wrong time with subsequent temporal disruption of anabolic and catabolic processes. These findings indicate that individuals who eat at the wrong time may have progeny with hepatic stress and dysfunction. The circadian system is a hierarchically organized multi-oscillator network consisting of a central clock located in the suprachiasmatic nucleus (SCN) and oscillators in peripheral organs [6, 7]. The SCN coordinates peripheral oscillators to align local clocks with geophysical time [8]. These clocks are responsive to environmental cues (e.g., light, food, temperature) to accommodate daily recurring environmental changes by altering clock gene expression or protein levels (9–11). Although the master clock SCN is sensitive to light, feeding and fasting are the dominant environmental cues for many other tissues, such as the liver [12]. Short-term studies of DRF in mice showed a 12 h phase shift of the expression of liver and white adipose tissue (WAT) clock genes [13]. In liver and visceral adipose tissue (VAT), $61\%$ and $80.5\%$ rhythmic genes reach the antiphase state after a 7-day DRF, which are enriched in fatty acid metabolism, complex carbohydrates metabolism, transport of lipids, and glucose metabolism [14]. Although changes in rhythmic gene expression caused by out-of-phase feeding are well described in mice, it is not known whether inverted meal timing induces a cumulative impact on rhythmic gene expression over generations. Endoplasmic reticulum (ER) is an important organelle that serves many functions, particularly in protein synthesis, folding, and transport. ER stress signaling and the unfolded protein response (UPR) are triggered to restore cellular homeostasis, but a prolonged ER stress response can activate apoptotic signals, leading to damage to the target cells. External influences, such as nutrient supply, can disrupt the function and homeostasis of the ER, contributing to the progression of metabolic diseases [15]. However, food intake with an improper dietary time and nutrient deprivation also disrupts ER homeostasis, representing an additional source of ER stress. Indeed, historical nutrition challenges influence the health and disease risk of offspring [16]. Therefore, the question remains whether ER stress dysfunction in response to inverted meal timing could show an incremental impact on further generations. In this study, we identified two clusters of rhythmic liver RNA expression that display a tendency to gain or lose rhythmicity from F2 to F10 in response to multigenerational DRF. We focused on the ER stress pathway, including the rhythmic changes in ER stress-associated genes and unfolded protein-mediated apoptotic cell death. ## Animals All C57BL/6 mice were housed in a specific pathogen-free animal facility and maintained on a 12 h light:12 h dark cycle (lights on at 8 a.m.). F1 mice were derived from ad libitum-fed (Ad) mice and assigned to DRF from weaning. Mice of the following generations were under DRF for life until they were euthanized for experiments. DRF mice had access to food for 6 h from ZT2 to ZT8 (ZT = Zeitgeber time, ZT0 is light onset, ZT12 is light off). ## Body weight, food intake, metabolic rhythm measurement, and sperm count Body weights were measured at birth and every 5 days from weaning until 2 months of age. Then, mice were placed into individual cages to record daily food intake. Consecutive mouse activity, food consumption and respiratory exchange ratio (RER) were measured by a comprehensive animal monitoring system (Oxymax; Columbus Instruments). Sperm was collected from epididymis as previously reported and calculated by MAKLER COUNTING CHAMBER [17]. ## Triglyceride measurements Blood was collected from the orbital sinus into sterile 1.5 ml tubes containing citrate sodium (3 M). Then, blood cells were removed by centrifugation at 2,000 g for 20 min at 4°C, the supernatant was immediately aliquoted and stored at -80°C. Plasma was prepared for TG analysis with 7100 automatic biochemical analyzer (Hitachi). Hepatic TG was performed according to the protocol of the TG assay kit (Nanjing Jiancheng Bioengineering Institute, A110-1-1) ## Tunicamycin administration For the ER stress model, mice were intraperitoneally injected with tunicamycin (APExBIO) dissolved in dimethyl sulfoxide (DMSO) and diluted in sterile 150 mM dextrose at a dose of 0.5 μg/g body mass at ZT8, and tissues were sampled 24 h later. ## Hydrodynamic injection Mice were anesthetized, and then, plasmid DNA suspended in sterile PBS in a volume equal to $10\%$ of the body weight was injected in 5 to 7 s via the tail vein of mice. The amount of injected plasmid was 6 μg flag-tagged folding-deficient mutant proinsulin, and tissue samples were collected 20 h after injection. ## RNA sequencing Liver samples were collected at 4 h intervals from ZT0 to ZT20 and immediately stored at -80°C. Total RNA was then extracted by the TRIzol procedure (Thermo Fisher). RNA purity and concentration were assessed by RNA electrophoresis, NanoDrop, and Agilent 2200 TapeStation analysis. RNA-seq was performed on an Illumina MiSeq platform with PE 150 bp reads at the BGI Genome Center, Shenzhen, China. Raw sequence files were subjected to quality control analysis using FastQC. Reads were mapped to the mouse reference genome GRCm38/mm10 using bowtie (v1.1.2), and expression quantification was performed using RSEM (v1.2.8). The expression values were normalized by FPKM. ## Rhythmicity assessment and functional enrichment analysis To assess rhythmicity in gene expression, we used the meta2d algorithm, a function of the MetaCycle R package (v1.2.0) [18]. $P \leq 0.05$ was considered significant. Functional enrichment analysis of circadian reprogramming genes was performed via Metascape [19]. ## Quantitative real-time PCR Total RNA from liver tissues was extracted using TRIzol reagent according to the manufacturer’s instructions. RNA (1 μg) was reverse transcribed using a PrimeScript RT Reagent Kit (TaKaRa). Quantitative real-time PCR was performed with SYBR Green detection reagent (TaKaRa). Measured values from specific genes were analyzed by the ΔCt method normalized to Actin as an endogenous control. Primer sequences for quantitative real-time PCR are listed in Table S1. ## Western blotting Protein lysates from isolated tissues were extracted using RIPA buffer with protease inhibitors. The protein concentration was measured with a BCA assay, and an equal amount of protein was separated by SDS PAGE electrophoresis and transferred to PVDF membranes. Membranes were blocked with $5\%$ milk powder in Tris-buffered saline-tween 20 (TBST) for 1 h and then incubated with specific primary antibody (in blocking solution) at 4°C overnight. Next, the membranes were incubated with horseradish peroxidase (HRP)-conjugated secondary antibodies for 1 h at RT. After several washes with TBST, the membrane was incubated with Omni ECL reagent (EpiZyme) and imaged by a ChemiScope 6200 system (CLINX). The following primary antibodies were used for the experiments. Antibodies against β-actin (1:5,000, A5441; Sigma-Aldrich), β-Tubulin (1:5,000, #5346; Cell Signaling), CHOP (1:1,000, #2895; Cell Signaling), Procaspase-3 (1:1,000, #9662; Cell Signaling), Cleaved Caspase-3 (1:1,000, #9664; Cell Signaling), BAX (1:1,000, #2772; Cell Signaling), and BCL2 (1:1,000, #3498; Cell Signaling) were used. ## Immunofluorescence assay Fresh liver tissues were rapidly frozen in liquid nitrogen and mounted in OCT compound. Then, thin sections (6 μm) were cut, mounted onto poly-L-lysine-coated glass slides, fixed in $4\%$ paraformaldehyde in PBS for 20 min at RT, and washed in PBS. Terminal transferase-mediated dUTP nick-end labeling (TUNEL) staining was performed according to the protocol of the (TUNEL) BrightRed Apoptosis Detection Kit (Vazyme, A113-01). ## Quantification and statistical analysis All data are shown as the mean ± standard deviation (SD) unless otherwise specified. The results were evaluated using unpaired (two-tailed) Student’s t test and one- or two-way analysis of variance (ANOVA). $P \leq 0.05$ was considered statistically significant ($P \leq 0.05$, $P \leq 0.01$, and $P \leq 0.001$). All statistical analysis were performed using GraphPad Prism 9 (GraphPad Software, Inc., San Diego, CA, USA). ## Altered traits of mice with multigenerational DRF To illustrate the contribution of multigenerational DRF to physiology and rhythmic gene expression, we provided a normal chow diet to mice only in the inactive period (ZT2 to ZT8; ZT = Zeitgeber time, ZT0 is light onset, ZT12 is light off) (Figure 1A). F0 mice were derived from ad libitum mice and assigned to DRF after weaning. The offspring were fed from ZT2 to ZT8 until they were euthanized for the experiments (Figure 1B). To characterize the generational effect of DRF on physiology and behaviors, we analyzed three generations (F2, F5 and F10) of DRF mice and ad libitum mice. We found that birth weight, litter size and sperm counts were similar among ad libitum and DRF groups (Figure 1C). Body weight gain in both the DRF and ad libitum groups gradually increased from 1 to 2 months of postnatal age (Figure 1D). The mice fed with DRF weighed less than the mice fed ad libitum, but there was no difference among the DRF groups (Figure 1D). The DRF groups ate less food than the ad libitum group, but there was no significant difference among generations (Figure 1E). Plasma and hepatic triglyceride (TG) levels were comparable among ad libitum and DRF groups (Figure S1A). Then, to determine whether DRF altered the diurnal rhythms of feeding and other behavior or physiological outputs, we recorded foraging behavior and energy expenditure parameters of the DRF and ad libitum-fed mice using the Comprehensive Lab Animal Monitoring System (CLAMS, IC). The mice with imposed DRF shifted their feeding behavior to match the external food accessibility with an enhanced feeding peak in the antiphase to the ad libitum-fed mice (Figure 1F). The activity pattern was disrupted by DRF, which induced additional activity during food accessibility time (Figure 1G). Consistent with the results of feeding behavior, metabolic parameters, including heat production, oxygen consumption (VO2), carbon dioxide production (VCO2) and respiratory exchange ratio (RER), exhibited diurnal fluctuation induced by DRF, which was antiphase to those of the ad libitum-fed mice (Figures S1B–S1E). Taken together, these results demonstrate that multigenerational exposure to an inverted feeding regime causes low weight with behavior and metabolic adaptation to DRF. However, the absence of a significant difference among generations indicated that these changes are driven by external feeding time. Figure 1:** Basic behavior and metabolic measurements. (A) Schematic showing the DRF schedules. Day and night are indicated above with white and gray bars. The solid red box and the clock pattern indicate the timing of food access. (B) Schematic diagram of the multigenerational DRF mouse model. The sampling time points are indicated by black arrows. (C) Body weight of neonatal mice (n = 8-9), litter size (n = 8-14) and sperm number of adult mice (n = 5). (D) Body weight recorded at 5-day intervals from 31 days to 61 days (n= 5-10) and P value from statistical hypothesis test. (E) Daily food intake (n = 5). (F, G) Activity and food consumption recorded at 20 min intervals for 3 days; red bars indicate the timing of food access (Ad group were fed ad libitum). Histograms show the daily percentage of activity and food consumption; day and night are indicated with gray and black bars (n = 4). Significance was calculated by unpaired two-tailed Student’s t test, or two-way ANOVA; P < 0.05; P < 0.01; P < 0.001 were considered significant; n.s., not significant. Error bars represent the mean ± SD. ## Reprogramming of the rhythmic liver transcriptome in multigenerational DRF To study the reprogramming of rhythmic gene transcription by DRF over generations, we collected the liver tissues of DRF mice in three generations (F2, F5 and F10) every 4 h for 24 h, and the whole liver transcriptome based on RNA-seq was analyzed, followed by the detection of rhythmic transcripts using the meta2d algorithm. We detected robust rhythmic expression in 3,116 liver transcripts in F2, of which $49.5\%$ ceased to be oscillatory in F5 (Figure 2A). Similarly, only 1,406 rhythmic genes were shared between F5 and F10 (Figure 2A). Next, Sankey diagrams were used to illustrate the dynamics of liver rhythmicity across generations (Figure 2B). Of all detected genes, 1,005 maintained rhythmicity in three generations with comparable phase and amplitude (Figures 2B; S2A-S2D). Notably, the flow diagram showed that 4,720 genes changed their oscillatory statuses by DRF in at least one generation, while 3,133 ($66.4\%$) genes displayed reversed rhythmicity between F2 and F10 (Figure 2B). These genes were grouped into two categories (loss and gain of rhythmicity) based on their meta2d_P values from the meta2d algorithm, which determines the extent of rhythmicity (Figure 2B). Furthermore, we found that the diurnal oscillation of 677 genes was dampened, and 815 genes were gained over generations (Figure 2C). Collectively, our data indicated that the oscillatory liver transcriptome was dramatically reprogrammed via multigenerational DRF. Generational gain or loss of rhythmicity occurred in more than $30\%$ of reprogrammed genes. Figure 2: Liver diurnal rhythms are reprogrammed via multigenerational DRF. (A) Venn diagram displaying the total number of rhythmic genes in the liver from ad libitum, F2, F5 and F10, including common genes. (B) Changes in gene expression rhythms from F2 to F10. Vertical red bars represent rhythmic genes(rhy), while green bars represent arrhythmic genes(ar). Horizontal red bands represent changes in rhythmic genes, while green bands represent changes in arrhythmic genes in F2. (C) Left: heatmaps display generational rhythm-loss genes satisfying the following condition: meta2d_P<0.05 in F2, meta2d_P>0.05 in F10, and meta2d_P in F5 is between meta2d_P in F2 and F10. Right: heatmaps display generationally rhythm-gain genes satisfying the following condition: meta2d_P>0.05 in F2, meta2d_P<0.05 in F10, and meta2d_P in F5 is between meta2d_P in F2 and F10. ## Clock gene expression is less affected by DRF Then, we further detected the response of the liver clock to multigenerational DRF. The expression of different core clock genes (Arntl, Dbp, Nr1d1, Per1, Per2, Cry1) was analyzed by the meta2d algorithm to detect their rhythmicity and estimate their phases and relative amplitude (Figure S3A). Except for Per1 in F10, the expression of liver clock genes was approximately 8 h phase-shifted toward the feeding time by DRF, and the shifted expression pattern was maintained from F2 to F10 (Figures S3A, S3B). Loss of circadian rhythmicity in Per1 was observed in F10 (Figure S3A). Interestingly, DRF gradually decreased the amplitude of the circadian output gene Dbp generation by generation (Figure S3A). Moreover, the phases and amplitude of core clock genes were perturbed upon DRF, and the altered expression pattern was maintained over generations. ## Functional enrichment analysis of genes with generational changes in rhythmicity To archive the functional consequences of this reprogramming of rhythmicity in the liver transcriptome, we performed gene ontology (GO) analysis. We observed a significant enrichment for genes that lost rhythmicity gradually from F2 to F10 in GO categories such as response to DNA damage and ER stress (Figure 3A). Representative genes, such as Wrnip1, Fan1 and Nabp2, which are involved in DNA damage response, exhibited lower expression across 24 h in F10, which resulted in a plateau of expression (Figure 3B). Xbp1 and Ptpn2, which are involved in ER stress, fluctuated more in F10 (Figure 3B). However, genes that exhibited a generationally gain of rhythmicity in the liver were mainly enriched in fatty acid metabolic processes (Figure 3C). DRF in F10 generated harmonic oscillation of Acadl (Figure 3D). In addition, a deeper trough, but unstable local lowest point, of diurnal expression of Hadh, Acadm, Acox3 and Aldh1l1 was observed in F10 compared to F2 (Figure 3D). Figure 3: Functional analysis of genes with generational changes in rhythmicity. (A) Gene ontology (GO) biological process enrichment analysis of generational rhythm-loss genes during multigenerational DRF. (B) RNA-seq expression profiles of representative generational rhythm-loss genes involved in DNA damage and ER stress (n = 2). (C) GO biological process enrichment analysis of genes with generational rhythm-gain during multigenerational DRF. (D) RNA-seq expression profiles of representative generational rhythm-gain genes involved in fatty acid and lipid metabolism (n = 2). P-values were calculated using the meta2d algorithm; P < 0.05; P < 0.01; P < 0.001 were considered rhythmic; n.s., not rhythmic. Error bars represent the mean ± SD. ## Multigenerational DRF exacerbates ER stress To verify that there are indeed differences in the enriched ER stress pathways between generations, we challenged ad libitum, F2, F5 and F10 mice with half of the standard dose of the ER stress-inducing agent tunicamycin (TM), which impairs protein folding by blocking N-linked glycosylation [20, 21]. First, we examined ER stress-associated gene expression in liver tissues at ZT8 24 h after TM injection. Our data showed that the mRNA levels of Xbp1 and Ptpn2 displayed pronounced upregulation in the livers of F10 mice (Figure 4A). Conversely, the response of Crebrf was attenuated from F2 to F10 (Figure 4A). Xbp1 is a key activated modulator of the UPR, while Crebrf participates in the negative regulation of the UPR [22], indicating higher activation of the UPR in the livers of F10 mice. Thus, we next examined other representative genes downstream of the UPR involved in protein folding and degradation. Hepatic transcript levels of the chaperone proteins Bip and Grp94 increased significantly, whereas changes in the critical complex of endoplasmic reticulum-associated degradation (ERAD), an integral part of the UPR [23], were the opposite (Figure 4B). Significantly increased transcript levels of the E3 ubiquitin-protein ligase Syvn1 (Hrd1) and decreased transcript levels of the adaptor protein Sel1l in the livers of F10 mice were observed (Figure 4B). Figure 4: Multigenerational DRF exacerbates ER stress. (A) Relative mRNA levels of Xbp1s, Ptpn2 and Crebrf (n = 4). (B) Relative mRNA levels of Bip, Grp94, Syvn1, and Sel1l (n = 4). (C) Protein levels of flag-tagged folding-deficient mutant proinsulin 20 h after injection (one representative figure is shown) and quantification of FLAG/ACTIN protein levels (n = 4). All significance was calculated by one-way ANOVA. P < 0.05; P < 0.01; P < 0.001 were considered significant. Error bars represent the mean ± SD. To further investigate the efficiency of the ERAD machinery, we generated labeled folding-deficient proinsulin in mouse liver by hydrodynamic tail vein injection [24]. Intriguingly, more flag-tagged misfolded proinsulin was detected in the livers of F10 mice (Figure 4C). Altogether, these results suggested that multigenerational DRF can enhance fundamental processes attempting to restore ER homeostasis and chronically affect the clearance of misfolded proteins. ## Multigenerational DRF induces hepatocyte apoptosis by ER stress In the presence of unrelieved ER stress, a switch in the UPR from a prosurvival to prodeath phenotype leads to the activation of apoptosis [25]. Thus, we examined the response of Chop to TM challenge, which is the key apoptosis regulator in ER stress and increased sharply by overwhelming ER stress [26]. TM triggered more than 2-fold increases in Chop at both the mRNA and protein levels in the livers of F10 mice (Figures 5A, B). Furthermore, our results showed that both upstream (Atf4) and downstream (Ero1a and Gadd34) of Chop were upregulated in F10 (Figure 5C). GADD34 and ERO1A also exert proapoptotic functions downstream of CHOP [27]. Thus, we analyzed hepatic apoptosis by terminal transferase-mediated dUTP nick-end labeling (TUNEL) staining. More TUNEL-positive cells were detected in the livers of F10 mice (Figure 5D). Moreover, significantly higher mRNA levels of Bax and lower levels of Bcl2 were measured in the livers of F10 mice (Figure 5E). Furthermore, the protein levels of BAX and Cleaved Caspase-3 showed an increase, while BCL2 displayed a decreasing tendency in the liver from F2 to F10 (Figures 5F, G). These results indicate that CHOP-induced apoptosis was more activated under ER stress stimulation by multigenerational DRF. Figure 5: Multigenerational DRF increases ER stress-induced hepatocyte apoptosis. (A) Relative mRNA levels of Chop (n = 4). (B) CHOP protein levels (one representative figure is shown) and quantification of CHOP/ACTIN protein levels (n = 4). (C) Relative mRNA levels of Atf4, Ero1a and Gadd34 (n = 4). (D) Representative TUNEL staining (red) of mouse liver. Nuclei were stained with DAPI (blue); scale bar, 50 μm. Quantification of the percentage of TUNEL-positive cells (n = 3). (E) Relative mRNA levels of Bax and Bcl2 (n = 4). (F) Protein levels of BAX and BCL2. (G) Protein levels of Procaspase-3 and Cleaved Caspase-3. All significance was calculated by one-way ANOVA. P < 0.05; P < 0.01; P < 0.001 were considered significant. Error bars represent the mean ± SD. Additionally, without TM challenge, the mRNA levels of Xbp1, Ptpn2, Crebrf, Bip, Grp94, Syvn1, Sel1l, Chop, Atf4, Ero1a, and Gadd34 showed similar changes compared to those under stimulation in the livers of F10 mice, suggesting that multigenerational DRF increases basic hepatic ER stress levels (Figure S4A). However, few TUNEL-positive cells could be detected, suggesting that ER stress is mild under physiological conditions (Figure S4B). Taken together, our results provide evidence that multigenerational DRF impairs the efficiency of ERAD and activates the CHOP-induced apoptosis pathway under exogenous ER stress stimulation. ## Discussion Growing evidence emphasizes the importance of eating time, and eating at the wrong time of the day, such as inverted feeding, may have deleterious effects on health. Of particular interest, little attention has been focused on the effect of successive generations of inverted feeding exposure on health. To this end, we developed a mouse model of DRF for 10 successive generations and performed a comprehensive analysis of liver rhythmic transcriptome. By analyzing RNA-seq data from the livers of F2, F5 and F10 mice, we identified two trends of changes in gene rhythmicity, which were dampened or enhanced generation by generation. Genes that lost rhythmicity generationally mainly regulate the response to ER stress, which suggests that accumulated mistimed feeding disrupts protection against environmental stimuli. Indeed, our experimental data from a mouse model of ER stress demonstrated that multigenerational mistimed feeding exacerbates liver injury, as more apoptotic hepatocytes were observed after TM stimulation. DRF is known to change the phase of the circadian clock and gene expression programs, especially in primary metabolic organs such as the liver. As previously reported, the expression rhythms of most core clock and clock-controlled genes in our research were entrained by restricted feeding [14]. Intriguingly, the liver circadian transcriptome was also distinguishable between different generations under the same diet regimen. In addition to epigenetic factors such as histone modification, factors participating in nutrient-sensing pathways may play important roles in multigenerational DRF-induced rhythm reprogramming. The circadian system and nutrient-sensing pathways interact at the whole organism to individual molecule levels. Nutrient-sensing pathways can impact the circadian clock, and conversely, the circadian system also modulates nutrient sensing and response [28]. Further research is needed to explore whether nutrient-sensing signals originating from long-term circadian misalignment are disrupted during consecutive DRF. Interestingly, most liver clock genes kept oscillation during multigenerational DRF except Per1, whose rhythmicity was lost in F10. The expression of period genes was acutely affected by refeeding, and the phase change of Per1 is the most sensitive to external stimuli [29, 30]. It is very likely that confliction between antiphase diet and circadian feedback loop might counteract the rhythmicity of Per1. ARNTL is one of crucial transcription factors that regulating the expression of Per1 [31]. Although the expression patterns of Arntl were similar among DRF groups, other posttranscriptional regulation may dampen the Per1 mRNA level. MicroRNAs have been reported to be involved in the posttranscriptional regulation of Per1 by affecting the stability of mRNA, and microRNA levels are sensitive to nutrition (32–35). Moreover, IRE1α endoribonuclease is one of three main UPR signaling branches and decreases Per1 mRNA in tumor cells without affecting transcription [36]. Our results showed that higher UPR response was detected in the liver of F10 mice. Whether lower mRNA levels of Per1 at ZT8 were resulted from higher activity of IRE1α endoribonuclease needs to be explored in the future. In our ER stress model, overwhelming ER stress was observed in the livers of F10 mice upon TM challenge. Although molecular chaperones such as Bip and Grp94 were more highly activated, CHOP-induced hepatocyte apoptosis was increased remarkably. TM induces ER stress in cells by inhibiting the first step in the biosynthesis of N-linked glycans that are necessary for protein folding and maturation. The protein glycosylation process could be impaired by DRF and produce excessive misfolded proteins. The lower efficiency of mutant protein clearance in the livers of F10 mice suggests that the function of the proteasomal degradation system as well as autophagy might be impaired over generations. Thus, multigenerational DRF triggered unbalanced production and clearance of cell debris and resulted in prolonged ER stress. We have indeed made extensive efforts to phenotyping the effect of DRF on the sperm counts, motility by sperm swimming speed, fertility by in-vitro fertilization, mating ability and litter size, but no significant difference was found among generations. Here are some possible factors that mediate the intergenerational effects of multigenerational DRF on offspring. Increasing evidence indicates that non-DNA sequence-based epigenetic information, including DNA methylation, histone modifications and non-coding RNAs, can be inherited across several generations [37]. Considering all male and female mice were kept in DRF for life, not only somatic cells but also developing progenitor germ cells might be affected. Thus, epigenetic information both from paternal and maternal may play an important role in offspring during multigenerational DRF. Epigenetic factors modulated by the environmental conditions experienced by parents have the potential to affect the zygote at fertilization, thereby regulate early embryonic development and impact the health of descendants. Although DNA methylation and histone modifications are removed during spermatogenesis and embryogenesis, there are still a minimal fraction of sperm DNA methylation and histone modifications that are preserved and contribute to inter and transgenerational effects on offspring (38–43). Some findings suggest that sperm RNAs functions in the transmission of paternal metabolic disorders associated with diet challenge, thus sperm RNAs may mediate the cumulative effects of multigenerational DRF stimulation probably [44, 45]. For maternal factors, DRF exposure exists not only during oogenesis but also during early embryonic development and postnatal lactation. As previously reported, dysfunctional mitochondria in oocyte and abnormal oocyte development have enduring effects on the long-term health of the offspring [46, 47]. Furthermore, aberrant nutritional and endocrine levels caused by exposure to adverse environmental stimuli such as undernourishment in utero and lactation results in impaired health for progenies (48–50). Thus, the effects of multigenerational DRF on oocytes, intrauterine development and lactation may contribute to the abnormal phenotypes found in F10 mice. Overall, there may be a complex mechanism to explain generational effects caused by multigenerational DRF in our study. Multiple factors including sperm DNA methylation, histone modifications and sperm RNAs, as well as mitochondria in oocytes, intrauterine development and lactation environments are possible carriers of multigenerational inheritance. In summary, this study is the first to explore the impairment of DRF at long-term and multigenerational levels to the best of our knowledge, which advances our understanding of the importance of eating time. Our results reveal that multigenerational DRF reprograms the liver circadian transcriptome over generations and induces unhealthy alterations in subsequent progenies that are mainly associated with ER stress. The underlying mechanism of this intergenerational process is intriguing and worth exploring in the future, and this process might contribute to not only social-medical but also evolutionary development. ## Data availability statement The data presented in the study are deposited in the NCBI SRA repository, accession number PRJNA932189. ## Ethics statement The animal study was reviewed and approved by the Animal Care and Use Committee of the Cambridge-Su Genomic Resource Center, Soochow University (YX-3, YX-2017-2, YX-2021-2). ## Author contributions KH, ZL, TZ, and YX are responsible for designing research; KH, ZL, WZ, LS, and PY performed research; KH, YG, TZ, and YX performed bioinformatic analysis; KH, TZ, TW, and YX analyzed data; and KH, TZ, and YX wrote 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/fendo.2023.1157165/full#supplementary-material ## References 1. Kecklund G, Axelsson J. Health consequences of shift work and insufficient sleep. Bmj (2016) 355. DOI: 10.1136/bmj.i5210 2. Woller A, Gonze D. Circadian misalignment and metabolic disorders: A story of twisted clocks. Biology (2021) 10 207. DOI: 10.3390/biology10030207 3. Vujovic N, Piron MJ, Qian J, Chellappa SL, Nedeltcheva A, Barr D. Late isocaloric eating increases hunger, decreases energy expenditure, and modifies metabolic pathways in adults with overweight and obesity. Cell Metab (2022) 34 1486-98.e7. DOI: 10.1016/j.cmet.2022.09.007 4. Yamamoto R, Tomi R, Shinzawa M, Yoshimura R, Ozaki S, Nakanishi K. Associations of skipping breakfast, lunch, and dinner with weight gain and Overweight/Obesity in university students: A retrospective cohort study. Nutrients (2021) 13 271. DOI: 10.3390/nu13010271 5. de Almeida Faria J, de Araujo TM, Mancuso RI, Meulman J, da Silva Ferreira D, Batista TM. Day-restricted feeding during pregnancy and lactation programs glucose intolerance and impaired insulin secretion in Male rat offspring. Acta Physiol (Oxf) (2016) 217. DOI: 10.1111/apha.12684 6. Mohawk JA, Green CB, Takahashi JS. Central and peripheral circadian clocks in mammals. Annu Rev Neurosci (2012) 35 445. DOI: 10.1146/annurev-neuro-060909-153128 7. Albrecht U. Timing to perfection: The biology of central and peripheral circadian clocks. Neuron (2012) 74. DOI: 10.1016/j.neuron.2012.04.006 8. Dibner C, Schibler U, Albrecht U. The mammalian circadian timing system: Organization and coordination of central and peripheral clocks. Annu Rev Physiol (2010) 72. DOI: 10.1146/annurev-physiol-021909-135821 9. Patke A, Young MW, Axelrod S. Molecular mechanisms and physiological importance of circadian rhythms. Nat Rev Mol Cell Biol (2020) 21 67-84. DOI: 10.1038/s41580-019-0179-2 10. Jagannath A, Butler R, Godinho SI, Couch Y, Brown LA, Vasudevan SR. The Crtc1-Sik1 pathway regulates entrainment of the circadian clock. Cell (2013) 154. DOI: 10.1016/j.cell.2013.08.004 11. Crosby P, Hamnett R, Putker M, Hoyle NP, Reed M, Karam CJ. Insulin/Igf-1 drives period synthesis to entrain circadian rhythms with feeding time. Cell (2019) 177 896-909.e20. DOI: 10.1016/j.cell.2019.02.017 12. Pickel L, Sung H-K. Feeding rhythms and the circadian regulation of metabolism. Front Nutr (2020) 7. DOI: 10.3389/fnut.2020.00039 13. Manella G, Sabath E, Aviram R, Dandavate V, Ezagouri S, Golik M. The liver-clock coordinates rhythmicity of peripheral tissues in response to feeding. Nat Metab (2021) 3. DOI: 10.1038/s42255-021-00395-7 14. Xin H, Deng F, Zhou M, Huang R, Ma X, Tian H. A multi-tissue multi-omics analysis reveals distinct kineztics in entrainment of diurnal transcriptomes by inverted feeding. iScience (2021) 24. DOI: 10.1016/j.isci.2021.102335 15. Ramadori G, Konstantinidou G, Venkateswaran N, Biscotti T, Morlock L, Galie M. Diet-induced unresolved er stress hinders kras-driven lung tumorigenesis. Cell Metab (2015) 21. DOI: 10.1016/j.cmet.2014.11.020 16. Fontana L, Partridge L. Promoting health and longevity through diet: From model organisms to humans. Cell (2015) 161. DOI: 10.1016/j.cell.2015.02.020 17. Gomez Montoto L, Magana C, Tourmente M, Martin-Coello J, Crespo C, Luque-Larena JJ. Sperm competition, sperm numbers and sperm quality in muroid rodents. PloS One (2011) 6. DOI: 10.1371/journal.pone.0018173 18. Wu G, Anafi RC, Hughes ME, Kornacker K, Hogenesch JB. Metacycle: An integrated r package to evaluate periodicity in Large scale data. Bioinformatics (2016) 32. DOI: 10.1093/bioinformatics/btw405 19. Zhou Y, Zhou B, Pache L, Chang M, Khodabakhshi AH, Tanaseichuk O. Metascape provides a biologist-oriented resource for the analysis of systems-level datasets. Nat Commun (2019) 10 1523. DOI: 10.1038/s41467-019-09234-6 20. Abdullahi A, Stanojcic M, Parousis A, Patsouris D, Jeschke MG. Modeling acute er stress. Shock (2017) 47. DOI: 10.1097/SHK.0000000000000759 21. Yamada H, Hirano T, Miyazaki T, Takatsuki A, Tamura G. Effects of tunicamycin on cell adhesion and biosynthesis of glycoproteins in aggregation-competent cells of dictyostelium discoideum. J Biochem (1982) 92 399-406. DOI: 10.1093/oxfordjournals.jbchem.a133946 22. Audas TE, Li Y, Liang G, Lu R. A novel protein, Luman/Creb3 recruitment factor, inhibits luman activation of the unfolded protein response. Mol Cell Biol (2008) 28. DOI: 10.1128/MCB.01439-07 23. Hwang J, Qi L. Quality control in the endoplasmic reticulum: Crosstalk between erad and upr pathways. Trends Biochem Sci (2018) 43 593-605. DOI: 10.1016/j.tibs.2018.06.005 24. Liu F, Song Y, Liu D. Hydrodynamics-based transfection in animals by systemic administration of plasmid DNA. Gene Ther (1999) 6. DOI: 10.1038/sj.gt.3300947 25. Bhat TA, Chaudhary AK, Kumar S, O’Malley J, Inigo JR, Kumar R. Endoplasmic reticulum-mediated unfolded protein response and mitochondrial apoptosis in cancer. Biochim Biophys Acta Rev Cancer (2017) 1867 58-66. DOI: 10.1016/j.bbcan.2016.12.002 26. Ron D, Habener JF. Chop, a novel developmentally regulated nuclear protein that dimerizes with transcription factors C/Ebp and lap and functions as a dominant-negative inhibitor of gene transcription. Genes Dev (1992) 6. DOI: 10.1101/gad.6.3.439 27. Hu H, Tian M, Ding C, Yu S. The C/Ebp homologous protein (Chop) transcription factor functions in endoplasmic reticulum stress-induced apoptosis and microbial infection. Front Immunol (2018) 9. DOI: 10.3389/fimmu.2018.03083 28. Chaix A, Manoogian ENC, Melkani GC, Panda S. Time-restricted eating to prevent and manage chronic metabolic diseases. Annu Rev Nutr (2019) 39 291-315. DOI: 10.1146/annurev-nutr-082018-124320 29. Vollmers C, Gill S, DiTacchio L, Pulivarthy SR, Le HD, Panda S. Time of feeding and the intrinsic circadian clock drive rhythms in hepatic gene expression. Proc Natl Acad Sci U.S.A. (2009) 106. DOI: 10.1073/pnas.0909591106 30. Park J, Lee K, Kim H, Shin H, Lee C. Endogenous circadian reporters reveal functional differences of period paralogs and the significance of period : Ck1 stable interaction. Proc Natl Acad Sci U.S.A. (2023) 120 e2212255120. DOI: 10.1073/pnas.2212255120 31. Partch CL, Green CB, Takahashi JS. Molecular architecture of the mammalian circadian clock. Trends Cell Biol (2014) 24. DOI: 10.1016/j.tcb.2013.07.002 32. Nagel R, Clijsters L, Agami R. The mirna-192/194 cluster regulates the period gene family and the circadian clock. FEBS J (2009) 276. DOI: 10.1111/j.1742-4658.2009.07229.x 33. Zhao X, Zhu X, Cheng S, Xie Y, Wang Z, Liu Y. Mir-29a/B/C regulate human circadian gene Hper1 expression by targeting its 3’utr. Acta Biochim Biophys Sin (Shanghai) (2014) 46. DOI: 10.1093/abbs/gmu007 34. Cheng HY, Papp JW, Varlamova O, Dziema H, Russell B, Curfman JP. Microrna modulation of circadian-clock period and entrainment. Neuron (2007) 54. DOI: 10.1016/j.neuron.2007.05.017 35. Ross SA, Davis CD. Microrna, nutrition, and cancer prevention. Adv Nutr (2011) 2. DOI: 10.3945/an.111.001206 36. Pluquet O, Dejeans N, Bouchecareilh M, Lhomond S, Pineau R, Higa A. Posttranscriptional regulation of Per1 underlies the oncogenic function of irealpha. Cancer Res (2013) 73. DOI: 10.1158/0008-5472.CAN-12-3989 37. Fitz-James MH, Cavalli G. Molecular mechanisms of transgenerational epigenetic inheritance. Nat Rev Genet (2022) 23. DOI: 10.1038/s41576-021-00438-5 38. Guibert S, Forne T, Weber M. Global profiling of DNA methylation erasure in mouse primordial germ cells. Genome Res (2012) 22. DOI: 10.1101/gr.130997.111 39. Hajkova P, Ancelin K, Waldmann T, Lacoste N, Lange UC, Cesari F. Chromatin dynamics during epigenetic reprogramming in the mouse germ line. Nature (2008) 452. DOI: 10.1038/nature06714 40. Wei Y, Yang CR, Wei YP, Zhao ZA, Hou Y, Schatten H. Paternally induced transgenerational inheritance of susceptibility to diabetes in mammals. Proc Natl Acad Sci U.S.A. (2014) 111. DOI: 10.1073/pnas.1321195111 41. Zheng X, Li Z, Wang G, Wang H, Zhou Y, Zhao X. Sperm epigenetic alterations contribute to inter- and transgenerational effects of paternal exposure to long-term psychological stress. Cell Discovery (2021) 7 101. DOI: 10.1038/s41421-021-00343-5 42. Lismer A, Siklenka K, Lafleur C, Dumeaux V, Kimmins S. Sperm histone H3 lysine 4 trimethylation is altered in a genetic mouse model of transgenerational epigenetic inheritance. Nucleic Acids Res (2020) 48. DOI: 10.1093/nar/gkaa712 43. Kaneshiro KR, Egelhofer TA, Rechtsteiner A, Cockrum C, Strome S. Sperm-inherited H3k27me3 epialleles are transmitted transgenerationally in cis. Proc Natl Acad Sci U.S.A. (2022) 119. DOI: 10.1073/pnas.2209471119 44. Chen Q, Yan M, Cao Z, Li X, Zhang Y, Shi J. Sperm tsrnas contribute to intergenerational inheritance of an acquired metabolic disorder. Science (2016) 351 397-400. DOI: 10.1126/science.aad7977 45. Grandjean V, Fourre S, De Abreu DA, Derieppe MA, Remy JJ, Rassoulzadegan M. Rna-mediated paternal heredity of diet-induced obesity and metabolic disorders. Sci Rep (2015) 5. DOI: 10.1038/srep18193 46. Saben JL, Boudoures AL, Asghar Z, Thompson A, Drury A, Zhang W. Maternal metabolic syndrome programs mitochondrial dysfunction. Cell Rep (2016) 16 1-8. DOI: 10.1016/j.celrep.2016.05.065 47. Chen B, Du YR, Zhu H, Sun ML, Wang C, Cheng Y. Maternal inheritance of glucose intolerance. Nature (2022) 605. DOI: 10.1038/s41586-022-04756-4 48. Radford EJ, Ito M, Shi H, Corish JA, Yamazawa K, Isganaitis E. Science (2014) 345. DOI: 10.1126/science.1255903 49. Ren J, Cheng Y, Ming ZH, Dong XY, Zhou YZ, Ding GL. Intrauterine hyperglycemia exposure results in intergenerational inheritance. Epigenet Chromatin (2018) 11 20. DOI: 10.1186/s13072-018-0192-2 50. Govic A, Bell V, Samuel A, Penman J, Paolini AG. Calorie restriction and corticosterone elevation during lactation can each modulate adult Male fear and anxiety-like behaviour. Horm Behav (2014) 66 591-601. DOI: 10.1016/j.yhbeh.2014.08.013
--- title: 'Neuroprotective effect of angiotensin II receptor blockers on the risk of incident Alzheimer’s disease: A nationwide population-based cohort study' authors: - Hyun Woo Lee - Seungyeon Kim - Youngkwon Jo - Youjin Kim - Byoung Seok Ye - Yun Mi Yu journal: Frontiers in Aging Neuroscience year: 2023 pmcid: PMC10025478 doi: 10.3389/fnagi.2023.1137197 license: CC BY 4.0 --- # Neuroprotective effect of angiotensin II receptor blockers on the risk of incident Alzheimer’s disease: A nationwide population-based cohort study ## Abstract ### Background Recent studies on renin-angiotensin system (RAS) inhibitors have reported a reduced risk of Alzheimer’s disease (AD). Nevertheless, the effect of RAS inhibitor type and blood–brain barrier (BBB) permeability on the risk of AD is still unknown. ### Objectives To assess the effects of RAS inhibitors on the risk of AD based on the type and BBB permeability and investigate the cumulative duration-response relationship. ### Methods This was a population-based retrospective cohort study using the Korean Health Insurance Review and Assessment database records from 2008 to 2019. The data of patients diagnosed with ischemic heart disease between January 2009 and June 2009 were identified for inclusion in the analyses. Propensity score matching was used to balance RAS inhibitor users with non-users. The association between the use of RAS inhibitors and incident AD was evaluated using a multivariate Cox proportional hazard regression model. The results are presented in adjusted hazard ratios (aHRs) and $95\%$ confidence intervals (CIs). ### Results Among the 57,420 matched individuals, 7,303 developed AD within the follow-up period. While the use of angiotensin-converting enzyme inhibitors (ACEIs) was not significantly associated with AD risk, the use of angiotensin II receptor blockers (ARBs) showed a significant association with reduced risk of incident AD (aHR = 0.94; $95\%$ CI = 0.90–0.99). Furthermore, the use of BBB-crossing ARBs was associated with a lower risk of AD (aHR = 0.83; $95\%$ CI = 0.78–0.88) with a cumulative duration-response relationship. A higher cumulative dose or duration of BBB-crossing ARBs was associated with a gradual decrease in AD risk (P for trend < 0.001). No significant association between the use of ACEIs and the risk of AD was observed regardless of BBB permeability. ### Conclusion Long-term use of BBB-crossing ARBs significantly reduced the risk of AD development. The finding may provide valuable insight into disease-modifying drug options for preventing AD in patients with cardiovascular diseases. ## Introduction During the past few decades, Alzheimer’s disease (AD) has emerged as a leading global health concern (Alzheimer’s Disease International, 2019). Despite constant efforts to develop new drugs, the complexity of pathologies has left no promising treatments for AD (Cummings et al., 2014; Bachurin et al., 2017). Owing to the limitations and uncertainty of current treatments, targeting modifiable risk factors for preventing AD incidence and delaying progression has gained importance in recent years. Management of hypertension with antihypertensive drugs has been recommended for the primary prevention of AD (Crous-Bou et al., 2017; Livingston et al., 2020; Hefner et al., 2021). The experimental results, however, remain controversial, and further analyzes are required to reach a general consensus on the use of antihypertensive drugs (Yasar et al., 2013; Chuang et al., 2014; Walker et al., 2020). Among the antihypertensive agent classes, renin-angiotensin system (RAS) inhibitors, particularly angiotensin-converting enzyme inhibitors (ACEIs) and angiotensin II type 1 receptor blockers (commonly used as angiotensin II receptor blockers, ARBs), have been reported to have potential benefits on reducing the risk of incident AD (Li et al., 2010; Davies et al., 2011; Barthold et al., 2018). RAS is a hormonal system responsible for regulating blood pressure (BP), fluid balance, electrolyte homeostasis, and vascular resistance. This system is mediated by angiotensin (Ang) ligands that interact with various receptors, including angiotensin II type 1 receptor (AT1R), angiotensin II type 2 receptor (AT2R), angiotensin IV receptor (AT4R), and Mas receptor (MASR; Gebre et al., 2018). In addition to the peripheral RAS, the receptors in the central nervous system (CNS) are involved in oxidative stress, neuroinflammation, and neuronal apoptosis, causing neurodegeneration (Abiodun and Ola, 2020; Xu et al., 2021). Accordingly, blood–brain barrier (BBB)-crossing RAS inhibitors that block neurodegenerative pathways may provide neuroprotective effects in brain disorders including AD. A few clinical studies have reported that CNS penetration of RAS inhibitors was associated with reduced cognitive decline and a lower conversion rate from mild cognitive impairment (MCI) to AD (Ohrui et al., 2004; Sink et al., 2009; O’Caoimh et al., 2014; Wharton et al., 2015). However, other studies have shown unclear relationship between BBB permeability and increased neuroprotection by RAS inhibitors (Hebert et al., 2013; Fazal et al., 2017). Considering that most studies focused on ACEIs and the results were inconclusive, further studies are essential in clarifying the effect of BBB-crossing RAS inhibitors, independent from blood pressure lowering effect, on the risk of AD incidence. This study aimed to assess the effects of RAS inhibitors on the risk of incident AD by using a longitudinal national health insurance database in Korea. In addition to comparing the effect of RAS inhibitors on AD based on class and BBB permeability, we further investigated the cumulative duration-response relationship. ## Study design and data source This population-based retrospective cohort study was conducted using Health Insurance Review and Assessment Service (HIRA) data collected for reimbursing healthcare providers between 2008 and 2019. In 2000, South Korea implemented a centralized single-insurer system, achieving healthcare coverage for almost the entire Korean population (Kwon, 2009). Universal coverage of health insurance allowed HIRA to develop a database containing socio-economic information and clinical details, including healthcare services, diagnoses, and prescriptions of 50 million beneficiaries (Kim et al., 2017). Restricted access to encrypted datasets is granted to generate public statistics and clinical research. This study was approved by the Institutional Review Board (IRB) of Yonsei University (IRB number: 7001988-202,004-HR-846-01E). The requirement for informed consent was waived owing to the anonymity of the data and the retrospective design of the study. ## Study population The study population consisted of Korean National Health Insurance beneficiaries aged 60 years or older who were diagnosed with ischemic heart disease (IHD) during the identification period (January 1, 2009, to June 30, 2009). The diagnosis of IHD was defined as the use of diagnostic codes for angina or myocardial infarction (MI) according to the International Classification of Diseases 10th Revision (ICD-10): I20.0-I22.0. Study participants were followed from the index date (July 1, 2009) to the occurrence of the outcome, the date of death, or the end of the claims record (December 31, 2019), whichever occurred first. Users of RAS inhibitors were defined as patients with the first prescription record of RAS inhibitors during the identification period. Patients who had never been prescribed RAS inhibitors during the study period were classified as non-users. Patients were excluded from the study based on the following criteria: [1] RAS inhibitor use prior to the identification period; [2] first use of RAS inhibitors after the identification period; [3] death record or last claims record prior to follow-up; and [4] diagnosis of dementia (F00-F03, G30-G31), MCI (F06.7, R41), or Parkinson’s disease (G20) before the follow-up. The outcome variable was assessed with a 1 year lag time to control and minimize reverse causality (Rea et al., 2005). ## Exposure assessment The RAS inhibitors included in this study were ACEIs and ARBs, based on the Anatomic Therapeutic Chemical (ATC) classification system provided by the World Health Organization (WHO) Collaborating Center for Drug Statistics Methodology (WHO Collaborating Centre for Drug Statistics Methodology, 2021). RAS inhibitor users were further categorized into the following four subgroups according to the type of RAS inhibitors and BBB permeability: poor BBB-crossing ACEIs (alacepril, benazepril, cilazapril, enalapril, imidapril, moexipril, and quinapril), BBB-crossing ACEIs (captopril, delapril, fosinopril, lisinopril, perindopril, ramipril, temocapril, trandolapril, and zofenopril), poor BBB-crossing ARBs (eprosartan, irbesartan, losartan, and olmesartan), and BBB-crossing ARBs (azilsartan, candesartan, fimasartan, telmisartan, and valsartan). Categorization was based on available evidence from previous analysis and review research. Drugs were considered poor BBB-crossing if the BBB permeability were inconclusive and/or if they showed relatively low lipophilicity (Oka et al., 1988; Takai et al., 2004; Sink et al., 2009; Wharton et al., 2012; Michel et al., 2013; Yagi et al., 2013; Kim et al., 2015; Alzahrani et al., 2020; Ho et al., 2021; Ouk et al., 2021; Jo et al., 2022). The dosage, frequency, and prescription days of RAS inhibitors during the identification and follow-up periods were multiplied and used as the cumulative dose. The cumulative doses were then converted into the cumulative defined daily dose (DDD) using the ATC-DDD toolkit provided by the WHO (WHO Collaborating Centre for Drug Statistics Methodology, 2021), as the claims database does not provide information regarding individual body weight. DDD is the assumed average maintenance dose per day for a drug used for its main indication in adults. The cumulative exposure duration and daily equivalent dose of RAS inhibitors were defined as the sum of the prescription days and ratio of cumulative DDD to cumulative exposure duration, respectively. ## Definition of outcome The primary outcome was the incidence of AD during the follow-up period. To enhance the accuracy of AD outcome measurement, new onset AD was defined as the presence of two or more prescription records of any AD treatment drug with an AD diagnostic code (ICD-10 F00, G30) generated from neurology or psychiatry department. Diagnostic codes obtained without the restriction on departments were included in the sensitivity analysis. The AD treatment drugs included donepezil, galantamine, rivastigmine, and memantine, which have been approved by the Food and Drug Administration for AD. Aducanumab was excluded because it was newly approved in June 2021. The outcome date was defined as the first occurrence of an AD diagnostic code within the follow-up period. ## Covariates Sociodemographic data, including patients’ age, sex, and type of insurance (health insurance and medical aid), were collected during the identification period. Comorbid diseases and concomitant medications were recorded up to the date of outcome, death, or last claims record. Patients were defined as having comorbid diseases or concurrent medications when diagnostic or drug codes appeared annually during the follow-up period. In this study, comorbid diseases reported to be potential risk factors of AD included atrial fibrillation (AF), atherosclerosis, bipolar disorder, cerebrovascular disease (hemorrhagic infarction, ischemic cortical infarction, and vasculopathy), depression, diabetes mellitus (DM), dyslipidemia, hypertension, Parkinson’s disease (PD), schizophrenia, sleep disorder, traumatic brain injury (TBI), and vascular dementia (VD; Profenno et al., 2010; Ballard et al., 2011; Xu et al., 2015). A list of ICD-10 codes for comorbid diseases is presented in Supplementary Table S1. Concomitant medications reported as potential protective or risk factors of AD included antidepressants, antiepileptics, antihistamines, antiparkinsonian agents, antipsychotics, antispasmodics, beta-blockers, benzodiazepines, bladder antimuscarinics, dihydropyridine calcium channel blockers (CCB-D), non-dihydropyridine calcium channel blockers (CCB-ND), 3-hydroxy-3-methylglutaryl coenzyme A (HMG-CoA) reductase inhibitors, skeletal muscle relaxants, and zolpidem (Carnahan et al., 2006; Risacher et al., 2016; American Geriatrics Society Beers Criteria® Update Expert Panel, 2019). Diseases or medications with a frequency of less than 30 in the population were excluded from the covariates (Yu et al., 2017). A detailed list of the concomitant medications is provided in Supplementary Table S2. ## Statistical analyzes We adopted propensity score matching (PSM) to control covariate imbalance and minimize treatment assignment bias with a multivariate logistic regression model. RAS inhibitor users were matched to non-users in a 1:1 ratio with no replacement, using the greedy matching method. The caliper width was set to 0.2 of the pooled standard deviation of the logit of the propensity score (Austin, 2011). The matching variables included age, sex, type of insurance, follow-up duration, comorbid diseases, and concurrent medications. PSM was validated by performing balance diagnostics using standardized mean difference (SMD). The absolute value of the SMD less than 0.1 was considered well balanced. Graphical distributions of propensity scores in cohorts before and after matching are presented in Supplementary Figure S1. Demographics and clinical characteristics of RAS inhibitor users and non-users were compared using descriptive statistics. Categorical variables were presented as frequencies with percentages using Pearson’s chi-squared test. Continuous variables were described as mean ± standard deviation (SD) using the Student’s t-test. The incidence of AD per 1,000 person-years was calculated by dividing the number of incident AD cases by the total follow-up person-years and multiplying the rate by 1,000. The proportional hazard assumption for the BBB-crossing ARBs was graphically validated using the log minus log plot of the Kaplan–Meier estimation (Supplementary Figure S2). The hazard ratio (HR) and $95\%$ confidence interval (CI) of the AD incidence were estimated using multivariate Cox proportional hazard regression models adjusted for sex, age, insurance type, follow-up period, comorbid diseases, and concurrent medications. The adjusted hazard ratio (aHR) between RAS inhibitor use and incident AD was computed based on RAS type and BBB permeability. Further analyzes of cumulative dose and duration-dependent responses were performed according to the cumulative DDD, cumulative exposure duration, and daily equivalent dose with a trend test using the Cox model. The cumulative hazard of AD is also graphically shown using Kaplan–Meier curves. Subgroup analyzes to identify the effect of RAS inhibitors on AD based on the type and BBB permeability within the sex were conducted. Sensitivity analyzes were conducted with four different designs for the index date, lag time, outcome definition, and exclusion criteria. First, the index date was shifted to July 1, 2010, and July 1, 2011. Second, the lag time of the outcome was extended to 3 and 5 years. Incidents during the lag time were excluded. Third, the AD incidence was re-defined with the following definitions: [1] AD diagnostic code (no restriction on departments) with at least two or more AD treatment prescriptions; and [2] AD diagnostic code with a neurology or psychiatry department subject code. Finally, the exclusion criteria were expanded with the following conditions for the entire study period 2008–2019: [1] presence of PD diagnostic code, and [2] concurrent use of ACEI and ARB. The study population was re-matched, and the outcome, comorbidity, and exposure to medications were re-assessed based on the new four designs of sensitivity analysis. Statistical analyzes were performed using the SAS software (version 9.4; SAS Institute, Cary, NC, United States). Pooled estimation with a value of $p \leq 0.05$ was considered significant. ## Baseline characteristics The cohort of patients diagnosed with IHD between January 2009 and June 2009 consisted of 537,116 participants. After the eligibility assessment and PSM, 57,420 patients with 490,384 person-years were identified for the analysis (Figure 1). The demographic and clinical characteristics of the study population are summarized in Table 1. The mean ± SD age was 69.6 ± 6.8 years and 69.9 ± 6.8 for the matched RAS inhibitor users and non-users, respectively; the proportion of the male participants was 45.0 and $46.2\%$ in RAS inhibitor users and non-users, respectively. The mean ± SD follow-up duration was 8.6 ± 2.9 years for RAS inhibitors users and 8.5 ± 2.9 years for non-users. Among the comorbid diseases, hypertension and dyslipidemia had a high prevalence of 61.0 and $48.0\%$, respectively. Figure 1: Flow chart of study population inclusion. AD, Alzheimer’s disease; MCI, mild cognitive impairment; PD, Parkinson’s disease; RAS, renin-angiotensin system; VD, vascular dementia. TABLE_PLACEHOLDER:Table 1 Before matching, significant differences in baseline characteristics were observed for the type of insurance, hypertension, DM, AF, and CCB-ND. Post-PSM results showed that the absolute SMD values for all variables were remodeled below 0.1. This indicated that the differences between the covariates were statistically well balanced. ## Renin-angiotensin system inhibitor use and AD risk A total of 7,303 new AD cases were observed, with an overall incidence of 14.9 per 1,000 person-years. The median time of censoring was 10.5 years (interquartile range, 8.3–10.5 years) in patients with AD. The use of RAS inhibitors was not significantly associated with AD risk. Females showed an increased risk of AD. The population aged 65 years or older had an increased risk of AD compared to those aged 60–64 years (aHR = 2.61; $95\%$ CI = 2.41–2.83), and the risk was even greater in those aged 80 years or older (aHR = 5.71; $95\%$ CI = 5.16–6.31). Comorbid diseases including atherosclerosis, AF, bipolar disease, cerebrovascular disease, depression, DM, dyslipidemia, hypertension, PD, schizophrenia, sleep disorder, TBI, and VD were all significant risk factors for AD incidence. Prescriptions of antidepressants, antiepileptics, antihistamines, antiparkinsonian agents, antipsychotics, antispasmodics, benzodiazepines, beta-blockers, benzodiazepines, bladder antimuscarinics, dihydropyridine CCB-D, CCB-ND, skeletal muscle relaxants, and zolpidem were also associated with an increased risk of AD. HMG-CoA reductase inhibitors were observed to be significant protective factors against the incidence of AD (Table 2). Table 2 \| Characteristics \| Number of subjects \| Person-years \| Number of events \| Incidence ratea \| Unadjusted HR (95% CI) \| Adjusted HR (95% CI)b \| value of p \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| RAS inhibitors \| \| \| \| \| \| \| \| \| Non-users \| 28710.0 \| 244738.0 \| 3689.0 \| 15.07 \| Ref. \| Ref. \| \| \| Users \| 28710.0 \| 245646.0 \| 3614.0 \| 14.71 \| 0.97 (0.93–1.02) \| 0.99 (0.94–1.03) \| 0.5886 \| \| Sex \| \| \| \| \| \| \| \| \| Men \| 26205.0 \| 220155.0 \| 2499.0 \| 11.35 \| Ref. \| Ref. \| \| \| Women \| 31215.0 \| 270229.0 \| 4804.0 \| 17.78 \| 1.55 (1.48–1.63) \| 1.29 (1.22–1.35) \| <0.0001 \| \| Age \| \| \| \| \| \| \| \| \| Under 65 \| 14406.0 \| 139775.0 \| 672.0 \| 4.81 \| Ref. \| Ref. \| \| \| Between 65 and 80 \| 37607.0 \| 318706.0 \| 5616.0 \| 17.62 \| 3.87 (3.57–4.19) \| 2.61 (2.41–2.83) \| <0.0001 \| \| Over 80 \| 5407.0 \| 31903.0 \| 1015.0 \| 31.82 \| 8.36 (7.58–9.22) \| 5.71 (5.16–6.31) \| <0.0001 \| \| Insurance type \| \| \| \| \| \| \| \| \| Health insurance \| 52408.0 \| 451995.0 \| 6484.0 \| 14.35 \| Ref. \| Ref. \| \| \| Medical aid \| 5012.0 \| 38389.0 \| 819.0 \| 21.33 \| 1.55 (1.45–1.67) \| 1.06 (0.98–1.14) \| 0.1288 \| \| Comorbid diseases \| \| \| \| \| \| \| \| \| Atherosclerosis \| 3801.0 \| 30269.0 \| 933.0 \| 30.82 \| 2.28 (2.13–2.44) \| 1.14 (1.06–1.22) \| 0.0003 \| \| Atrial fibrillation \| 5533.0 \| 42740.0 \| 1022.0 \| 23.91 \| 1.78 (1.66–1.90) \| 1.08 (1.01–1.16) \| 0.0243 \| \| Bipolar disorder \| 3695.0 \| 25224.0 \| 2160.0 \| 85.63 \| 9.02 (8.57–9.49) \| 1.21 (1.13–1.30) \| <0.0001 \| \| Cerebrovascular disease \| 11919.0 \| 86778.0 \| 3625.0 \| 41.77 \| 4.97 (4.75–5.21) \| 2.16 (2.05–2.27) \| <0.0001 \| \| Depression \| 8247.0 \| 58952.0 \| 3637.0 \| 61.69 \| 8.03 (7.67–8.41) \| 2.54 (2.40–2.69) \| <0.0001 \| \| Diabetes mellitus \| 16932.0 \| 135354.0 \| 3302.0 \| 24.4 \| 2.23 (2.13–2.34) \| 1.40 (1.33–1.47) \| <0.0001 \| \| Dyslipidemia \| 27533.0 \| 237258.0 \| 4798.0 \| 20.22 \| 2.03 (1.93–2.13) \| 1.46 (1.38–1.56) \| <0.0001 \| \| Hypertension \| 34998.0 \| 289386.0 \| 5526.0 \| 19.1 \| 2.21 (2.10–2.33) \| 1.34 (1.27–1.42) \| <0.0001 \| \| Parkinson’s disease \| 1675.0 \| 12199.0 \| 829.0 \| 67.96 \| 5.48 (5.10–5.89) \| 1.31 (1.21–1.42) \| <0.0001 \| \| Schizophrenia \| 1752.0 \| 10923.0 \| 1119.0 \| 102.44 \| 9.57 (8.97–10.21) \| 1.18 (1.10–1.27) \| <0.0001 \| \| Sleep disorder \| 6705.0 \| 46658.0 \| 2284.0 \| 48.95 \| 4.77 (4.54–5.01) \| 1.48 (1.40–1.57) \| <0.0001 \| \| Traumatic brain injury \| 1791.0 \| 12303.0 \| 753.0 \| 61.2 \| 4.95 (4.59–5.34) \| 1.31 (1.21–1.42) \| <0.0001 \| \| Vascular dementia \| 10013.0 \| 76196.0 \| 4254.0 \| 55.83 \| 8.15 (7.78–8.54) \| 3.17 (3.01–3.34) \| <0.0001 \| \| Concurrent medications \| \| \| \| \| \| \| \| \| Antidepressants \| 2533.0 \| 17400.0 \| 1132.0 \| 65.06 \| 5.46 (5.13–5.82) \| 1.21 (1.13–1.30) \| <0.0001 \| \| Antiepileptics \| 405.0 \| 2710.0 \| 189.0 \| 69.74 \| 5.27 (4.56–6.09) \| 1.36 (1.17–1.57) \| <0.0001 \| \| Antihistamines \| 2251.0 \| 13229.0 \| 979.0 \| 74.0 \| 6.54 (6.11–6.992) \| 2.19 (2.04–2.36) \| <0.0001 \| \| Antiparkinsonian agents \| 570.0 \| 3601.0 \| 381.0 \| 105.8 \| 8.60 (7.75–9.53) \| 1.28 (1.14–1.43) \| <0.0001 \| \| Antipsychotics \| 3691.0 \| 23943.0 \| 2353.0 \| 98.28 \| 12.50 (11.90–13.13) \| 2.12 (1.97–2.28) \| <0.0001 \| \| Antispasmodics \| 589.0 \| 2561.0 \| 463.0 \| 180.79 \| 18.49 (16.81–20.34) \| 2.69 (2.43–2.97) \| <0.0001 \| \| Benzodiazepines \| 4354.0 \| 28465.0 \| 1756.0 \| 61.69 \| 5.75 (5.45–6.07) \| 1.48 (1.39–1.57) \| <0.0001 \| \| Beta–blockers \| 14498.0 \| 120723.0 \| 2506.0 \| 20.76 \| 1.61 (1.54–1.69) \| 1.13 (1.07–1.18) \| <0.0001 \| \| Bladder antimuscarinics \| 2422.0 \| 16746.0 \| 1177.0 \| 70.29 \| 5.96 (5.59–6.34) \| 1.95 (1.82–2.08) \| <0.0001 \| \| CCB (Dihydropyridine) \| 15813.0 \| 135062.0 \| 2794.0 \| 20.69 \| 1.62 (1.55–1.70) \| 1.09 (1.04–1.15) \| 0.0009 \| \| CCB (Non–dihydropyridine) \| 5128.0 \| 43905.0 \| 828.0 \| 18.86 \| 1.30 (1.21–1.39) \| 1.10 (1.03–1.19) \| 0.0091 \| \| HMG–CoA reductase inhibitors \| 25227.0 \| 226480.0 \| 3922.0 \| 17.32 \| 1.31 (1.25–1.37) \| 0.83 (0.79–0.88) \| <0.0001 \| \| Skeletal muscle relaxants \| 771.0 \| 4191.0 \| 549.0 \| 130.99 \| 11.68 (10.70–12.75) \| 1.95 (1.78–2.15) \| <0.0001 \| \| Zolpidem \| 2380.0 \| 14746.0 \| 1224.0 \| 83.01 \| 7.55 (7.10–8.03) \| 1.32 (1.23–1.42) \| <0.0001 \| A significant reduction in the risk of AD was observed with ARB use (aHR = 0.94; $95\%$ CI = 0.90–0.99). ACEIs did not show significant association with the AD incidence (aHR = 1.03; $95\%$ CI = 0.97–1.10). Regarding the risk of AD based on the type of RAS inhibitor and BBB permeability, only the use of BBB-crossing ARBs demonstrated a significant protective effect (aHR = 0.83; $95\%$ CI = 0.78–0.88; Table 3). BBB-crossing ARBs were subdivided for additional analyzes based on cumulative DDD, cumulative exposure duration, and daily equivalent dose. Responses depending on cumulative DDD, and duration were observed at 1-year intervals (Supplementary Table S3). Longer exposure to BBB-crossing ARBs was significantly associated with a gradual reduction in AD risk in the trend analysis ($p \leq 0.001$). The Kaplan–Meier curves of the cumulative hazard according to the cumulative DDD of BBB-crossing ARBs with a 2-year interval are shown in Figure 2. Both ≥4 years of cumulative DDD and ≥ 4 years of cumulative exposure duration showed significantly reduced AD incidence, regardless of daily equivalent dose (Figure 3). The subgroup analysis showed that ARBs were superior to ACEIs in AD risk both in men and women, and there was no difference in the protective effect of BBB-crossing ARBs between men and women (Supplementary Table S4). ## Sensitivity analyzes Sensitivity analyzes for the AD risk of BBB-crossing ARB use for ≥4 DDD years are presented in Table 4. In all sensitivity analyzes with index date shift, lag time extension, outcome definition change, and exclusion criteria expansion, the aHRs of incident AD in users of BBB-crossing ARBs ≥4 DDD-years remained significantly lower than those in RAS inhibitor non-users. Sensitivity analyzes for AD risk of BBB-crossing ARB use <4 DDD-years are shown in Supplementary Table S5. Table 4 \| Unnamed: 0 \| Number of subjects \| Person-years \| Number of events \| Incidence ratea \| Adjusted HRs (95% CI)b \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| Index date shift \| Index date shift \| Index date shift \| Index date shift \| Index date shift \| Index date shift \| \| July 1, 2009 (main) \| 6869 \| 67460 \| 507 \| 7.52 \| 0.59 (0.53–0.65) \| \| July 1, 2010 \| 2387 \| 21451 \| 174 \| 8.11 \| 0.62 (0.53–0.73) \| \| July 1, 2011 \| 1729 \| 13977 \| 121 \| 8.66 \| 0.68 (0.56–0.83) \| \| Lag time extension \| Lag time extension \| Lag time extension \| Lag time extension \| Lag time extension \| Lag time extension \| \| 1-year lagged (main) \| 6869 \| 67460 \| 507 \| 7.52 \| 0.59 (0.53–0.65) \| \| 3-year lagged \| 6347 \| 62795 \| 435 \| 6.93 \| 0.64 (0.57–0.71) \| \| 5-year lagged \| 5516 \| 55517 \| 316 \| 5.69 \| 0.73 (0.64–0.83) \| \| Outcome definition switch \| Outcome definition switch \| Outcome definition switch \| Outcome definition switch \| Outcome definition switch \| Outcome definition switch \| \| ICD-10 + neuropsychiatry subject code + ≥2 drug prescriptions (main) \| 6869 \| 67460 \| 507 \| 7.52 \| 0.59 (0.53–0.65) \| \| ICD-10 + ≥2 drug prescriptions \| 6853 \| 66973 \| 663 \| 9.90 \| 0.59 (0.55–0.65) \| \| ICD-10 + neuropsychiatry subject code \| 6830 \| 66802 \| 655 \| 9.81 \| 0.59 (0.54–0.64) \| \| Exclusion criteria expansion \| Exclusion criteria expansion \| Exclusion criteria expansion \| Exclusion criteria expansion \| Exclusion criteria expansion \| Exclusion criteria expansion \| \| Main \| 6869 \| 67460 \| 507 \| 7.52 \| 0.59 (0.53–0.65) \| \| Main + presence of PD diagnostic code \| 6607 \| 64984 \| 454 \| 6.99 \| 0.59 (0.53–0.65) \| \| Main + concurrent use of ACEI and ARB \| 6459 \| 63550 \| 479 \| 7.54 \| 0.61 (0.55–0.67) \| ## Discussion As the protective effect of antihypertensive agents on cognitive decline beyond their blood pressure-lowering effects has emerged, the potential effect of reducing the risk of AD via the renin-angiotensin system has been demonstrated in animal and human studies (Li et al., 2010; Davies et al., 2011; Barthold et al., 2018; Abiodun and Ola, 2020). However, the neuroprotective effects of RAS inhibitors reported in previous studies have been conflicting (Ohrui et al., 2004; Sink et al., 2009; Hebert et al., 2013; Hsu et al., 2013; O’Caoimh et al., 2014; Qiu et al., 2014; Wharton et al., 2015). In this nationwide population-based cohort study, patients with IHD who used BBB-crossing ARBs had a lower risk of incident AD than those who did not use RAS inhibitors. Notably, our study showed a significant reduction in the risk of incident AD in patients who used BBB-permeable ARBs at higher cumulative doses. While previous studies have focused on comparing the effects of ARBs and ACEIs (Marcum et al., 2022) or BBB permeability within RAS inhibitors (Hebert et al., 2013; Qiu et al., 2014; Ho et al., 2021), to the best of our knowledge, this is the first study to simultaneously assess risk reduction considering both BBB permeability and cumulative doses. Our results were robust owing to a valid study design with a long-term follow-up based on nationwide study samples with appropriate comparisons and sensitivity analyzes. In this study, we revealed that the use of ARBs, but not ACEIs, was associated with a reduced risk of AD. This result is consistent with some clinical studies that have shown the advantageous effect of ARBs over ACEIs in reducing AD risk (Li et al., 2010; Davies et al., 2011; Barthold et al., 2018; Marcum et al., 2022). Additionally, a number of animal studies have supported this difference in protective effect by suggesting potential underlying mechanisms. ACEIs target the angiotensin-converting enzyme (ACE), which is responsible for converting Ang I to Ang II, thus attenuating AT1R and AT2R activation, whereas ARBs selectively block the Ang II/AT1R axis (Gebre et al., 2018). AT1R activation induces oxidative stress, neuroinflammation, and apoptosis, whereas AT2R counteracts AT1R-mediated neurodegeneration by various mechanisms (Lanz et al., 2010; Faraco et al., 2016; Abiodun and Ola, 2020). Hence, the blockade of AT1R by ARBs may induce indirect activation of the Ang II/AT2R axis to provide neuroprotection (Mogi and Horiuchi, 2013). Enhancement of cognitive function by direct stimulation of AT2R was also demonstrated in the animal study (Jing et al., 2012). Moreover, the benefits of ARBs can be attributed to the conversion of Ang II into Ang IV and Ang (1–7), which are selective for AT4R and MASR, respectively. AT4R has been suggested to have a positive effect on cerebral blood flow, memory, and neuroprotection (Näveri et al., 1994; Royea and Hamel, 2020), whereas enhancement of the Ang (1–7)/MASR axis has been reported to have a potential anti-inflammatory effect and facilitate hippocampal long-term potentiation (Hellner et al., 2005; Wright and Harding, 2019). Several studies have also found that the expression of ACE and ACE2 is related to a decreased amyloid-beta (Aβ) load (Hemming and Selkoe, 2005; Zou et al., 2007; Kehoe et al., 2016). However, further translational investigation is essential to confirm the association between the neuroprotective effects of ARBs and Aβ pathology (Loera-Valencia et al., 2021). Our analyzes indicated that the use of BBB-crossing ARBs was associated with a reduced risk of AD. The finding was especially significant as CI did not overlap with those of the other three types of RAS inhibitors. Additional benefit of using BBB-crossing RAS inhibitors has been demonstrated in previous studies assessing the effect on cognitive function and MCI to AD conversion (Wharton et al., 2015; Ouk et al., 2021). Another longitudinal study, which investigated the effect of ACEIs depending on central exposure, reported no significant association between BBB-crossing ACEIs and AD and a risk of incident AD with poor BBB-crossing ACEIs (Sink et al., 2009). However, a previous meta-analysis (Ho et al., 2021) assessing the effect of BBB-crossing RAS inhibitors on seven cognitive domains reported that poor BBB-crossing RAS inhibitors demonstrated a better effect in the attention domain compared to that of BBB-crossing RAS inhibitors. As previous studies on cognitive decline and incident AD by BBB permeability of RAS inhibitors largely focused on ACEIs (Ohrui et al., 2004; Sink et al., 2009; Hebert et al., 2013; O’Caoimh et al., 2014; Ouk et al., 2021), our results on BBB-crossing ARBs are noteworthy, but the benefit of using BBB-crossing ARBs as potential drugs for preventing AD should be carefully interpreted. In addition to BBB-crossing effects, some in vitro/vivo and animal studies have reported that some BBB-crossing ARBs, such as telmisartan, showed partial peroxisome proliferator-activated receptors (PPAR) gamma activation effects that have beneficial effects on cognitive functions (Mogi et al., 2008; Pang et al., 2012; Garg et al., 2021). However, the PPAR-gamma activation effect of ARBs is still controversial, with limited evidence for the attenuation of cognitive functions in only some ARBs (Benson et al., 2004; Erbe et al., 2006; Kajiya et al., 2011). Therefore, further comprehensive studies that have considered the PPAR-gamma binding affinities on ARBs, as well as the BBB-crossing characteristics, are needed. Remarkably, the significantly reduced risk of AD by BBB-crossing ARBs was robust in patients with a larger cumulative dose or longer duration, regardless of the daily equivalent dose. These results implied that the cumulative exposure duration was a more crucial factor in the neuroprotective effect of ARBs than the daily exposure dose. Risk-reducing effect of ARBs on AD with larger cumulative dose and longer exposure were also demonstrated in a previous longitudinal study, supporting our findings (Chiu et al., 2014). Considering that antihypertensive drugs are generally used for an extended period and our results showed a cumulative effect of BBB-crossing ARBs on AD, they could be suggested as promising targets for drug repurposing. The current treatment for AD shows modest effects only on symptoms (Atri, 2019; Cummings et al., 2020), and the efficacy of the newly approved drug, aducanumab, is also controversial (Whitehouse et al., 2022). Moreover, midlife hypertension has been associated with an increased risk of AD, and blood pressure control is a modifiable risk factor for cognitive decline (Lennon et al., 2019). Taken together, BBB-crossing ARBs might be a promising disease-modifying drug option for reducing the risk of AD in patients with cardiovascular diseases, such as hypertension. In the subgroup analysis, no difference in the protective effect of BBB-crossing ARBs was identified between men and women. A study by Barthold et al. reported that ARBs were superior to ACEIs in risk of AD incidence for white men and women, but no association was observed for the black and Hispanic populations (Barthold et al., 2018). Estrogen lowers AT1R expression, prevents the production and action of angiotensin II, and decreases NADPH-oxidase activity and expression of neuroinflammatory markers (De Silva and Faraci, 2012; O’Hagan et al., 2012; Rodriguez-Perez et al., 2015). Aging men with aromatization of androgens to estrogens have a higher estrogen level than that of aging women with dramatic ovarian loss of 17β-estradiol (Rosario et al., 2011). A study on the pathophysiology of sex differences in the protective effect of ARB owing to race is needed. Moreover, further studies in other Asian countries are needed to confirm the sex difference in the protective effect of ARBs in Asians. This study has several limitations. First, this study used a secondary claims database; therefore, we could not verify detailed clinical information, including symptoms, body weight, blood pressure, smoking, alcohol intake, education level, and genetic factors, such as APOE ε4. In addition, the limitation related to the accuracy of incident AD needs to be considered because the outcome variable was identified based on ICD-10 diagnostic codes. Given that the diagnosis of AD is based on the patient’s symptoms (Atri, 2019) and the protective effects of RAS inhibitors have been reported to vary according to cognitive symptoms (Ho et al., 2021), our results should be carefully interpreted. However, we attempted to use the medication prescription claims along with the diagnostic codes for enhanced accuracy of outcome definitions. Moreover, we adopted various outcome definitions in sensitivity analyzes to confirm the robustness of the study results. Second, the possibility of a selection bias cannot be neglected, as our study population was selected based on a very short identification period of 6 months. Moreover, we could not consider active comparators and make a direct comparison of the effect of drugs with different mechanisms, such as beta-blockers, CCBs, or thiazides, because antihypertensive agents are usually used in combination. To minimize this selection bias, we balanced RAS inhibitor users and non-users by PS matching and adjusted for various confounders using rigorous definitions. Moreover, our sensitivity analysis by shifting the index date provided comparable results. Third, it is difficult to generalize the study results to the entire population, as our study population included patients with IHD, who have a high cardiovascular profile. RAS inhibitors, possessing strong vascular effects, has been used for treating various cardiovascular diseases, and conflicting results have been reported on the neuroprotective effect of ARBs, depending on the study population. Further research on the effects of RAS inhibitors on AD in patients with various cardiovascular diseases is required. Finally, the duration or cumulative doses of concomitant medications could not be considered in our study. Instead of considering the variability of confounder status by using time-varying Cox regression, this study used the precise definition of comorbidities and concurrent medications that appeared at least once every year during the follow-up period. To the best of our knowledge, this is the first longitudinal study to demonstrate the effect of BBB-crossing ARBs on the incidence of AD with cumulative dose and duration subgroups using a population-based cohort. In this study, we highlighted the neuroprotective effect of ARBs, particularly BBB-crossing ARBs, on AD. Additionally, we present a novel finding of the protective effects against AD conferred by long-term use of BBB-crossing ARBs. In addition to existing evidence, these results are expected to provide valuable insights for AD-targeted drug development. ## Data availability statement The datasets presented in this article are not readily available because the primary data analyzed in this study are handled and stored by the Health Insurance Review and Assessment Service. Requests to access the datasets should be directed to Health Insurance Review and Assessment Service, https://www.hira.or.kr. ## Ethics statement The studies involving human participants were reviewed and approved by Institutional Review Board (IRB) of Yonsei University. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements. ## Author contributions HWL and SK contributed for study design, data analysis, data interpretation, and writing of the manuscript. YMY contributed for study conceptualization, data interpretation, critical revision of the manuscript, and supervision of the study. YJ and YK contributed to data analysis and manuscript revision. BSY contributed to clinical interpretation of the data and critical revision of the manuscript. All authors contributed to the article and approved the submitted version. ## Funding This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Ministry of Science and ICT (MSIT) of the Korea government (No. 2020R1G1A110120513). ## 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/fnagi.2023.1137197/full#supplementary-material ## References 1. Abiodun O. A., Ola M. S.. Role of brain renin angiotensin system in neurodegeneration: an update. Saudi J. Biol. Sci. (2020) 27 905-912. DOI: 10.1016/j.sjbs.2020.01.026 2. Alzahrani Y. M., Alim A. S. M. A., Kamel F. O., Ramadan W. S., Alzahrani Y. A.. Possible combined effect of perindopril and Azilsartan in an experimental model of dementia in rats. Saudi Pharm. J. (2020) 28 574-581. DOI: 10.1016/j.jsps.2020.03.009 3. Alzheimer’s Disease International. (2019). World Alzheimer Report 2019: Attitudes to Dementia. London: Alzheimer’s Disease International.. World Alzheimer Report 2019: Attitudes to Dementia (2019) 4. American Geriatrics Society 2019 updated Ags beers criteria® for potentially inappropriate medication use in older adults. J. Am. Geriatr. Soc. (2019) 67 674-694. DOI: 10.1111/jgs.15767 5. Atri A.. The Alzheimer's disease clinical Spectrum: diagnosis and management. Med. Clin. North Am. (2019) 103 263-293. DOI: 10.1016/j.mcna.2018.10.009 6. Austin P. C.. Optimal caliper widths for propensity-score matching when estimating differences in means and differences in proportions in observational studies. Pharm. Stat. (2011) 10 150-161. DOI: 10.1002/pst.433 7. Bachurin S. O., Bovina E. V., Ustyugov A. A.. Drugs in clinical trials for Alzheimer's disease: the major trends. Med. Res. Rev. (2017) 37 1186-1225. DOI: 10.1002/med.21434 8. Ballard C., Gauthier S., Corbett A., Brayne C., Aarsland D., Jones E.. Alzheimer's disease. Lancet (2011) 377 1019-1031. DOI: 10.1016/S0140-6736(10)61349-9 9. Barthold D., Joyce G., Wharton W., Kehoe P., Zissimopoulos J.. The association of multiple anti-hypertensive medication classes with Alzheimer's disease incidence across sex, race, and ethnicity. PLoS One (2018) 13 e0206705. DOI: 10.1371/journal.pone.0206705 10. Benson S. C., Pershadsingh H. A., Ho C. I., Chittiboyina A., Desai P., Pravenec M.. Identification of Telmisartan as a unique angiotensin ii receptor antagonist with selective Ppargamma-modulating activity. Hypertension (2004) 43 993-1002. DOI: 10.1161/01.HYP.0000123072.34629.57 11. Carnahan R. M., Lund B. C., Perry P. J., Pollock B. G., Culp K. R.. The anticholinergic drug scale as a measure of drug-related anticholinergic burden: associations with serum anticholinergic activity. J. Clin. Pharmacol. (2006) 46 1481-1486. DOI: 10.1177/0091270006292126 12. Chiu W. C., Ho W. C., Lin M. H., Lee H. H., Yeh Y. C., Wang J. D.. Angiotension receptor blockers reduce the risk of dementia. J. Hypertens. (2014) 32 938-947. DOI: 10.1097/HJH.0000000000000086 13. Chuang Y. F., Breitner J. C. S., Chiu Y. L., Khachaturian A., Hayden K., Corcoran C.. Use of diuretics is associated with reduced risk of Alzheimer's disease: the Cache County study. Neurobiol. Aging (2014) 35 2429-2435. DOI: 10.1016/j.neurobiolaging.2014.05.002 14. Crous-Bou M., Minguillõn C., Gramunt N., Molinuevo J. L.. Alzheimer’s disease prevention: from risk factors to early intervention. Alzheimers Res. Ther. (2017) 9 71. DOI: 10.1186/s13195-017-0297-z 15. Cummings J., Lee G., Ritter A., Sabbagh M., Zhong K.. Alzheimer’s disease drug development pipeline: 2020. Alzheimers Dement. (2020) 6 E12050. DOI: 10.1002/trc2.12050 16. Cummings J. L., Morstorf T., Zhong K.. Alzheimer’s disease drug-development pipeline: few candidates, frequent failures. Alzheimers Res. Ther. (2014) 6 37. DOI: 10.1186/alzrt269 17. Davies N. M., Kehoe P. G., Ben-Shlomo Y., Martin R. M.. Associations of anti-hypertensive treatments with Alzheimer's disease, vascular dementia, and other dementias. J. Alzheimers Dis. (2011) 26 699-708. DOI: 10.3233/JAD-2011-110347 18. De Silva T. M., Faraci F. M.. Effects of angiotensin ii on the cerebral circulation: role of oxidative stress. Front. Physiol. (2012) 3 484. DOI: 10.3389/fphys.2012.00484 19. Erbe D. V., Gartrell K., Zhang Y. L., Suri V., Kirincich S. J., Will S.. Molecular activation of Ppargamma by angiotensin ii type 1-receptor antagonists. Vasc. Pharmacol. (2006) 45 154-162. DOI: 10.1016/j.vph.2006.05.002 20. Faraco G., Sugiyama Y., Lane D., Garcia-Bonilla L., Chang H., Santisteban M. M.. Perivascular macrophages mediate the neurovascular and cognitive dysfunction associated with hypertension. J. Clin. Invest. (2016) 126 4674-4689. DOI: 10.1172/JCI86950 21. Fazal K., Perera G., Khondoker M., Howard R., Stewart R.. Associations of centrally acting ace inhibitors with cognitive decline and survival in Alzheimer's disease. BJPsych Open (2017) 3 158-164. DOI: 10.1192/bjpo.bp.116.004184 22. Garg S., Khan S. I., Malhotra R. K., Sharma M. K., Kumar M., Kaur P.. Cardioprotective effects of Azilsartan compared with that of Telmisartan on an in vivo model of myocardial ischemia-reperfusion injury. J. Biochem. Mol. Toxicol. (2021) 35 E22785. DOI: 10.1002/jbt.22785 23. Gebre A. K., Altaye B. M., Atey T. M., Tuem K. B., Berhe D. F.. Targeting Renin-Angiotensin System Against Alzheimer's Disease. Front. Pharmacol. (2018) 9 440. DOI: 10.3389/fphar.2018.00440 24. Hebert P. L., Mcbean A. M., O'connor H., Frank B., Good C., Maciejewski M. L.. Time until incident dementia among Medicare beneficiaries using centrally acting or non-centrally acting ace inhibitors. Pharmacoepidemiol. Drug Saf. (2013) 22 641-648. DOI: 10.1002/pds.3449 25. Hefner M., Baliga V., Amphay K., Ramos D., Hegde V.. Cardiometabolic modification of amyloid Beta in Alzheimer's disease pathology. Front. Aging Neurosci. (2021) 13 721858. DOI: 10.3389/fnagi.2021.721858 26. Hellner K., Walther T., Schubert M., Albrecht D.. Angiotensin-(1-7) enhances Ltp in the hippocampus through the G-protein-coupled receptor mas. Mol. Cell. Neurosci. (2005) 29 427-435. DOI: 10.1016/j.mcn.2005.03.012 27. Hemming M. L., Selkoe D. J.. Amyloid Beta-protein is degraded by cellular angiotensin-converting enzyme (ace) and elevated by an ace inhibitor. J. Biol. Chem. (2005) 280 37644-37650. DOI: 10.1074/jbc.M508460200 28. Ho J. K., Moriarty F., Manly J. J., Larson E. B., Evans D. A., Rajan K. B.. Blood-brain barrier crossing renin-angiotensin drugs and cognition in the elderly: a meta-analysis. Hypertension (2021) 78 629-643. DOI: 10.1161/HYPERTENSIONAHA.121.17049 29. Hsu C. Y., Huang C. C., Chan W. L., Huang P. H., Chiang C. H., Chen T. J.. Angiotensin-receptor blockers and risk of Alzheimer’s disease in hypertension population. Circ. J. (2013) 77 405-410. DOI: 10.1253/circj.CJ-12-0658 30. Jing F., Mogi M., Sakata A., Iwanami J., Tsukuda K., Ohshima K.. Direct stimulation of angiotensin ii type 2 receptor enhances spatial memory. J. Cereb. Blood Flow Metab. (2012) 32 248-255. DOI: 10.1038/jcbfm.2011.133 31. Jo Y., Kim S., Ye B. S., Lee E., Yu Y. M.. Protective effect of renin-angiotensin system inhibitors on Parkinson’s disease: a Nationwide cohort study. Front. Pharmacol. (2022) 13 837890. DOI: 10.3389/fphar.2022.837890 32. Kajiya T., Ho C., Wang J., Vilardi R., Kurtz T. W.. Molecular and cellular effects of Azilsartan: a new generation angiotensin ii receptor blocker. J. Hypertens. (2011) 29 2476-2483. DOI: 10.1097/HJH.0b013e32834c46fd 33. Kehoe P. G., Wong S., Al Mulhim N., Palmer L. E., Miners J. S.. Angiotensin-converting enzyme 2 is reduced in Alzheimer's disease in association with increasing amyloid-Β and tau pathology. Alzheimers Res. Ther. (2016) 8 50. DOI: 10.1186/s13195-016-0217-7 34. Kim C. K., Yang X. L., Kim Y. J., Choi I. Y., Jeong H. G., Park H. K.. Effect of long-term treatment with Fimasartan on transient focal ischemia in rat brain. Biomed. Res. Int. (2015) 2015 295925. DOI: 10.1155/2015/295925 35. Kim J.-A., Yoon S., Kim L.-Y., Kim D.-S.. Towards actualizing the value potential of Korea health insurance review and assessment (Hira) data as a resource for Health Research: strengths, limitations, applications, and strategies for optimal use of Hira data. J. Korean Med. Sci. (2017) 32 718-728. DOI: 10.3346/jkms.2017.32.5.718 36. Kwon S.. Thirty years of National Health Insurance in South Korea: lessons for achieving universal health care coverage. Health Policy Plan. (2009) 24 63-71. DOI: 10.1093/heapol/czn037 37. Lanz T. V., Ding Z., Ho P. P., Luo J., Agrawal A. N., Srinagesh H.. Angiotensin ii sustains brain inflammation in mice via Tgf-Beta. J. Clin. Invest. (2010) 120 2782-2794. DOI: 10.1172/JCI41709 38. Lennon M. J., Makkar S. R., Crawford J. D., Sachdev P. S.. Midlife hypertension and Alzheimer's disease: a systematic review and meta-analysis. J. Alzheimers Dis. (2019) 71 307-316. DOI: 10.3233/JAD-190474 39. Li N. C., Lee A., Whitmer R. A., Kivipelto M., Lawler E., Kazis L. E.. Use of angiotensin receptor blockers and risk of dementia in a predominantly male population: prospective cohort analysis. BMJ (2010) 340 b5465. DOI: 10.1136/bmj.b5465 40. Livingston G., Huntley J., Sommerlad A., Ames D., Ballard C., Banerjee S.. Dementia prevention, intervention, and care: 2020 report of the lancet commission. Lancet (2020) 396 413-446. DOI: 10.1016/S0140-6736(20)30367-6 41. Loera-Valencia R., Eroli F., Garcia-Ptacek S., Maioli S.. Brain renin-angiotensin system as novel and potential therapeutic target for Alzheimer's disease. Int. J. Mol. Sci. (2021) 22 10139. DOI: 10.3390/ijms221810139 42. Marcum Z. A., Cohen J. B., Zhang C., Derington C. G., Greene T. H., Ghazi L.. Association of Antihypertensives that stimulate vs inhibit types 2 and 4 angiotensin ii receptors with cognitive impairment. JAMA Netw. Open (2022) 5 E2145319. DOI: 10.1001/jamanetworkopen.2021.45319 43. Michel M. C., Foster C., Brunner H. R., Liu L.. A systematic comparison of the properties of clinically used angiotensin ii type 1 receptor antagonists. Pharmacol. Rev. (2013) 65 809-848. DOI: 10.1124/pr.112.007278 44. Mogi M., Horiuchi M.. Effect of angiotensin ii type 2 receptor on stroke, cognitive impairment and neurodegenerative diseases. Geriatr Gerontol Int (2013) 13 13-18. DOI: 10.1111/j.1447-0594.2012.00900.x 45. Mogi M., Li J. M., Tsukuda K., Iwanami J., Min L. J., Sakata A.. Telmisartan prevented cognitive decline partly due to Ppar-gamma activation. Biochem. Biophys. Res. Commun. (2008) 375 446-449. DOI: 10.1016/j.bbrc.2008.08.032 46. Näveri L., Strømberg C., Saavedra J. M.. Angiotensin iv reverses the acute cerebral blood flow reduction after experimental subarachnoid hemorrhage in the rat. J. Cereb. Blood Flow Metab. (1994) 14 1096-1099. DOI: 10.1038/jcbfm.1994.143 47. O’Caoimh R., Healy L., Gao Y., Svendrovski A., Kerins D. M., Eustace J.. Effects of centrally acting angiotensin converting enzyme inhibitors on functional decline in patients with Alzheimer's disease. J. Alzheimers Dis. (2014) 40 595-603. DOI: 10.3233/JAD-131694 48. O’Hagan T. S., Wharton W., Kehoe P. G.. Interactions between Oestrogen and the renin angiotensin system–potential mechanisms for gender differences in Alzheimer's disease. Am. J. Neurodegener. Dis. (2012) 1 266-279. PMID: 23383397 49. Ohrui T., Matsui T., Yamaya M., Arai H., Ebihara S., Maruyama M.. Angiotensin-converting enzyme inhibitors and incidence of Alzheimer's disease in Japan. J. Am. Geriatr. Soc. (2004) 52 649-650. DOI: 10.1111/j.1532-5415.2004.52178_7.x 50. Oka Y., Nishikawa K., Kito G., Mayahara H., Tanayama S., Muto H.. Delapril. Cardiovasc. Drug Rev. (1988) 6 192-205. DOI: 10.1111/j.1527-3466.1988.tb00376.x 51. Ouk M., Wu C. Y., Rabin J. S., Jackson A., Edwards J. D., Ramirez J.. The use of angiotensin-converting enzyme inhibitors vs. angiotensin receptor blockers and cognitive decline in Alzheimer’s disease: the importance of blood-brain barrier penetration and APOE ε4 carrier status. Alzheimers Res. Ther. (2021) 13 43. DOI: 10.1186/s13195-021-00778-8 52. Pang T., Benicky J., Wang J., Orecna M., Sanchez-Lemus E., Saavedra J. M.. Telmisartan ameliorates lipopolysaccharide-induced innate immune response through peroxisome proliferator-activated receptor-Γ activation in human monocytes. J. Hypertens. (2012) 30 87-96. DOI: 10.1097/HJH.0b013e32834dde5f 53. Profenno L. A., Porsteinsson A. P., Faraone S. V.. Meta-analysis of Alzheimer's disease risk with obesity, diabetes, and related disorders. Biol. Psychiatry (2010) 67 505-512. DOI: 10.1016/j.biopsych.2009.02.013 54. Qiu W. W. Q., Lai A., Mon T., Mwamburi M., Taylor W., Rosenzweig J.. Angiotensin converting enzyme inhibitors and Alzheimer disease in the presence of the apolipoprotein E4 allele. Am. J. Geriatr. Psychiatr. (2014) 22 177-185. DOI: 10.1016/j.jagp.2012.08.017 55. Rea T. D., Breitner J. C., Psaty B. M., Fitzpatrick A. L., Lopez O. L., Newman A. B.. Statin use and the risk of incident dementia: the cardiovascular health study. Arch. Neurol. (2005) 62 1047-1051. DOI: 10.1001/archneur.62.7.1047 56. Risacher S. L., Mcdonald B. C., Tallman E. F., West J. D., Farlow M. R., Unverzagt F. W.. Association between anticholinergic medication use and cognition, brain metabolism, and brain atrophy in cognitively Normal older adults. JAMA Neurol. (2016) 73 721-732. DOI: 10.1001/jamaneurol.2016.0580 57. Rodriguez-Perez A. I., Borrajo A., Valenzuela R., Lanciego J. L., Labandeira-Garcia J. L.. Critical period for dopaminergic neuroprotection by hormonal replacement in menopausal rats. Neurobiol. Aging (2015) 36 1194-1208. DOI: 10.1016/j.neurobiolaging.2014.10.028 58. Rosario E. R., Chang L., Head E. H., Stanczyk F. Z., Pike C. J.. Brain levels of sex steroid hormones in men and women during Normal aging and in Alzheimer's disease. Neurobiol. Aging (2011) 32 604-613. DOI: 10.1016/j.neurobiolaging.2009.04.008 59. Royea J., Hamel E.. Brain angiotensin ii and angiotensin iv receptors as potential Alzheimer's disease therapeutic targets. Geroscience (2020) 42 1237-1256. DOI: 10.1007/s11357-020-00231-y 60. Sink K. M., Leng X., Williamson J., Kritchevsky S. B., Yaffe K., Kuller L.. Angiotensin-converting enzyme inhibitors and cognitive decline in older adults with hypertension: results from the cardiovascular health study. Arch. Intern. Med. (2009) 169 1195-1202. DOI: 10.1001/archinternmed.2009.175 61. Takai S., Jin D., Sakaguchi M., Miyazaki M.. Significant target organs for hypertension and cardiac hypertrophy by angiotensin-converting enzyme inhibitors. Hypertens. Res. (2004) 27 213-219. DOI: 10.1291/hypres.27.213 62. Walker V. M., Davies N. M., Martin R. M., Kehoe P. G.. Comparison of antihypertensive drug classes for dementia prevention. Epidemiology (2020) 31 852-859. DOI: 10.1097/EDE.0000000000001245 63. Wharton W., Goldstein F. C., Zhao L., Steenland K., Levey A. I., Hajjar I.. Modulation of renin-angiotensin system may slow conversion from mild cognitive impairment to Alzheimer's disease. J. Am. Geriatr. Soc. (2015) 63 1749-1756. DOI: 10.1111/jgs.13627 64. Wharton W., Stein J. H., Korcarz C., Sachs J., Olson S. R., Zetterberg H.. The effects of Ramipril in individuals at risk for Alzheimer's disease: results of a pilot clinical trial. J. Alzheimers Dis. (2012) 32 147-156. DOI: 10.3233/JAD-2012-120763 65. Whitehouse P., Gandy S., Saini V., George D. R., Larson E. B., Alexander G. C.. Making the case for accelerated withdrawal of Aducanumab. J. Alzheimers Dis. (2022) 87 1003-1007. DOI: 10.3233/JAD-220262 66. WHO Collaborating Centre for Drug Statistics Methodology. (2021). Atc/Ddd index 2022. World Health Organization. Available at: Https://Www.Whocc.No/Atc_Ddd_Index/ (Accessed February 10, 2022).. (2021) 67. Wright J. W., Harding J. W.. Contributions by the brain renin-angiotensin system to memory, cognition, and Alzheimer's disease. J. Alzheimers Dis. (2019) 67 469-480. DOI: 10.3233/JAD-181035 68. Xu X., Du L., Jiang J., Yang M., Wang Z., Wang Y.. Microglial Trem2 mitigates inflammatory responses and neuronal apoptosis in angiotensin ii-induced hypertension in middle-aged mice. Front. Aging Neurosci. (2021) 13 716917. DOI: 10.3389/fnagi.2021.716917 69. Xu W., Tan L., Wang H. F., Jiang T., Tan M. S., Tan L.. Meta-analysis of modifiable risk factors for Alzheimer's disease. J. Neurol. Neurosurg. Psychiatry (2015) 86 1299-1306. DOI: 10.1136/jnnp-2015-310548 70. Yagi S., Akaike M., Ise T., Ueda Y., Iwase T., Sata M.. Renin-angiotensin-aldosterone system has a pivotal role in cognitive impairment. Hypertens. Res. (2013) 36 753-758. DOI: 10.1038/hr.2013.51 71. Yasar S., Xia J., Yao W., Furberg C. D., Xue Q. L., Mercado C. I.. Antihypertensive drugs decrease risk of Alzheimer disease: ginkgo evaluation of memory study. Neurology (2013) 81 896-903. DOI: 10.1212/WNL.0b013e3182a35228 72. Yu Y. M., Lee J. Y., Lee E.. Access to anti-osteoporosis medication after hip fracture in Korean elderly patients. Maturitas (2017) 103 54-59. DOI: 10.1016/j.maturitas.2017.06.021 73. Zou K., Yamaguchi H., Akatsu H., Sakamoto T., Ko M., Mizoguchi K.. Angiotensin-converting enzyme converts amyloid Beta-protein 1-42 (Abeta(1-42)) to Abeta(1-40), and its inhibition enhances brain Abeta deposition. J. Neurosci. (2007) 27 8628-8635. DOI: 10.1523/JNEUROSCI.1549-07.2007
--- title: Niclosamide does not modulate airway epithelial function through blocking of the calcium activated chloride channel, TMEM16A authors: - Henry Danahay - Sarah Lilley - Kathryn Adley - Holly Charlton - Roy Fox - Martin Gosling journal: Frontiers in Pharmacology year: 2023 pmcid: PMC10025480 doi: 10.3389/fphar.2023.1142342 license: CC BY 4.0 --- # Niclosamide does not modulate airway epithelial function through blocking of the calcium activated chloride channel, TMEM16A ## Abstract Niclosamide and benzbromarone have been described as inhibitors of the calcium activated chloride channel, TMEM16A, and on this basis have been considered and tested as clinical candidates for the treatment of airway diseases. However, both compounds have previously demonstrated activity on a range of additional biological targets and it is unclear from the literature to what extent any activity on TMEM16A may contribute to efficacy in these models of airway disease. The aim of the present study was therefore to examine the pharmacology and selectivity of these clinical candidates together with a structurally unrelated TMEM16A blocker, Ani9, in a range of functional assays to better appreciate the putative role of TMEM16A in the regulation of both epithelial ion transport and the development of an airway epithelial mucus secretory phenoptype. Benzbromarone and Ani9 both attenuated recombinant TMEM16A activity in patch clamp studies, whereas in contrast, niclosamide induced a paradoxical potentiation of the TMEM16A-mediated current. Niclosamide and benzbromarone were also demonstrated to attenuate receptor-dependent increases in intracellular Ca2+ levels ([Ca2+]i) which likely contributed to their concomitant attenuation of the Ca2+-stimulated short-circuit current responses of FRT-TMEM16A and primary human bronchial epithelial (HBE) cells. In contrast, Ani9 attenuated the Ca2+-stimulated short-circuit current responses of both cell systems without influencing [Ca2+]i which supports a true channel blocking mechanism for this compound. Additional studies using HBE cells revealed effects of both niclosamide and benzbromarone on global ion transport processes (absorptive and secretory) as well as signs of toxicity (elevated LDH levels, loss of transepithelial resistance) that were not shared by Ani9. Ani9 also failed to influence the IL-13 induced differentiation of HBE towards a goblet cell rich, mucus hypersecreting epithelium, whereas niclosamide and benzbromarone attenuated numbers of both goblet and multiciliated cells, that would be consistent with cellular toxicity. Together these data challenge the description of niclosamide as a TMEM16A blocker and illustrate a range of off-target effects of both niclosamide and benzbromarone which may contribute to the reported activity in models of airway function. ## Introduction The TMEM16A protein (anoctamin-1) has been established as a calcium activated chloride channel (CaCC) in both recombinant and native cell systems (Caputo et al., 2008; Schroeder et al., 2008; Yang et al., 2008). Physiological roles for TMEM16A protein function have been described in numerous processes including the regulation of ion transport, smooth muscle contraction, neuronal conduction and cell proliferation (reviewed in Pedemonte & Galietta, 2014). To this end, therapeutic opportunities may exist in the regulation of TMEM16A function for the treatment of a host of human diseases including cystic fibrosis, asthma, pulmonary hypertension, secretory diarrhoea and cancer. Low molecular weight compounds have been identified which are reported as selective blockers of TMEM16A function and include: T16Ainh-A01, CaCCinh-A01, MONNA, niclosamide, benzbromarone and Ani9 (De La Fuente et al., 2008; Namkung et al., 2011; Huang et al., 2012; Seo et al., 2016; Miner et al., 2019). These compounds have been widely utilised to support the identification of the biological activities associated with TMEM16A. In the case of both benzbromarone and niclosamide, repurposing of these agents as inhibitors of TMEM16A for the treatment of pulmonary hypertension and respiratory diseases have been proposed and clinical studies performed (Cabrita et al., 2019; Miner et al., 2019; Papp et al., 2019; Singh et al., 2022). The selectivity of some of these compounds for TMEM16A has however been recently questioned. For example, T16Ainh-A01, CaCCinh-A01, and MONNA were demonstrated to relax rodent arteries with a mechanism of action that was independent of CaCC activity (Boedtkjer et al., 2015). Additional pharmacological activities for these three compounds include inhibition of TMEM16B (ANO2) and volume-regulated anion channels (VRAC) (Seo et al., 2016) as well as effects on the regulation of intracellular Ca2+ ([Ca2+]i) homeostasis (Cabrita et al., 2017; Genovese et al., 2022). Understanding the selectivity of niclosamide and benzbromarone are important to enable the interpretation of the pre-clinical data supporting the repurposing of these agents. The aim of the present study was to test the hypothesis that benzbromarone, niclosamide and Ani9 would all selectively block TMEM16A ion channel function and that this activity would be independent of effects on intracellular Ca2+ levels. Furthermore, the effects of these agents on models of airway epithelial function would be tested to support the utility of benzbromarone and niclosamide as potential candidates for therapeutic repurposing. ## Methods Unless otherwise stated, all chemicals were purchased from Sigma (Poole, Dorset, United Kingdom) and cell culture/immunofluorescence reagents from Thermo Fisher Scientific (Inchinnan, Renfrew, United Kingdom). ## Cell culture FRT-hTMEM16Aabc cells were provided by Dr Luis Galietta (Genoa, IT) and were cultured as previously described for either QPatch or ion transport studies (Caputo et al., 2008). In brief, cells were cultured in Coon’s modified Ham’s F-12 medium. Media was supplemented with $10\%$ fetal calf serum, 2 mM L-glutamine, 100 U/mL penicillin, 100 μg/mL streptomycin, G418 (750 μg/mL) and sodium bicarbonate ($0.26\%$). Cells were seeded onto polycarbonate Snapwell inserts (0.5 × 106 cells/insert) and used between 4 and 10 days after seeding. Cells were fed every 48 h and the day before an ion transport experiment. Human bronchial epithelial (HBE) cells were provided by Dr Scott Randell from both the University of North Carolina Chapel Hill collection and the Cystic Fibrosis Foundation Therapeutics repository. Normal and CF (CF-HBE) cells were cultured at air-liquid interface as previously described (Danahay et al., 2020a). In brief, following an expansion step on plastic, HBE were seeded onto either Snapwell or Transwell inserts in submerged culture followed by 2 weeks at air-liquid interface in DMEM:F12 media supplemented with Ultroser G ($2\%$ v/v; Pall, Cergy, France). At all stages of culture, cells were maintained at 37°C in $5\%$ CO2 in an air incubator. ## Whole-cell patch clamp assay Whole-cell voltage-clamp recordings from FRT-hTMEM16Aabc, HEK-hTMEM16Aacd, and HEK293 cells were made using the QPatch planar patch clamp system as described previously (Bill et al., 2015; Danahay et al, 2020a). TMEM16A currents were assessed using chloride selective solutions with calculated free [Ca2+]i tightly buffered at either 415 or 0 nM. Current rectification ratios were calculated by dividing the magnitude of the outward current at +90 mV by the magnitude of the inward current at −90 mV. FRT-TMEM16Aabc cells, with [Ca2+]i clamped at 415 nM, displayed large outwardly rectifying currents with slow activation and inactivation kinetics in whole-cell patch clamp recordings (Figure 1), the key biophysical characteristics of TMEM16A. In vehicle ($0.3\%$ v/v DMSO) treated cells, the mean peak outward currents at +90 mV slowly declined from 204 ± 30 pA/pF to 150 ± 23 pA/pF ($$n = 7$$) over the 30 min recording time of the assay (Supplementary Figure S1), which equated to a reduction of $26\%$ ± $3\%$ of the peak current. There was no change in the voltage-dependence of the current over the 30 min recording time, with a similar $23\%$ ± $4\%$ reduction in current observed at −90 mV (−15 ± 1 pA/pF to −11 ± 1 pA/pF, $$n = 7$$). FIGURE 1: Effects of Ani9, benzbromarone and niclosamide on hTMEM16Aabc currents. Whole-cell patch clamp recordings of the effects of Ani9 (A), benzbromarone (B) and niclosamide (C) on hTMEM16Aabc currents measured in FRT cells with [Ca2+]i clamped at 415 nM. Sample current-time curves (i) for the effects of the indicated compound concentration on peak current at +90mV; inset shows raw current traces for the time points indicated by the coloured dots. Mean current-voltage relationship for each concentration tested (ii) and concentration-response curves (iii) are also shown. Symbols in (ii) and (iii) represent the mean data ±SEM. (n = 4–10 experiments in each group). IC50 = concentration of test compound required to inhibit 50% of the response. Ani9 induced a concentration-dependent inhibition TMEM16Aabc currents with an IC50 value of 68 ± 11 nM, as assessed at +90 mV ($$n = 9$$). The current was almost completely inhibited by Ani9 at 1 µM ($93\%$ ± $1\%$ inhibition, $$n = 9$$; Figure 1A). There was minor voltage dependence to the blockade with the IC50 at −90 mV being reduced slightly to 37 ± 3 nM ($$n = 9$$, $p \leq 0.05$). Like Ani9, benzbromarone induced a concentration-dependent inhibition of TMEM16Aabc currents (Figure 1B), with an IC50 value of 3.0 ± 0.6 µM ($$n = 10$$), as assessed at +90 mV. As with Ani9, benzbromarone inhibition was slightly more potent when the assessed at −90 mV (IC50 = 1.3 ± 0.3 µM, $$n = 10$$, $p \leq 0.05$). Benzbromarone inhibited the TMEM16Aabc current by $96\%$ ± $1\%$ ($$n = 10$$) at the highest concentration tested (30 µM). In contrast to the clear inhibitory effects of Ani9 and benzbromarone on TMEM16Aabc currents, niclosamide produced a “bell-shaped” response, with potentiation of currents at concentrations up to 1.1 µM (Figure 1C; $$n = 4$$). At higher concentrations of niclosamide, the increase in current magnitude was reduced. The effects of niclosamide were more pronounced on the inward than the outward currents such that the rectification ratio (ratio of peak outward current at +90 mV to the peak inward current at −90 mV) was reduced from 10.0 ± 1.0 in the presence of vehicle alone, to 2.3 ± 0.3 ($$n = 4$$, $p \leq 0.05$) by 1.1 µM niclosamide. The change in rectification was also accompanied by a leftward shift in the current reversal potential from −12.3 ± 2.0 mV with vehicle, to −29.8 ± 1.3 mV ($$n = 4$$, $p \leq 0.05$) in the presence of 30 µM niclosamide. To provide additional insight into the unexpected profile, the effects of niclosamide were evaluated on an additional TMEM16A isoform, TMEM16Aacd. The effects of niclosamide on TMEM16Aacd, stably expressed in HEK cells, with [Ca2+]i clamped at 415 nM, were identical to those observed in the FRT-TMEM16Aabc cells, i.e., a “bell shaped” concentration response, with enhancement of currents at low µM accompanied by loss of rectification and a left shift in the current reversal potential (see Supplementary Figure S2A). When [Ca2+]i in the cells was clamped to 0 nM to remove activation of the TMEM16Aacd channels, the cells exhibited small (<5 pA/pF) currents that lacked the biophysical characteristics of TMEM16A, i.e., no outward rectification and time-dependent activation/inactivation (Supplementary Figure S2B). However, under these conditions niclosamide induced a clear concentration-dependent increase in current with an EC50 value of 2.7 ± 0.3 µM ($$n = 6$$). The current retained linear rectification characteristics, lacked time dependent activation or inactivation kinetics and reversed at a more hyperpolarised reversal potential than the baseline current of approximately −50 mV. Quantitatively similar effects were induced by niclosamide on the parental HEK cell line, which lacked expression of TMEM16A, suggesting that niclosamide can induce activation of a non-TMEM16A conductance in these cells (Supplementary Figure S3). ## Short-circuit current (ISC) measurements FRT-hTMEM16Aabc cells were mounted in Ussing chambers in asymmetrical Ringers solutions. The basolateral Ringers contained (in mM): 120 NaCl, 25 NaHCO3, 3.3 KH2PO4, 0.8 K2HPO4, 1.2 CaCl2, 1.2 MgCl2, and 10 glucose. In the apical Ringers solution, NaCl was reduced to 80 mM and 40 mM sodium gluconate was included to establish a Cl− gradient. The solution osmolarity was between 280 and 300 mosmol kg H2O−1 and was continuously gassed ($5\%$ CO2 in O2; pH 7.4) and maintained at 37°C. Cells were voltage clamped to 0 mV (model EVC4000; WPI) and the short-circuit current (ISC) was measured. Data were recorded using a PowerLab workstation and LabChart software (ADInstruments, Abingdon, Oxon, United Kingdom). TMEM16A blockers were added to the apical side of the epithelium for 5 min before the addition of UTP, that was used to stimulate a TMEM16A-mediated chloride secretory responses. HBE cultured at air-liquid interface (ALI) for 14–21 days were mounted in Ussing chambers in symmetrical Ringers solution containing (in mM): 120 NaCl, 25 NaHCO3, 3.3 KH2PO4, 0.8 K2HPO4, 1.2 CaCl2, 1.2 MgCl2, and 10 glucose, and voltage clamped as described above. In some studies, HBE were pre-treated with IL-13 (10 ng/mL; Peprotech, United Kingdom) for 48 h to increase TMEM16A expression levels (Caputo et al., 2008). Amiloride (10 μM; apical) was used to block the spontaneous, ENaC-mediated current, whilst UTP (10 μM; apical) and forskolin (10 μM; apical and basolateral) were added to the bath solutions to activate Ca2+ and cAMP-dependent anion secretory responses, respectively. Inh172 (30 μM; apical and basolateral) was also used to inhibit CFTR function. The TMEM16A blockers were added to the apical side of the Ussing chambers to evaluate effects on each of these currents. ## Intracellular Ca2+ measurements CF-HBE cultured at air-liquid interface for 14–21 days were pre-treated with IL-13 (10 ng/mL; 48 h). CF-HBE were then loaded with the Ca2+-sensitive fluorescent reporter dye, Calcium6 for 120 min at 37°C in Hanks balanced salt solution (HBSS) buffered with 20 mM HEPES (pH 7.4). Calcium6 was used as its Kd for calcium (320 nM) covers the reported physiological range of [Ca2+]i mobilisation by purinoceptors in these cells (0.1–1 μM, Paradiso et al., 2001; Ribeiro et al., 2005). The direct effects of apically administered TMEM16A blockers on [Ca2+]i as well as effects on the subsequent responses to UTP were measured using a PHERAstar plate reader (BMG Lab Tech, United Kingdom). Equivalent measurements were performed in FRT-TMEM16Aabc cells that had been cultured in 96 well, clear bottom plastic plates and loaded with Calcium6 using the same protocol. ## Goblet cell formation HBE cultured at air-liquid interface for 14–21 days were treated with TMEM16A blockers ± IL-13 (10 ng/mL; basolateral) for 96 h to establish any effects on goblet cell numbers. At 48h, LDH levels in the media were measured according to the manufacturer’s instructions (Roche Cytotoxicity Detection Kit #11644793001), and blockers, IL-13 and media were refreshed. At 96 h, wells were fixed in $4\%$ formaldehyde and were stained with antibodies to MUC5AC (45M1; Thermo Fisher, United Kingdom) and acetylated α-tubulin (6-11B-1; Sigma, United Kingdom) as previously described (Danahay et al., 2015; Danahay et al., 2020b) with secondary antibodies to enable fluorescence detection of both proteins (Alexa Fluor; Thermo Fisher, United Kingdom). The MUC5AC+ stained area was visualised using a Zeiss Axiovert epifluorescence microscope with a motorised stage that was used to image the same 9 regions of interest on each insert. ImageJ was used to quantify the MUC5AC+ and stained area per insert which was normalised to the vehicle control group. The process was repeated for the acetylated α-tubulin+ stained area. ## GPCR profiling Profiling of compound activity versus a select panel of diverse GPCRs was undertaken by ThermoFisher Scientific (United Kingdom) via their SelectScreen cell-based GPCR assay panel. Detailed experimental protocols on this assay platform can be found at (https://www.thermofisher.com/content/dam/LifeTech/migration/en/filelibrary/services/discovery-research/pdfs.par.36410.file.dat/sscg-brochure.pdf). Briefly these assays detect agonist or antagonist activity using cell lines with the GPCR of interest over-expressed, driving the expression of a reporter gene encompassing a mammalian-optimized Beta-lactamase. Test compounds are incubated with the specific cell lines at the defined final concentration for 5 or 16 h (depending on the cell line specifics) before quantification of effect. Statistical tests: For comparisons between multiple test compounds and a single vehicle control group, a one-way ANOVA with post hoc Dunnett’s test was used with significance assumed when $p \leq 0.05.$ ## Ion transport and [Ca2+]i measurements Under an imposed basolateral to apical chloride gradient, the addition of UTP (10 µM) to the apical side of FRT-TMEM16Aabc monolayers induced a transient increase in ISC (Figure 2) of 158.4 ± 13.0 µAcm-2 ($$n = 11$$). Treatment with either Ani9, benzbromarone or niclosamide for 5 min attenuated the subsequent UTP-stimulated peak increase in ISC in addition to the integrated ISC response (AUC) in a concentration-dependent manner ($p \leq 0.0001$; $$n = 8$$–11). Of note, Ani9 and benzbromarone were without effect on the baseline ISC prior to the addition of UTP. In contrast, niclosamide (10 µM) induced a bi-phasic response immediately on addition to the cells. Initially niclosamide stimulated a transient increase in ISC (4.4 ± 0.3 µAcm-2 compared to 0.0 ± 0.1 µAcm-2 in the vehicle control; $p \leq 10$–5) and a subsequent attenuation of the baseline current (−6.7 ± 0.7 µAcm-2 compared to −0.3 ± 0.1 µAcm-2 in the vehicle control; $p \leq 10$–4). In separate experiments, the UTP-stimulated increase in [Ca2+]i in FRT-TMEM16Aabc cells was unaffected by pre-treatment with Ani9 (Figure 3). In contrast, both benzbromarone and niclosamide significantly attenuated the maximal UTP-stimulated elevation of [Ca2+]i. FIGURE 2: Ani9, niclosamide and benzbromarone attenuate UTP-stimulated anion secretion in FRT-TMEM16Aabc monolayers. Sample short-circuit current traces illustrating the effects of Ani9, niclosamide (Nic) and benzbromarone (Benz) on the apical UTP-stimulated response of FRT-TMEM16Aabc monolayers (A). The peak increase in ISC (B) as well as the integrated AUC (C) were measured. Mean data ±SEM (n = 8–11 inserts per group) are shown. Concentrations of test compounds are shown in brackets as µM. and ** denote p < 0.0002 and p < 0.0001 respectively following a one-way ANOVA using post hoc Dunnett’s test.* FIGURE 3: Benzbromarone and niclosamide affect UTP-stimulated calcium mobilisation in FRT-hTMEM16Aabc cells. Mean data ±SEM illustrating the effects of a 10 min incubation of FRT-hTMEM16Aabc cells with test compounds on a subsequent UTP-stimulated increase in [Ca2+]i. Cells had been loaded with Calcium6 dye and were stimulated with UTP (0.03–69 µM) and the changes in fluorescence recorded. Peak changes in fluorescence minus the baseline levels are plotted for each compound. ** denotes p < 0.001 (n = 6 experiments per groups).* Similar results were obtained with the three TMEM16A blockers in IL-13 pre-treated CF-HBE (Figure 4). After the ENaC-mediated ISC had been blocked with amiloride, Ani9 and benzbromarone treatment for 5 min attenuated the subsequent UTP-stimulated increase in ISC, quantified as both the peak increase in ISC and AUC that was concentration-dependent ($p \leq 0.0001$; $$n = 3$$–5). Neither Ani9 nor benzbromarone had any direct effect on the amiloride-insensitive baseline ISC when added to the cells. In contrast, niclosamide (10 µM) induced an immediate, transient increase in ISC (6.8 ± 1.0 µAcm-2 compared to 0.5 ± 0.2 µAcm-2 in the vehicle control; $p \leq 0.001$) and returned to a lower baseline ISC than before compound addition (−0.9 ± 0.2 µAcm-2 compared to −0.3 ± 0.1 µAcm-2 in the vehicle control; $$p \leq 0.02$$). The subsequent UTP-stimulated increase in ISC was also attenuated by niclosamide. In separate experiments, but again using IL-13 pre-treated CF-HBE cultured at ALI, neither baseline [Ca2+]i or the UTP-stimulated increase in [Ca2+]i was affected by pre-treatment with Ani9 (Figure 5). In contrast, both benzbromarone and niclosamide induced an immediate and transient increase in baseline [Ca2+]i when added to the epithelium and attenuated the subsequent UTP-stimulated elevation of [Ca2+]i. FIGURE 4: Ani9, niclosamide and benzbromarone attenuate UTP-stimulated anion secretion in CF-HBE cells. Sample short-circuit current traces illustrating the effects of Ani9, niclosamide (Nic) and benzbromarone (Benz) on the apical UTP-stimulated response of IL-13 pre-treated CF-HBE cells (A). The peak increase in ISC (B) as well as the integrated AUC (C) were measured. Mean data ±SEM (n = 3-5 inserts per group) are shown. Concentrations of test compounds are shown in brackets as µM. * denotes p < 0.0001 respectively following a one-way ANOVA using post hoc Dunnett’s test.* FIGURE 5: Benzbromarone and niclosamide affect calcium handling in differentiated cystic fibrosis human bronchial epithelial (HBE) cells. Mean data ±SEM illustrating the direct effects of Ani9, benzbromarone and niclosamide on [Ca2+]i (A) and also on the increases in [Ca2+]i stimulated by a maximal concentration of UTP (B) in CF-HBE. Peak changes in fluorescence minus the baseline levels are plotted for the effects of each compound under both conditions. Mean data ±s. e.m. (4–10 inserts per group). and ** denote p < 0.001 and p < 0.0001 respectively.* To understand the specificity of these TMEM16A blockers in HBE, Ani9, benzbromarone and niclosamide were added to the apical side of the epithelium at the start of the protocol and before any other manoeuvres. Non-CF HBE were used for these studies to enable an evaluation of both the ENaC and CFTR mediated ion transport processes. Ani9 (10 µM) had no effect on: 1) the baseline ISC or transepithelial resistance, 2) on the magnitude of the subsequent response to amiloride or 3) on the forskolin stimulated, Inh172-sensitive ISC, when compared with the vehicle (Figure 6A). There was a small but significant inhibitory effect of Ani9 on the peak increase in forskolin-stimulated current. In contrast, niclosamide (10 µM) induced an approximate $90\%$ reduction in the baseline ISC within 30 min of addition to the cells (Figure 6B), that was associated with a significant attenuation of the transepithelial resistance (−174 ± 19 Ω cm2 versus +108 ± 14 Ω cm2 in the vehicle control; $p \leq 10$–6, $$n = 6$$). The subsequent amiloride-sensitive, forskolin-stimulated and Inh172-sensitive currents were all also attenuated after niclosamide treatment. Benzbromarone induced a similar, $50\%$ decline in the baseline ISC (Figure 6C) but tended to increase the transepithelial resistance (+170 ± 25 Ω cm2 versus +108 ± 14 Ω cm2 in the vehicle control; $$p \leq 0.053$$, $$n = 6$$). Benzbromarone treatment also attenuated the subsequent amiloride-sensitive, forskolin-stimulated and Inh172-sensitive currents. FIGURE 6: Multiple effects of niclosamide and benzbromarone on epithelial ion transport mechanisms in HBE cells. Sample short-circuit current traces (A) illustrating the effects of Ani9, niclosamide (Nic) and benzbromarone (Benz) on the ion transport properties of HBE cells (non-CF). The mean ± SEM changes in baseline (B), amiloride-sensitive (C), forskolin-stimulated (D) and Inh172-sensitive (E) ISC responses are shown (n = 6 inserts per group). Concentrations of test compounds are shown in brackets as µM. and ** denote p < 0.01 and p < 0.0001 respectively following a one-way ANOVA using post hoc Dunnett’s test.* ## Effect of TMEM16A inhibitors on goblet cell formation TMEM16A inhibitors have been widely reported to influence the differentiation of airway epithelial cells and to specifically attenuate the formation of mucin producing goblet cells. Using HBE in an assay format to quantify goblet cell formation, IL-13 treatment for 96 h induced a significant, 7.0 ± 1.1 fold increase in the number of goblet cells (Figure 7A), measured as the increase in MUC5AC+ stained area ($p \leq 0.001$, $$n = 5$$). Concurrent treatment with Ani9 (10 μM; 96 h) did not affect the MUC5AC+ stained area in either the IL-13 naïve or IL-13 treated cells. There was likewise no effect of Ani9 on the acetylated α-tubulin (ciliated) stained area (Figure 7B). In contrast, treatment of HBE with either niclosamide or benzbromarone significantly attenuated the MUC5AC+ stained area of naïve and IL-13 exposed cultures. These treatments also significantly attenuated the acetylated α-tubulin-stained area. It was observed in these experiments that by 48 h after initiating treatment of HBE, the air-liquid interface was compromised in the niclosamide treated group, with media observed on to the mucosal surface of the cells. Analysis of basolateral media collected at this 48 h timepoint revealed detectable levels of LDH (0.098 ± 0.003 U/mL) in the niclosamide treated group compared to undetectable levels in the other treatment groups. FIGURE 7: Effects of TMEM16A blockers on IL-13 stimulated goblet cell formation in HBE. The effects of Ani9, niclosamide (Nic) and benzbromarone (Benz) on normal (black bars) or IL-13 treated (red bars) HBE cultures. Mean ± SEM stained areas of MUC5AC + goblet cells (A) and acetylated α-tubulin + multiciliated cells (B) are shown (n = 5 inserts per group). Concentrations of test compounds are shown in brackets as µM. , and * denote p < 0.02, p < 0.006 and p < 0.0002 respectively following a one-way ANOVA using post hoc Dunnett’s test. Sample immunofluorescence images are shown illustrating the respective staining for MUC5AC+ goblet cells (C) and acetylated α-tubulin+ multiciliated cells (D).* ## GPCR selectivity profiling When profiled against a small panel of 22 functional GPCR assays (Supplementary Figure S4), Ani9 (10 µM) inhibited 2 GPCR targets by ≥ $50\%$ (5-HT2B, delta opioid). In contrast benzbromarone and niclosamide (10 µM), inhibited 6 and 11 GPCRs respectively by ≥ $50\%$. ## Discussion Inhibiting TMEM16A function has been proposed as a novel approach to treat several respiratory diseases including asthma, cystic fibrosis, COVID-19 infection and pulmonary arterial hypertension (PAH) (Cabrita et al., 2019; Kunzelmann et al., 2019; Miner et al., 2019; Papp et al., 2019; Singh et al., 2022). Benzbromarone and niclosamide have been used clinically as a uricosuric agent and anti-helmintic respectively and have been more recently described as blockers of TMEM16A (Huang et al., 2012; Miner et al., 2019). To this end, both niclosamide and benzbromarone have entered clinical trials as repurposed therapies for the treatment of some of these respiratory conditions (ClinicalTrials.gov NCT04644705; Papp et al., 2019) and in view of this it is surprising that data supporting their selectivity and even mechanism of action on TMEM16A are limited. The most illuminating data in the present study, is the observation that niclosamide does not block TMEM16A but rather attenuates channel activity through an indirect inhibitory effect on intra-cellular Ca2+ signalling. Benzbromarone does block TMEM16A activity, but together with niclosamide, shares a number of TMEM16A-independent activities that complicate the interpretation of preclinical and now clinical studies. In patch-clamp experiments using conditions where [Ca2+]i was tightly clamped and therefore changes in current were independent of changes in [Ca2+]i (Bill et al., 2015; Danahay et al., 2020a), benzbromarone and Ani9 both inhibited TMEM16Aabc activity. Effects of these inhibitors were similar on both inward and outward currents and displayed profiles consistent with a bona fide block of channel activity. In contrast, niclosamide failed to inhibit the activity of both TMEM16Aabc and TMEM16Aacd isoforms, paradoxically potentiating the current at concentrations ≤1 μM and induced a significant leftward shift in the reversal potential. In view of the niclosamide-induced activation of a current with [Ca2+]i clamped to 0 nM in both HEK-TMEM16Aacd cells as well as in the parental HEK cell, it is likely that niclosamide activates an additional, non-TMEM16A conductance(s) in both the HEK and FRT cell lines. Niclosamide has however previously been reported to block TMEM16A in patch-clamp experiments (Centeio et al., 2020). In the present studies niclosamide attenuated UTP-stimulated increases in short circuit current in monolayers of both FRT-TMEM16Aabc and CF-HBE. In each of these assay formats where niclosamide has demonstrated an inhibition of TMEM16A activity [Ca2+]i has been unbuffered and is thus under normal physiological control. We hypothesised that niclosamide may be affecting the physiological regulation [Ca2+]i signalling leading to the artefactual observation and description of niclosamide as a TMEM16A blocker. Consistent with this hypothesis, niclosamide attenuated the UTP-stimulated increase in [Ca2+]i in both FRT-TMEM16Aabc and CF-HBE. In contrast, Ani9 did not affect the UTP-stimulated rise in [Ca2+]i. These data support the concept that niclosamide does not block TMEM16A directly, but does attenuate [Ca2+]i in response to purinergic stimulation. Ani9 blocks TMEM16A but importantly, has no influence on [Ca2+]i. Benzbromarone shared characteristics of both niclosamide and Ani9 in that it both blocked TMEM16A activity under conditions of [Ca2+]i buffering but also attenuated stimulated [Ca2+]i responses. Similar data confirming the effect of niclosamide on [Ca2+]i and the lack of effect of Ani9 have been recently reported (Genovese et al., 2022) and suggest that niclosamide may be inhibiting the SERCA pump. An inhibitory effect of niclosamide on the SERCA pump would also be consistent with the transient increase in ISC that was observed in both FRT-TMEM16A and HBE ion transport experiments (Figures 2, 4). To further evaluate the selectivity profiles of niclosamide, benzbromarone and Ani9, compounds were added to non-CF HBE under voltage clamp conditions to examine potential effects on ENaC and CFTR-dependent ion transport processes. Ani9 was without effect on either the baseline or amiloride-sensitive currents. The small but significant effect of Ani9 on the forskolin stimulated ISC likely reflects a component of this current being TMEM16A-mediated rather than an effect on CFTR directly as this compound has previously been shown to not affect CFTR function (Seo et al., 2016) and is further supported by the lack of effect on the Inh172-sensitive current which is CFTR-mediated (Figure 6E). In contrast, both niclosamide and benzbromarone significantly attenuated the baseline, ENaC-mediated and CFTR currents. In addition, niclosamide significantly reduced the transepithelial resistance whilst benzbromarone induced a trend towards a tightening of the epithelium ($$p \leq 0.053$$). It would appear unlikely that both niclosamide and benzbromarone are direct blockers of ENaC and CFTR. It is more likely that both compounds have a generalised effect on active ion transport processes secondary to the previously described effects on [Ca2+]i homeostasis or perhaps due to activity of these agents on mitochondrial function (Felser et al., 2014; Tao et al., 2014). The relative selectivity of the three compounds was also assessed using a small panel of 22 GPCR assays. Ani9 inhibited 2 GPCR targets when tested at a concentration >130-fold over it is IC50 for TMEM16A. In contrast, benzbromarone inhibited six GPCR targets at 4x it is IC50 for TMEM16A whilst niclosamide inhibited 11 GPCRs, that would be indicative of a greater potential for off-target effects in biological systems compared with Ani9. Finally, we evaluated the effects of these compounds on cellular differentiation in HBE following IL-13 treatment. IL-13 both increases the expression of functional TMEM16A and promotes the formation of MUC5AC+ goblet cells (Caputo et al., 2008; Danahay et al., 2015) and previous studies have proposed a mechanism whereby increased TMEM16A channel activity drives goblet cell formation (Lin et al., 2015; Qin et al., 2016; Kondo et al., 2017; Benedetto et al., 2019). This proposed activity of TMEM16A has been used as one of the salient arguments to progress TMEM16A blockers as novel therapeutics to treat mucus hypersecretion in asthma and CF (Benedetto et al., 2019; Cabrita et al., 2019). As previously reported, Ani9 was without effect on the numbers of MUC5AC+ goblet cells or acetylated α-tubulin+ ciliated cells under both naïve and IL-13 stimulated conditions (Danahay et al., 2020b). In contrast, niclosamide and benzbromarone treatment attenuated numbers of both goblet and ciliated cells irrespective of IL-13 treatment. These data do not however support a role for TMEM16A channel function as a regulator of cellular differentiation but rather suggest that previous reports to the contrary that used either benzbromarone or niclosamide (Kondo et al., 2017; Benedetto et al., 2019) are likely due to a non-selective, potentially toxic effect of these agents. The broad effect of both niclosamide and benzbromarone on active ion transport processes in these cells, the increase in LDH release (niclosamide only) and loss of epithelial integrity, together with published effects on mitochondrial function further support this proposal. In view of these data, the published studies supporting the repurposing of niclosamide and benzbromarone as TMEM16A blocker therapies for respiratory diseases should be carefully reviewed. In addition to the data reported in the present study, niclosamide has been demonstrated to relax freshly isolated human bronchial smooth muscle and on this basis proposed TMEM16A blockade as a novel bronchodilator mechanism (Miner et al., 2019). This effect of niclosamide has been confirmed but also demonstrated to be an effect that is not shared by Ani9 (Danahay et al., 2020b). Furthermore, ETX001, a recently described TMEM16A potentiator shows no effect on freshly isolated human bronchial smooth muscle tone or on lung function in vivo. Together, these data suggest that niclosamide is likely relaxing airway smooth muscle through an effect on [Ca2+]i that is independent of TMEM16A. Benzbromarone has similarly been demonstrated to relax vascular smooth muscle in vitro through a proposed TMEM16A blocking mechanism (Papp et al., 2019). It is however noteworthy that the structurally diverse TMEM16A inhibitors T16Ainh-A01 and Ani9 (Papp et al., 2019; Danahay et al., 2020b) failed to show activity in human vascular smooth muscle preparations, that would support an off-target effect of benzbromarone as having driven the vasodilator responses. Of note, when subsequently evaluated in a small clinical study in PAH patients, benzbromarone induced a paradoxical increase in pulmonary artery pressure (Papp et al., 2019). In summary, these data do not support the continued use or description of niclosamide as a blocker of TMEM16A and historical reports to this end should be interpreted with caution. Benzbromarone can be considered as a TMEM16A blocker, although the potential for off-target effects on [Ca2+]i should be carefully considered during experimental design and in the interpretation of historical data. Based on available data, Ani9 is presently the most potent and selective TMEM16A inhibitor, displaying selectivity over related family members. ## Data availability statement The raw data supporting the conclusion of this article will be made available by the authors, without undue reservation. ## Author contributions HD and MG conceptualised the work. SL, RF, and MG developed, ran and analysed data QPatch and calcium assays. HC, KA, and HD developed, ran and alaysed ion transport assays. HD developed, ran and alaysed goblet cell assays. All authors contributed to the writing and editing of the manuscript. ## Conflict of interest Authors HD and MG were employed by the company Enterprise Therapeutics 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. ## 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.2023.1142342/full#supplementary-material ## References 1. Benedetto R., Cabrita I., Schreiber R., Kunzelmann K.. TMEM16A is indispensable for basal mucus secretion in airways and intestine. FASEB J. (2019) 33 4502-4512. DOI: 10.1096/fj.201801333RRR 2. Bill A., Gutierrez A., Kulkarni S., Kemp C., Bonenfant D., Voshol H.. ANO1/TMEM16A interacts with EGFR and correlates with sensitivity to EGFR-targeting therapy in head and neck cancer. Oncotarget (2015) 6 9173-9188. DOI: 10.18632/oncotarget.3277 3. Boedtkjer D. M., Kim S., Jensen A. B., Matchkov V. M., Andersson K. E.. New selective inhibitors of calcium-activated chloride channels - T16A(inh) -A01, CaCC(inh) -A01 and MONNA - what do they inhibit?. Br. J. Pharmacol. (2015) 172 4158-4172. DOI: 10.1111/bph.13201 4. Cabrita I., Benedetto R., Fonseca A., Wanitchakool P., Sirianant L., Skryabin B. V.. Differential effects of anoctamins on intracellular calcium signals. FASEB J. (2017) 31 2123-2134. DOI: 10.1096/fj.201600797RR 5. Cabrita I., Benedetto R., Schreiber R., Kunzelmann K.. Niclosamide repurposed for the treatment of inflammatory airway disease. JCI Insight (2019) 4 e128414. DOI: 10.1172/jci.insight.128414 6. Caputo A., Caci E., Ferrera L., Pedemonte N., Barsanti C., Sondo E.. TMEM16A, a membrane protein associated with calcium-dependent chloride channel activity. Science (2008) 322 590-594. DOI: 10.1126/science.1163518 7. Centeio R., Cabrita I., Benedetto R., Talbi K., Ousingsawat J., Schreiber R.. Pharmacological inhibition and activation of the Ca2+ activated Cl- channel TMEM16A. Int. J. Mol. Sci. (2020) 21 2557. DOI: 10.3390/ijms21072557 8. Danahay H. L., Lilley S., Fox R., Charlton H., Sabater J., Button B.. TMEM16A potentiation: A novel therapeutic approach for the treatment of cystic fibrosis. Am. J. Respir. Crit. Care Med. (2020a) 201 946-954. DOI: 10.1164/rccm.201908-1641OC 9. Danahay H., Fox R., Lilley S., Charlton H., Adley K., Christie L.. Potentiating TMEM16A does not stimulate airway mucus secretion or bronchial and pulmonary arterial smooth muscle contraction. FASEB Bioadv (2020b) 2 464-477. DOI: 10.1096/fba.2020-00035 10. Danahay H., Pessotti A. D., Coote J., Montgomery B. E., Xia D., Wilson A.. Notch2 is required for inflammatory cytokine-driven goblet cell metaplasia in the lung. Cell Rep. (2015) 10 239-252. DOI: 10.1016/j.celrep.2014.12.017 11. De La Fuente R., Namkung W., Mills A., Verkman A. S.. Small-molecule screen identifies inhibitors of a human intestinal calcium-activated chloride channel. Mol. Pharmacol. (2008) 73 758-768. DOI: 10.1124/mol.107.043208 12. Felser A., Lindinger P. W., Schnell D., Kratschmar D. V., Odermatt A., Mies S.. Hepatocellular toxicity of benzbromarone: Effects on mitochondrial function and structure. Toxicology (2014) 324 136-146. DOI: 10.1016/j.tox.2014.08.002 13. Genovese M., Buccirossi M., Guidone D., De Cegli R., Sarnataro S., di Bernardo D.. Analysis of inhibitors of the anoctamin-1 chloride channel (transmembrane member 16A, TMEM16A) reveals indirect mechanisms involving alterations in calcium signalling. Br. J. Pharmacol. (2022) 180 775-785. DOI: 10.1111/bph.15995 14. Huang F., Zhang H., Wu M., Yang H., Kudo M., Peters C. J.. Calcium-activated chloride channel TMEM16A modulates mucin secretion and airway smooth muscle contraction. Proc. Natl. Acad. Sci. U. S. A. (2012) 109 16354-16359. DOI: 10.1073/pnas.1214596109 15. Kondo M., Tsuji M., Hara K., Arimura K., Yagi O., Tagaya E.. Chloride ion transport and overexpression of TMEM16A in a Guinea-pig asthma model. Clin. Exp. Allergy (2017) 47 795-804. DOI: 10.1111/cea.12887 16. Kunzelmann K., Ousingsawat J., Cabrita I., Doušová T., Bähr A., Janda M.. TMEM16A in cystic fibrosis: Activating or inhibiting?. Front. Pharmacol. (2019) 10. DOI: 10.3389/fphar.2019.00003 17. Lin J., Jiang Y., Li L., Liu Y., Tang H., Jiang D.. TMEM16A mediates the hypersecretion of mucus induced by Interleukin-13. Exp. Cell Res. (2015) 334 260-269. DOI: 10.1016/j.yexcr.2015.02.026 18. Miner K., Labitzke K., Liu B., Wang P., Henckels K., Gaida K.. Drug repurposing: The anthelmintics niclosamide and nitazoxanide are potent TMEM16A antagonists that fully bronchodilate airways. Front. Pharmacol. (2019) 10 51. DOI: 10.3389/fphar.2019.00051 19. Namkung W., Phuan P. W., Verkman A. S.. TMEM16A inhibitors reveal TMEM16A as a minor component of calcium-activated chloride channel conductance in airway and intestinal epithelial cells. J. Biol. Chem. (2011) 286 2365-2374. DOI: 10.1074/jbc.M110.175109 20. Papp R., Nagaraj C., Zabini D., Nagy B. M., Lengyel M., Skofic Maurer D.. Targeting TMEM16A to reverse vasoconstriction and remodelling in idiopathic pulmonary arterial hypertension. Eur. Respir. J. (2019) 53 1800965. DOI: 10.1183/13993003.00965-2018 21. Paradiso A. M., Ribeiro C. M., Boucher R. C.. Polarized signaling via purinoceptors in normal and cystic fibrosis airway epithelia. J. Gen. Physiol. (2001) 117 53-67. DOI: 10.1085/jgp.117.1.53 22. Pedemonte N., Galietta L. J.. Structure and function of TMEM16 proteins (anoctamins). Physiol. Rev. (2014) 94 419-459. DOI: 10.1152/physrev.00039.2011 23. Qin Y., Jiang Y., Sheikh A. S., Shen S., Liu J., Jiang D.. Interleukin-13 stimulates MUC5AC expression via a STAT6-TMEM16A-ERK1/2 pathway in human airway epithelial cells. Int. Immunopharmacol. (2016) 40 106-114. DOI: 10.1016/j.intimp.2016.08.033 24. Ribeiro C. M., Paradiso A. M., Carew M. A., Shears S. B., Boucher R. C.. Cystic fibrosis airway epithelial Ca2+ i signaling: The mechanism for the larger agonist-mediated Ca2+ i signals in human cystic fibrosis airway epithelia. J. Biol. Chem. (2005) 280 10202-10209. DOI: 10.1074/jbc.M410617200 25. Schroeder B. C., Cheng T., Jan Y. N., Jan L. Y.. Expression cloning of TMEM16A as a calcium-activated chloride channel subunit. Cell (2008) 134 1019-1029. DOI: 10.1016/j.cell.2008.09.003 26. Seo Y., Lee H. K., Park J., Jeon D. K., Jo S., Jo M.. Ani9, A novel potent small-molecule ANO1 inhibitor with negligible effect on ANO2. PLoS One (2016) 11 e0155771. DOI: 10.1371/journal.pone.0155771 27. Singh S., Weiss A., Goodman J., Fisk M., Kulkarni S., Lu I.. Niclosamide-A promising treatment for COVID-19. Br. J. Pharmacol. (2022) 179 3250-3267. DOI: 10.1111/bph.15843 28. Tao H., Zhang Y., Zeng X., Shulman G. I., Jin S.. Niclosamide ethanolamine-induced mild mitochondrial uncoupling improves diabetic symptoms in mice. Nat. Med. (2014) 20 1263-1269. DOI: 10.1038/nm.3699 29. Yang Y. D., Cho H., Koo J., Tak M. H., Cho Y., Shim W.. TMEM16A confers receptor-activated calcium-dependent chloride conductance. Nature (2008) 455 1210-1215. DOI: 10.1038/nature07313
--- title: Development of the thin film solid phase microextraction (TF-SPME) method for metabolomics profiling of steroidal hormones from urine samples using LC-QTOF/MS authors: - Wiktoria Struck-Lewicka - Beata Karpińska - Wojciech Rodzaj - Antoni Nasal - Bartosz Wielgomas - Michał Jan Markuszewski - Danuta Siluk journal: Frontiers in Molecular Biosciences year: 2023 pmcid: PMC10025495 doi: 10.3389/fmolb.2023.1074263 license: CC BY 4.0 --- # Development of the thin film solid phase microextraction (TF-SPME) method for metabolomics profiling of steroidal hormones from urine samples using LC-QTOF/MS ## Abstract In the present study, the development and optimization of a thin film solid phase microextraction method (TF-SPME) was conducted for metabolomics profiling of eight steroid compounds (androsterone, dihydrotestosterone, dihydroepiandrosterone, estradiol, hydroxyprogesterone, pregnenolone, progesterone and testosterone) from urine samples. For optimization of extraction method, two extraction sorbents (PAN-C18 and PS-DVB) were used as they are known to be effective for isolation of low-polarity analytes. The stages of sample extraction and analyte desorption were considered as the most crucial steps in the process. Regarding the selection of the most suitable desorption solution, six different mixtures were analyzed. As a result, the mixture of ACN: MeOH (1:1, v/v) was chosen in terms of the highest analytes’ abundances that were achieved using the chosen solvent. Besides other factors were examined such as the volume of desorption solvent and the time of both extraction and desorption processes. The analytical determination was carried out using the ultra-high performance liquid chromatography coupled with high resolution tandem mass spectrometry detection in electrospray ionization and positive polarity in a scan mode (UHPLC-ESI-QTOF/MS). The developed and optimized TF-SPME method was validated in terms of such parameters as extraction efficiency, recovery as well as matrix effect. As a result, the extraction efficiency and recovery were in a range from $79.3\%$ to $99.2\%$ and from $88.8\%$ to $111.8\%$, respectively. Matrix effect, calculated as coefficient of variation was less than $15\%$ and was in a range from $1.4\%$ to $11.1\%$. The values of both validation parameters (recovery and matrix effect) were acceptable in terms of EMA criteria. The proposed TF-SPME method was used successfully for isolation of steroids hormones from pooled urine samples before and after enzymatic hydrolysis of analytes. ## 1 Introduction Steroid hormones are a group of hormones possessing in their chemical structure a sterane skeleton (Pawłowski, 2020). They could be divided in two classes: corticosteroids and sex steroids (Zeelen, 1997). Within those two classes are five types according to the receptors engaged in their pharmacological action: glucocorticoids, mineralocorticoids, estrogens, progestins and androgens (Brunton et al., 2011). The natural steroid hormones are generally synthesized from cholesterol in adrenal cortex (gluco- and mineralocorticoids) and in the gonads (estrogens, progestins and androgens) (Miller, 1988). They bind to specific serum carrier proteins (e.g., globulins) and are transported through the bloodstream to various target organs. Prior to their pharmacological effects presentation, steroids are liberated from carrier proteins, and, due to their high lipophilicity they easily pass the cell membrane, and are translocated to the nucleus where they bind to nuclear receptors. In the nucleus, the steroid-receptor ligand complex binds to specific DNA sequences and induces transcription of its target genes (Gupta and Lalchhandama, 2002; Rousseau, 2012). Steroid hormones carry out regulation of such physiological processes as e.g., carbohydrate, protein and fat metabolism (glucocorticoids) (Mantha et al., 1999; Kuo et al., 2015), and water-mineral balance (mineralocorticoids) (Rogerson and Fuller, 2000). Steroid hormones are also responsible for sexual differentiation and reproduction (sex steroids, gonadocorticoids). Estrogens, including estradiol, take part in the development of the primary and secondary female sex characteristics, while progestins (e.g., progesterone) in maintaining pregnancy. Androgens (e.g., testosterone) control the development and maintenance of reproductive function and are responsible for the secondary sex characteristics in the male (Brunton et al., 2011). Solid phase microextraction method is a modern analytical sample pretreatment approach that allows for efficient isolation, matrix purification and analytes’ concentration in one process. SPME can be easily automated and requires low amounts of chemical solvents. It can be used for isolation of compounds of various physico-chemical properties and from variety of matrices (biological, environmental samples and others) (Pawliszyn, 2009). Therefore, it can also be adopted in various metabolomics studies (Vuckovic and Pawliszyn, 2011; Bojko et al., 2014; Jaroch et al., 2019; Mousavi et al., 2019). SPME process is composed of only a few steps that allows for matrix purification and analytes concentration which decreases the risk of possible contaminations, analytical errors or loss of analytes. Depending of the type of sorbent used, the SPME can be more or less selective in terms of the range of metabolome coverage from extracted samples (Olcer et al., 2019). There are various types of SPME techniques among which the thin-film one (TF-SMPE) is of interest. Compared to the traditional microextraction techniques, the most important advantage of TF-SPME is that a larger volume of extraction phase (the so-called thin-film geometry) which is reflected as larger surface area, leads to higher extraction efficiency and enhanced sensitivity of extraction process. This approach provides more effective agitation as well as increased extraction recovery with shortened analysis time. When combined with automatic 96-blade system, TF-SPME provides high-throughput sample preparation involving preconditioning, sample extraction, washing, and sample desorption as automation process (Mirnaghi et al., 2011). Following Equation 1 (Eq. 1), the mass balance under equilibrium conditions, which is fundamental to SPME, presents the correlation between the amount of analytes extracted into the extractive phase (n) and its original concentration in the extracted sample (C0). According to Eq. 1, increase in the volume of extractive phase will lead to the enhancement of extraction sensitivity (Olcer et al., 2019). n=K∗V1∗V2K∗V2+V1C0 [1] Where, n—number of moles of extracted compound, K- the distribution constant of the analyte for the extractive phase and sample matrix, V 1 - volume of sample, V 2 —volume of extractive phase, C 0 - original concentration of compound in sample. The sensitivity enhancement is crucial in terms of extraction of trace amounts of compounds or when there is a need for extraction of compounds at different concentration levels from complex biological matrices. Therefore, some of the untargeted metabolomics studies have utilized TF-SPME for extraction and effective isolation of endogenous metabolites (Maciążek-Jurczyk et al., 2020; Łuczykowski et al., 2021). The purpose of the so-called metabolomic fingerprinting approach is to detect possibly all metabolites present in analysed sample. Hence, the sample pretreatment procedure is often limited only to dilution, deprotenization (in case of plasma), homogenisation (in terms of tissues) and filtration steps. The efficiency of matrix purification in SPME technique makes it an exceptionally useful analytical tool, however, some metabolites not absorbed to the SPME sorbent will not be ultimately detected. Therefore, this procedure can be used in metabolomics profiling approach which also belongs to the untargeted study but is focused on a selected group of metabolites with common chemical properties like semi-polar, non-polar metabolites or which belong to the same biochemical class such as nucleosides, steroid hormones, amino acids, fatty acids and others. Taking into account metabolomics profiling approach, the most crucial would be the selection of a proper SPME sorbent, dedicated to the metabolites of interests, as well the type of desorption solvent that can effectively desorb the compounds from the extractive phase. In regard with determination of steroid hormones, these analytes have been extracted using different modifications of SPME from various matrices like waste water, fish plasma, urine, milk, saliva or fish and chicken meat (Peñalver et al., 2002; Zhang et al., 2009; Kataoka et al., 2013; Lima Gomes et al., 2013; Olędzka et al., 2017; do Carmo et al., 2019; Mirzajani et al., 2019; Maciążek-Jurczyk et al., 2020; Wang et al., 2020). Although many SPME studies present the extraction of steroid hormones, only few adopting Thin-Film type of SPME have been found. In the work of Maciążek-Jurczyk et al. six steroid hormones (cortisol, testosterone, progesterone, estrone, 17β-estradiol and 17α-ethinylestradiol) were extracted using TF-SPME on C18 fibers from sucker fish plasma (Maciążek-Jurczyk et al., 2020). TF-SPME was also applied in a work of do Carmo et al. ( do Carmo et al., 2019) for extraction of estrogens (estriol, estrone, 17β-estradiol and 17α-ethinylestradiol) from urine samples using specially synthetized biosorbent (bract) produced by the conifer Araucaria angustifolia. The determination was performed with the use of LC-FLD determination technique. Any other TF-SPME approaches have not been carried out for analysis of endogenous steroid hormones from human urine samples. For the first time in the present study, the polystyrene divinylbenzene (PS-DVB) as a thin-film extraction sorbent was chosen for isolation of steroid hormones from human urine samples. Steroid hormones from human urine were also extracted by Olędzka et al. ( Olędzka et al., 2017) but the Authors utilized dispersive liquid-liquid microextraction which appeared to be more efficient than tested conventional type of SPME. Other SPME approaches for extraction of steroid hormones from pig urine (Zhang et al., 2009) river water (Lima Gomes et al., 2013), milk (Wang et al., 2020), white meat (Mirzajani et al., 2019) and saliva (Kataoka et al., 2013) were presented but those studies also utilized conventional SPME technique. Above all, the diversity of above mentioned approaches indicates that SPME is a prominent and modern analytical sample preparation technique which can be employed for effective isolation of endogenous compounds from matrices critical for bioanalytical or clinical studies. In an advent of Thin Film type of SPME, it revealed to be more robust and efficient extraction approach on even larger sample amount in relatively short time than conventional SPME. In the present work, TF-SPME method was developed and optimized for extraction of eight steroid hormones (androsterone, dihydrotestosterone, dihydroepiandrosterone, estradiol, hydroxyprogesterone, pregnenolone, progesterone and testosterone). For optimization, the following types of parameters were tested: type of extraction phase, desorption solvent, time of both extraction and desorption processes as well as volume of the desorption solvent. The developed method was validated in terms of such parameters as the extraction efficiency, recovery as well as matrix effect according to the EMA regulations. The proposed TF-SPME method was applied for isolation of steroidal hormones from urine samples before and after enzymatic hydrolysis process. ## 2.1 Chemicals, reagents and apparatus The eight reference standards as estradiol, progesterone, androsterone, testosterone, dehydroepiandrosterone (DHEA), 17α-hydroxyprogesterone, 4,5α-dihydrotestosterone (DHT) were obtained from Merck, Darmstadt, Germany. Another reference standard pregnenolone was purchased from Avanti Polar Lipids, AL, United States. Deuterium labeled-d3-testosterone [d3-T] (100 μg/mL methanol solution) was purchased from Cerilliant Corporation (Austin, TX, United States). Methanol (MeOH), acetonitrile (ACN) and isopropanol (IPA), all of MS grade were purchased from Thermo Fisher Scientific (Massachusetts, United States). Deionized water was obtained using Milli Ro and Milli Qplus apparatus (Millipore, Vienna, Austria). Formic acid $98\%$–$100\%$ and glacial acetic acid ($99\%$) of LC-MS grade, Surine™ negative urine control used as a blank urine were obtained from Supelco (Merck, Darmstadt, Germany). Sodium acetate trihydrate and β-glucuronidase from Helix pomatia (Type HP-2, aqueous solution, ≥100,000 units/mL), phosphate buffer saline (PBS) were also obtained from Supelco (Merck, Darmstadt, Germany). Reference mass solution and 10 times diluted ESI low calibration tuning mix were purchased from Agilent Technologies, Waldbronn, Germany). The SPME was performed using apparatus Concept 96.2 (PAS Technology, Magdala, Germany) composed of 96 well plates, the arm with mixing table and brush with blades coated in sorbent. The bladed brush was made from steel while the coatings were purchased from PAS Technology (Magdala, Germany). Extraction was performed by using steel blades coated with a polystyrene divinylbenzene (PS-DVB) and polyacrylonitrile C18 (PAN-C18) sorbents. Coating preparation procedures were based on the spraying method described by Mirnaghi et al. ( Mirnaghi et al., 2011). The analyses were performed with the use of ultrahigh performance liquid chromatography UHPLC 1290 Infinity II Series (Agilent Technologies, Waldbronn, Germany) coupled with electrospray ionization (ESI) and high resolution tandem mass spectrometry 6546 QTOF/MS (Agilent Technologies, Waldbronn, Germany). The analyses were performed with the use of Mass Hunter Acquisition software whereas the obtained data were monitored and integrated using Mass Hunter Qualitative Analysis B.07.00 and Mass Hunter Profinder B.10.0 (Agilent Technologies, Waldbronn, Germany). ## 2.2 Chromatographic conditions Analyses of estradiol, progesterone, androsterone, testosterone, DHEA, DHT, 17α-hydroxyprogesterone, pregnenolone and d3-testosterone were accomplished with the use of ZORBAX Extend C18 chromatographic column (2.1 mm × 100 mm, 3.5 μm; Agilent Technologies, Waldbronn, Germany). The mobile phase was composed of $0.1\%$ aqueous solution of formic acid (phase A) and $0.1\%$ formic acid solution in methanol (phase B). The gradient elution was utilized starting from $60\%$ of phase B to $80\%$ of B in 10 min, then was set at $80\%$ of B for 4 min. The time for stationary phase equilibration was set at 6 min. The flow rate was 0.35 mL/min, the injection volume was 2 µL and the column temperature was maintained at 40°C. ## 2.3 Optimization of the mass spectra (MS) parameters Mass spectra were recorded using full scan in positive ion mode with a scan range from m/z 61 to 1,000 to cover all steroid hormones likely to be detected. The analyses were performed using electrospray ionization source (ESI) with the following optimized parameters: gas temperature (nitrogen) was set at 320°C with flow rate at 10 L/min, nebulizer pressure was set to 40 psi, sheath gas temperature and its flow rate were set at 350°C and 11 L/min, respectively. The capillary voltage was maintained at 3250 V and fragmentor voltage was 150 V. The data were collected as centroids. ## 2.4 Preparation of standard stock solutions The concentrated stock solutions of steroid hormones were prepared at 1 mg/mL in methanol. The working solutions of standards were prepared by dilution of stock solutions with methanol to obtain the following concentrations: 100 μg/mL, 10 μg/mL and 1 μg/mL. Standards at 1 μg/mL concentration level were analyzed separately using UHPLC- ESI-QTOF/MS in a scan mode to evaluate their retention time, ionization adducts and isotopic pattern. For the development of TF-SPME method another working standard solution was prepared by mixing proper amount of each 100 μg/mL standard and methanol to the final concertation of 10 μg/mL. Such a mixture was diluted with $1\%$ of PBS (1:10, v/v) to give the concentration of 1 μg/mL. The stock solutions were stored at −80°C while working standard solutions were kept in −20°C. Proper volume of each working solution was added to urine blank matrix (Surine™ negative urine control) in order to prepare quality control samples (QC) during validation process of SPME extraction. ## 2.5 SPME procedure The developed and optimized SPME procedure was composed of five steps: preconditioning, extraction, washing, desorption and cleaning of sorbents. Each step was performed at room temperature at 1,000 rpm agitation speed. The extraction sorbent was preconditioned with 1 mL of methanol/water (50:50, v/v) for 30 min. Then 1 mL of sample was extracted for 30 min. After this step the blades were washed with deionized water for 10 s and subsequently, the desorption was applied with the mixture of methanol/acetonitrile (50:50, v/v) for 45 min. The samples after desorption phase were evaporated to dryness using vacuum centrifuge at 45°C. The dry residues were dissolved with 200 µL of methanol, centrifuged at 140,000 rpm for 10 min and injected into UHPLC-QTOF/MS system. After desorption, the sorbents were cleaned with the use of the mixture composed of methanol, acetonitrile, isopropanol and water (25:25:25:25,v/v/v/v). All extraction steps were performed at room temperature with 1,000 rpm agitation. The optimization of TF-SPME relied on the evaluation of i) type of extraction sorbent, ii) type of desorption mixture, iii) time of both extraction and desorption processes and iv) volume of desorption mixture. The optimization of TF-SPME method was carried out with the use of mixture of standards as it was mentioned in 2.4. Section, wherein the 100 µL of mixture at concentration of 10 μg/mL spiked with internal standard (10 μL at 10 μg/mL), was dissolved with 900 µL of $1\%$ PBS. Such 1 mL of extraction solvent was transferred to 96 well plates. The exemplary bladed brush coated with two extraction TF sorbents was presented in Figure 1. FIGURE 1: Two types of extraction solid phases used in the study. A-polystyrene divinylbenzene (PS-DVB) and B-polyacrylonitrile C18 (PAN-C18) sorbents. The validation of TF-SPME was performed with the use of urine samples, wherein the 500 µL of urine spiked with internal standard (10 μL, 1 μg/mL) was diluted with 500 µL of $1\%$ PBS. The application of TF-SPME for pooled urine samples was performed following the enzymatic hydrolysis reaction using a modified method described by Klimowska et al. ( Klimowska and Wielgomas, 2018). ## 2.6 Preparation of urine samples and enzymatic hydrolysis procedure using β-glucuronidase from Helix pomatia Steroid hormones which are excreted into urine are mainly their glucuronic or sulphate conjugates as their undergo metabolic II phase biotransformation. In order to detect unconjugated forms of steroids, those hormones in a urine sample should be enzymatically hydrolyzed. In the present work, the enzymatic hydrolysis was utilized with the use of β-glucuronidase obtained from H. pomatia (Type HP-2, activity ≥100,000 units/mL). The procedure was applied for the pooled urine obtained from healthy volunteers ($$n = 6$$). The pooled urine samples were derived from three women and three men (mean age: 41.67 ± 5.32, BMI: 22.43 ± 2.35). Prior to this study, an ethical approval from an independent committee of bioethical research at the Medical University of Gdansk was obtained (number of consent: NKBBN/$\frac{252}{2014}$). The group of healthy volunteers have declared a good health status and did not undergo any medical treatment at the time of urine collection. The collected and pooled urine samples were immediately frozen and stored at −80°C. In each case, before SPME extraction procedure, the urine samples were thawed at room temperature. Then the urine was adjusted to pH = 5 using 1M of acetate buffer. Next, the 500 µL of centrifuged urine, spiked with 10 µL of internal standard (1 μg/mL) was hydrolyzed with 5 µL of β-glucuronidase during 8 h at 37°C. Then, the reaction was stopped by rapidly cooling samples in ice. The 500 µL of urine diluted with 500 µL of $1\%$ PBS was used for the next step of the TF-SPME procedure. ## 2.7.1 Matrix effect, recovery and process efficiency Matrix effect (ME), recovery (RE) and process efficiency (PE) were performed with the use of quality control standards which were set at three concentration levels (LQC, MQC and HQC) using steroid-free urine matrix. Three sets of samples were prepared as follows: Set A was composed of a set of steroid-free urine matrix extracted by SPME. Then the extract was evaporated to dryness and the dry residue was dissolved in methanol and subsequently spiked with QC standards. Such samples were then injected into the UHPLC-QTOF/MS system. Set B consisted of a set of steroid-free urine matrix spiked with QC standards and then extracted by SPME. After extraction, the samples were evaporated to dryness and the dry residue of each sample was dissolved in methanol and then injected into UHPLC-QTOF/MS system. Set C was a set of neat QC standards dissolved in methanol injected into the UHPLC-QTOF/MS system. In each set the area under the peak of each analyte versus area under the peak of internal standard was measured (EMA, 2011). The recovery (RE) was calculated by dividing obtained results from set A by the set B using Equation 2. RE=SET BSET A x 100 % [2] The process efficiency (PE) was calculated by dividing the results obtained from set B by the set C using the Equation 3. PE=SET BSET C x 100 % [3] The matrix effect (ME) was obtained by calculating matrix factor which is the result of dividing set A versus set C using the Equation 4. MF=SET ASET C [4] Next, the average value of MF and standard deviation of MF was calculated thanks to which the matrix effect could be expressed as coefficient of variation. This was obtained using the following equation: ME=MF standard deviationMF average x 100 % [5] ## 3.1 Chromatographic and mass spectra conditions The analyses of eight steroid hormones (estradiol, progesterone, androsterone, testosterone, DHEA, DHT, 17α-hydroxyprogesterone, pregnenolone) along with the internal standard d3-testosterone was accomplished in 14 min using gradient elution composed of $0.1\%$ FA in water and $0.1\%$ FA in methanol according to the gradient program briefly presented in 2.2. Section. The retention time and ionization adducts of each analyte was measured by separate analysis of each steroid hormone in a full scan range from 61 to 1,000 m/z. The exemplary total ion chromatogram (TIC) of steroid hormones mixtures was presented in Figure 2. The peaks of analytes were extracted using Find Compounds by Formula algorithm in Mass Hunter Qualitative Analysis software. FIGURE 2: The representative Total Ion Chromatogram (TIC) of eight steroid hormones along with internal standard detected in full scan mode: one- estradiol, 2-d3-testosterone, 3-testosterone, 4-17α-hydroxyprogesterone, 5-DHEA, 6-DHT, 7-progesterone, 8-androsterone, 9-pregnenolone. Although the last detected steroid hormone eluted at 7.2 min the method lasted 14 min. The gradient elution set from $60\%$ to $80\%$ of organic modifier (phase B) was achieved from 0 to 10 min and then $80\%$ of B was stated till 14 min. Such elution was developed in order to apply this method for separation of other steroid-related hormones in untargeted metabolomics profiling approach. Taking into account the structure of steroid hormones and their derivates possibly detected in urine samples, the gradient elution set to $80\%$ of B seems to be enough to chromatographically separate other steroid-related compounds (Stanczyk et al., 1997; Boyaci et al., 2016). Regarding ionization efficiency, it was observed that some steroid hormones appeared to have higher intensity when ionized with the loss of water molecule rather than by only protonation process. Therefore, such steroid hormones like estradiol, DHEA, androsterone and pregnenolone were monitored as their protonated adducts along with the loss of one molecule of water. The rest steroid hormones were monitored as their protonated adducts. The monitored precursor ions along with their retention times were presented in Table 1. Besides, in Supplementary Material (Supplementary Figure S1) the mass spectra of each detected steroid hormone were presented. TABLE 1 \| Analyte \| Molecular formula \| Retention time [min] \| Molecular mass [amu] \| Monitored ion (m/z) \| \| --- \| --- \| --- \| --- \| --- \| \| Estradiol \| C18H24O2 \| 2.68 \| 272.1776 \| 255.1743 (M + H+-H2O) \| \| D3-Testosterone (ISTD) \| C19H25O2D3 \| 3.19 \| 291.2278 \| 292.2394 (M + H+) \| \| Testosterone \| C19H28O2 \| 3.23 \| 288.2089 \| 289.2167 (M + H+) \| \| 17α-hydroxyprogesterone \| C21H30O3 \| 3.56 \| 330.2195 \| 331.2273 (M + H+) \| \| DHEA \| C19H28O2 \| 3.77 \| 288.2089 \| 271.2059 (M + H+-H2O) \| \| DHT \| C19H30O2 \| 4.74 \| 290.2246 \| 291.2322 (M + H+) \| \| Progesterone \| C21H30O2 \| 5.73 \| 314.2246 \| 315.2319 (M + H+) \| \| Androsterone \| C19H30O2 \| 6.11 \| 290.2246 \| 273.2216 (M + H+-H2O) \| \| Pregnenolone \| C21H32O2 \| 7.22 \| 316.2402 \| 299.2372 (M + H+-H2O) \| ## 3.2 Development and optimization of TF-SPME procedure The main objective of the present study was to develop and optimize the thin-film solid phase microextraction procedure for isolation of eight steroid hormones from urine samples. These steroid hormones are only the example of the widespread application of TF-SPME that is environmentally friendly, fast (if automated) and solvent-saving procedure. The optimized conditions of TF-SPME was optimized on various types of steroid hormones taking into account extraction of other steroid hormones in untargeted metabolomics profiling approach. Due to the capacious extraction phase in thin-film type of SPME in comparison with classical one, the sensitivity enhancement of the method is observed. Besides, the time of extraction can be shortened without the risk of sensitivity reduction. Here, two extraction sorbents were tested, namely, with the use of polystyrene divinylbenzene (PS-DVB) and polyacrylonitrile C18 (PAN-C18) sorbents. Both of these sorbents can be used for steroid-related compounds due to their affinity to low polar and hydrophobic molecules (Boyaci et al., 2016). For these fibers, various types of desorption mixtures were evaluated such as a) ACN:H2O (70:30, v/v); b) ACN:H2O (80:20, v/v); c) ACN:H2O (85:15, v/v); d) ACN:MeOH (50:50, v/v); e) ACN:MeOH:H2O (40:40:20, v/v/v) and f) ACN:MeOH:H2O (45:45:10, v/v/v). For the optimization of the type of sorbent and the desorption mixture, the 60 min extraction and 60 min desorption time was applied. The extraction was performed using three separate replicates of steroid hormones mixture. The results of TF-SPME extraction are presented in Figures 3A–D. FIGURE 3: The peak areas of extracted steroid hormones depending on type of extraction sorbent (A, B): PAN-C18; (C, D): PS-DVB) and type of desorption mixtures. As it is observed in Figure 3, the most efficient desorption mixture was ACN:MeOH (50:50, v/v) where the peak area of each steroid hormone is the highest. Among sorbent types, the PS-DVB one, resulted in higher intensities of analytes than in the case of PAN-C18 sorbent. The exact values of peak area along with the coefficient of variation of the results are presented in Supplementary Materials Table S1. After the selection of the fiber type and desorption mixture, the time of both extraction and desorption processes were evaluated. Due to the thin-film type of SPME, it is supposed that time of each step can be shortened in comparison with a classical SPME, as larger surface area leads to enhanced sensitivity of extraction process. The tested time of extraction and desorption processes were as follows: 30 min, 45 min and 60 min. Firstly, the time of extraction process was evaluated using 60 min desorption time. After final selection of extraction time, the time of desorption process was assessed. These experiments were carried out using four separate replicates and the results are presented in Figure 4. The exact values along with the coefficient of variations are presented in Supplementary Table S2, S3. FIGURE 4: The comparison of peak area of steroid hormones normalized by peak area of internal standard for eight steroid hormones extracted (A, B) and desorbed (C, D) during various time points (30, 45 and 60 min). According to the obtained results related with the assessment of the extraction time, the intensities of analytes normalized by internal standard differ slightly and each time do not enhance significantly the sensitivity of the method. Also the coefficients of variation do not differ between time of extraction process. Therefore, taking into account the time of total SPME procedure, we decided to use 30 min extraction time for the next step of method development. Using this time of extraction, the time of desorption process was assessed. According to the results presented in Figure 4 and Supplementary Table S3, the higher intensities of the steroid hormones normalized by internal standard were observed during 45 min desorption time. The coefficient of variation was in acceptable range (<$15\%$) but was varied between desorption time points. Due to the observed higher results obtained for 45 min desorption time, this time point for desorption step was chosen. The last parameter evaluated in TF-SPME method was the volume of desorption. Here, two types of volumes were tested: 1 mL and 1.5 mL. The influence of applied two volumes of desorption mixture on analyte intensities is presented in Figure 5. FIGURE 5: The comparison of the influence of desorption volume on the intensity of steroid hormones. As it is presented in Figure 4, the lower desorption mixture volume applied, the higher intensity of analytes was observed. Therefore, the final volume of desorption mixture was set to 1 mL. It is important to emphasize, that the total time of TF-SPME procedure without cleaning step lasts less than 2 h. Taking into account 96 well plates present in the SPME apparatus, the extraction could be applied as high-throughput approach in untargeted metabolic profiling of steroid-related compounds. The overall TF-SPME conditions applied for extraction and isolation of steroid hormones on DVB sorbent is presented in Table 2. TABLE 2 \| Name of the step \| Type of solvent \| Time [min] \| Temperature [°C] \| Agitation [rpm] \| \| --- \| --- \| --- \| --- \| --- \| \| Conditioning \| MeOH:H2O 1:1 (v/v) \| 30 \| 20 \| 1000 \| \| Extraction \| Sample \| 30 \| 20 \| 1000 \| \| Washing \| H2O \| 10 s \| 20 \| 0 \| \| Desorption \| ACN: MeOH 1:1 (v/v) \| 45 \| 20 \| 1000 \| \| Cleaning \| ACN: MeOH: IPA: H2O 1:1:1:1 (v/v/v/v) \| 60 \| 20 \| 1000 \| ## 3.3 Validation of the TF-SPME method The developed and optimized method which final conditions were presented in Table 2 was validated in terms of recovery (RE), process (extraction) efficiency (PE) and matrix effect (ME). The calculations were performed using commercially available steroid-free urine matrix that was spiked with QC standards at three concentration levels (LQC = 0.25 μg/mL, MQC = 0.5 μg/mL and HQC = 0.75 μg/mL). The validation steps were briefly explained in 2.7.1. Section. The recovery and process efficiency were calculated using Eqs. 2, 3, respectively. For these values also percentage of standard deviation was calculated. Matrix effect was calculated using Eq. 4 where matrix factor was obtained. The coefficient of variation of matrix factor was calculated using Eq. 5. The results from validation are presented in Table 3. As it can be observed from the Table 3, the recovery and matrix effect are in acceptable range according to EMA validation criteria. TABLE 3 \| Analyte \| Process efficiency \| Process efficiency.1 \| Process efficiency.2 \| Recovery \| Recovery.1 \| Recovery.2 \| Matrix effect [CV %] \| Matrix effect [CV %].1 \| Matrix effect [CV %].2 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Analyte \| LQC (n = 6) \| MQC (n = 6) \| HQC (n = 6) \| LQC (n = 6) \| MQC (n = 6) \| HQC (n = 6) \| LQC (n = 6) \| MQC (n = 6) \| HQC (n = 6) \| \| Androsterone \| 88.0 ± 11.2 \| 86.1 ± 7.9 \| 81.9 ± 5.6 \| 95.8 ± 3.4 \| 95.3 ± 11.0 \| 91.4 ± 6.5 \| 9.1 \| 2.8 \| 3.4 \| \| DHEA \| 95.3 ± 4.8 \| 93.1 ± 7.1 \| 84.9 ± 7.2 \| 106.8 ± 4.8 \| 104.1 ± 8.3 \| 96.0 ± 8.4 \| 4.8 \| 5.0 \| 5.6 \| \| DHT \| 94.6 ± 9.8 \| 91.3 ± 8.3 \| 85.2 ± 6.0 \| 102.8 ± 7.9 \| 104.2 ± 10.7 \| 98.0 ± 7.6 \| 8.9 \| 3.7 \| 3.2 \| \| Estradiol \| 99.2 ± 3.6 \| 90.9 ± 6.6 \| 90.9 ± 4.1 \| 107.1 ± 5.1 \| 111.8 ± 8.1 \| 100.2 ± 5.5 \| 4.7 \| 11.1 \| 1.4 \| \| Pregnenolone \| 93.6 ± 9.6 \| 98.7 ± 8.1 \| 87.7 ± 7.9 \| 102.1 ± 12 \| 108.2 ± 10.5 \| 96.3 ± 10.9 \| 8.9 \| 4.8 \| 3.6 \| \| Progesterone \| 84.8 ± 7.3 \| 83.3 ± 5.9 \| 79.3 ± 4.5 \| 89.9 ± 1.4 \| 91.1 ± 8.4 \| 88.8 ± 5.6 \| 5.7 \| 2.6 \| 2.9 \| \| Testosterone \| 96.9 ± 6.1 \| 91.3 ± 4.1 \| 86.6 ± 2.5 \| 106.2 ± 3.4 \| 105.1 ± 7.5 \| 99.3 ± 4.0 \| 6.0 \| 1.9 \| 2.7 \| \| 17αOHProgesterone \| 92.9 ± 8.5 \| 90.8 ± 4.1 \| 83.2 ± 3.5 \| 101.6 ± 5.6 \| 102.7 ± 8.5 \| 93.8 ± 4.7 \| 9.2 \| 4.0 \| 3.4 \| \| D3-Testosterone \| 85.4 ± 4.9 \| 85.8 ± 4.3 \| 87.8 ± 2.8 \| ----- \| ----- \| ----- \| ----- \| ----- \| ----- \| ## 3.4 Application of enzymatic hydrolysis of pooled urine samples using β-glucuronidase from Helix pomatia As steroid hormones are excreted into urine mainly as glucuronic conjugates, their TF-SPME extraction of unconjugated forms of analytes from urine samples has to be performed after enzymatic hydrolysis. The hydrolysis was utilized using β-glucuronidase from H. pomatia with enzymatic activity ≥100,000 units/mL on pooled urine samples ($$n = 6$$) from healthy volunteers. β-Glucuronidase Type HP-2 from H. pomatia is a crude solution of enzymes derived from the digestive juices of the Roman snail. This type of enzyme (HP-2) has documented β-glucuronidase activity to be more than 100,000 units/mL as well as arylsulfatase activity at 7,500 units/mL level, therefore both glucuronic and sulfate conjugated of steroid hormones can be hydrolysed. The applied hydrolysis procedure was briefly presented in 2.6. Section. After hydrolysis, the pooled urine samples were extracted using validated TF-SPME method. To compare the influence of hydrolysis process on the analytes levels, the pooled urine samples without hydrolysis step were simultaneously extracted and analyzed. The determination was performed using developed and optimized method involving UHPLC-ESI-QTOF/MS instrumentation in a full scan mode. The extracted ion chromatograms (EIC) were prepared using find compounds by formula algorithm applied in Mass Hunter Qualitative Analysis B.07.00. Software. The change in levels of each steroid hormone was presented in Table 4. TABLE 4 \| Analyte \| Average area in pooled urine samples (n = 2) \| Average area in pooled urine samples after enzymatic hydrolysis (n = 5) \| Average change: Hydrolysis vs. without hydrolysis±SD \| Average change: Hydrolysis vs. without hydrolysis±SD.1 \| \| --- \| --- \| --- \| --- \| --- \| \| Estradiol \| ND \| 124046 (detected in 2 samples) \| ----- \| ----- \| \| Testosterone \| 2374715 \| 4474798 \| 1.88 \| ±0.06 \| \| 17α-hydroxyprogesterone \| 3338358 \| 693096 \| 0.21 \| ±0.1 \| \| DHEA \| 176344 \| 17616511 \| 99.9 \| ±2.81 \| \| DHT \| 332358 \| 1265096 \| 3.81 \| ±1.36 \| \| Progesterone \| 5839251 \| 425985 \| 0.07 \| ±0.02 \| \| Androsterone \| 2210570 \| 27497309 \| 12.44 \| ±0.67 \| \| Pregnenolone \| 149226 \| 565234 \| 3.79 \| ±0.91 \| As it can be observed in Table 4, enzymatic hydrolysis significantly improved detection of DHEA. The level of DHEA after hydrolysis is almost 100 times higher than in comparison of its level without hydrolysis step. DHEA exists in urine also as sulphate conjugate but it is known that β-glucuronidases derived from molluscs often contain also sulfatase activity. The levels of such steroid hormones as testosterone, dihydrotestosterone, androsterone and pregnenolone were also from almost 2 to 4 times higher after enzymatic hydrolysis step. Concerning estradiol, this hormone was detected only in two out of five samples after hydrolysis so no comparisons were performed. The last hormones like progesterone and 17α-hydroxyprogesterone were found to have almost 14 and 5 times lower levels after enzymatic hydrolysis, respectively. The reason of that decreased level can be likely associated with another pathways of enzymatic biotransformation of these compounds. Progesterone can be metabolized to its main metabolite pregnanediol-3-glucuronide (PDG) (Stanczyk et al., 1997) so the balance between progesterone level itself can be moved to formation of other metabolites after enzymatic deconjugation like pregnanediol. Above all, the application of enzymatic hydrolysis step before the TF-SPME approach can be utilized in order to ensure better metabolome coverage in other untargeted metabolomics profiling studies. ## 3.5 Application of the TF-SPME method for untargeted metabolomics profiling studies In the present study, the TF-SPME method was developed and validated based on eight steroid hormones from urine samples. However, taking into account the applied PS-DVB sorbent as well as type of desorption mixture (ACN:MeOH, 50:50, v/v) other steroid-related metabolites can be efficiently extracted as well. Additionally, the chromatographic parameters were optimized for determination of wider spectrum of metabolites, while the mass spectra conditions allow for detection of compounds in a very wide range of m/z from 61 to 1,000. In our previous study PS-DVB sorbent in TF-SPME was already applied for untargeted metabolomics study from urine samples, however, the method was not optimized for steroid hormones profiling (Łuczykowski et al., 2021). In the present project the typical untargeted workflow has not been applied but the obtained set of data from pooled urine samples was processed using in-house database created based on Metlin Lipids library. Such database consisted of 43 steroid hormones and their derivatives. As a result, six additional steroids were annotated along with eight steroid hormones previously identified (based on reference standards). In Table 5, the list of additionally annotated steroid hormones is presented with the overall score of annotation set to be above $80\%$. The overall score includes match of isotopic pattern and molecular mass. Further studies in this untargeted steroid profiling are needed with the use of reference standards and MS/MS fragmentation pattern to confirm identity of additionally annotated compounds. TABLE 5 \| Name of analyte \| Molecular formula \| Molecular weight \| m/z \| Retention time [min] \| Overall score \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| Deoxycortisol \| C21H30O4 \| 346.2153 \| 347.2227 \| 2.406 \| 83.43 \| \| Dihydrocortisol \| C21H32O5 \| 364.2258 \| 365.2326 \| 2.422 \| 86.15 \| \| Hydroxyandrosterone \| C19H30O3 \| 306.2199 \| 289.2165 \| 3.225 \| 97.54 \| \| Tetrahydrocorticosterone \| C21H34O4 \| 350.2464 \| 333.2429 \| 3.39 \| 88.41 \| \| Androsterone glucuronide \| C25H38O8 \| 466.257 \| 489.2463 \| 4.027 \| 98.21 \| \| Hydroxypregnenanolone \| C21H34O3 \| 334.2513 \| 317.2482 \| 5.616 \| 98.24 \| ## 4 Conclusion The developed and validated TF-SPME method reported in this manuscript is simple, fast and with minimized influence of matrix effect on detection of steroid hormones in urine samples. The extraction method can be applied for isolation of steroid-related metabolites or other lipophilic compounds in untargeted/targeted metabolomics profiling approach. The utilized determination method involving UHPLC-ESI-QTOF/MS in a scan mode can be also applied for detection of urine samples to ensure the metabolome coverage of steroid related compounds. ## 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 NKBBN/$\frac{252}{2014.}$ The patients/participants provided their written informed consent to participate in this study. ## Author contributions WSL: investigation, methodology, formal analysis, writing the manuscript original draft, figures visualization, edition and revision of the manuscript; BK: performing SPME studies; WR: deglucuronidation protocol methodology; AN: writing the manuscript original draft; BW: supervision of deglucuronidation methodology; MM: supervision and project administration; DS: concept of the study, validation, review and editing of the manuscript, project administration, 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/fmolb.2023.1074263/full#supplementary-material ## References 1. Bojko B., Reyes-Garce´s N., Bessonneau V., Goryński K., Mousavi F., Souza Silva E. A.. Solid-phase microextraction in metabolomics. Trac. Trends Anal. Chem. (2014) 61 168-180. DOI: 10.1016/j.trac.2014.07.005 2. Boyaci E., Goryński K., Viteri C. R., Pawliszyn J.. A study of thin film solid phase microextraction methods for analysis of fluorinated benzoic acids in seawater. J. Chromatogr. A (2016) 1436 51-58. DOI: 10.1016/j.chroma.2016.01.071 3. Brunton L. L.. Goodman & gilman’s the pharmacological basis of therapeutics (2011) 4. do Carmo S. N., Merib J., Carasek E.. Bract as a novel extraction phase in thin-film SPME combined with 96-well plate system for the high-throughput determination of estrogens in human urine by liquid chromatography coupled to fluorescence detection. J. Chromatogr. B Anal. Technol. Biomed. Life Sci. (2019) 1118-1119 17-24. DOI: 10.1016/j.jchromb.2019.04.037 5. EMA Guideline on bioanalytical method validation. (2011) 6. Gupta B. B. P., Lalchhandama K.. Molecular mechanism of glucocorticoid action. Curr. Sci. (2002) 83 1103-1111 7. Jaroch K., Boyaci E., Pawliszyn J., Bojko B.. The use of solid phase microextraction for metabolomic analysis of non-small cell lung carcinoma cell line (A549) after administration of combretastatin. Sci. Rep. (2019) 9 1-9. DOI: 10.1038/s41598-018-36481-2 8. Kataoka H., Ehara K., Yasuhara R., Saito K.. Simultaneous determination of testosterone, cortisol, and dehydroepiandrosterone in saliva by stable isotope dilution on-line in-tube solid-phase microextraction coupled with liquid chromatography-tandem mass spectrometry. Anal. Bioanal. Chem. (2013) 405 331-340. DOI: 10.1007/s00216-012-6479-4 9. Klimowska A., Wielgomas B.. Off-line microextraction by packed sorbent combined with on solid support derivatization and GC-MS: Application for the analysis of five pyrethroid metabolites in urine samples. Talanta (2018) 176 165-171. DOI: 10.1016/j.talanta.2017.08.011 10. Kuo T., McQueen A., Chen T. C., Wang J. C.. Regulation of glucose homeostasis by glucocorticoids. Adv. Exp. Med. Biol. (2015) 872 99-126. DOI: 10.1007/978-1-4939-2895-8_5 11. Lima Gomes P. C. F., Barnes B. B., Santos-Neto Á. J., Lancas F. M., Snow N. H.. Determination of steroids, caffeine and methylparaben in water using solid phase microextraction-comprehensive two dimensional gas chromatography-time of flight mass spectrometry. J. Chromatogr. A (2013) 1299 126-130. DOI: 10.1016/j.chroma.2013.05.023 12. Łuczykowski K., Warmuzińska N., Operacz S., Stryjak I., Bogusiewicz J., Jacyna J.. Metabolic evaluation of urine from patients diagnosed with high grade (HG) bladder cancer by SPME-LC-MS method. Molecules (2021) 26 2194. DOI: 10.3390/molecules26082194 13. Maciążek-Jurczyk M., Bessonneau V., Ings J., Bragg L., McMaster M., Servos M. R.. Development of a thin-film solid-phase microextraction (TF-SPME) method coupled to liquid chromatography and tandem mass spectrometry for high-throughput determination of steroid hormones in white sucker fish plasma. Anal. Bioanal. Chem. (2020) 412 4183-4194. DOI: 10.1007/s00216-020-02657-x 14. Mantha L., Palacios E., Deshaies Y.. Modulation of triglyceride metabolism by glucocorticoids in diet-induced obesity. Am. J. Physiol. (1999) 277 R455-R464. DOI: 10.1152/ajpregu.1999.277.2.R455 15. Miller W. L.. Molecular biology of steroid hormone synthesis. Endocr. Rev. (1988) 9 295-318. DOI: 10.1210/edrv-9-3-295 16. Mirnaghi F. S., Chen Y., Sidisky L. M, Pawliszyn J.. Optimization of the coating procedure for a high-throughput 96-blade solid phase microextraction system coupled with LC–MS/MS for analysis of complex samples. Anal. Chem. (2011) 83 6018-6025. DOI: 10.1021/ac2010185 17. Mirzajani R., Kardani F., Ramezani Z.. A nanocomposite consisting of graphene oxide, zeolite imidazolate framework 8, and a molecularly imprinted polymer for (multiple) fiber solid phase microextraction of sterol and steroid hormones prior to their quantitation by HPLC. Microchim. Acta. (2019) 186 129. DOI: 10.1007/s00604-018-3217-4 18. Mousavi F., Bojko B., Pawliszyn R.. High-throughput solid-phase microextraction-liquid chromatography-mass spectrometry for microbial untargeted metabolomics. Methods. Mol. Biol. (2019) 1859 133-152. DOI: 10.1007/978-1-4939-8757-3_7 19. Olcer Y. A., Tascon M., Eroglu A. E., Boyacı E.. Thin film microextraction: Towards faster and more sensitive microextraction. Trac. Trends Anal. Chem. (2019) 113 93-101. DOI: 10.1016/j.trac.2019.01.022 20. Olędzka I., Kowalski P., Plenis A., Bączek T.. Evaluation of various approaches to the isolation of steroid hormones from urine samples prior to FASS-MEKC analysis. Electrophoresis (2017) 38 1632-1643. DOI: 10.1002/elps.201600509 21. Pawliszyn J.. Handbook of solid phase microextraction (2009) 22. Pawłowski M.. Pharmaceutical chemistry (in polish) (2020) 23. Peñalver A., Pocurull E., Borrull F., Marcé R. M., PenAlver A.. Method based on solid-phase microextraction-high-performance liquid chromatography with UV and electrochemical detection to determine estrogenic compounds in water samples. J. Chromatogr. A (2002) 964 153-160. DOI: 10.1016/s0021-9673(02)00694-5 24. Rogerson F. M., Fuller P. J.. Mineralocorticoid action. Steroids (2000) 65 61-73. DOI: 10.1016/s0039-128x(99)00087-2 25. Rousseau G. G.. Fifty years ago: The quest for steroid hormone receptors. Mol. Cell. Endocrinol. (2012) 375 10-13. DOI: 10.1016/j.mce.2013.05.005 26. Stanczyk F. Z., Gentzschein E., Ary B. A., Kojima T., Ziogas A., Lobo R. A.. Urinary progesterone and pregnanediol. Use for monitoring progesterone treatment. J. Reprod. Med. (1997) 42 216-222. PMID: 9131494 27. Vuckovic D., Pawliszyn J.. Systematic evaluation of solid-phase microextraction coatings for untargeted metabolomic profiling of biological fluids by liquid chromatography-mass spectrometry. Anal. Chem. (2011) 83 1944-1954. DOI: 10.1021/ac102614v 28. Wang S., Geng Y., Sun X., Wang R., Zheng Z., Hou S.. Molecularly imprinted polymers prepared from a single cross-linking functional monomer for solid-phase microextraction of estrogens from milk. J. Chromatogr. A (2020) 1627 461400. DOI: 10.1016/j.chroma.2020.461400 29. Zeelen F. J., Bittar E. E., Bittar N.. Medicinal chemistry of steroids” in molecular and cellular pharmacology (1997) 427-463 30. Zhang Z., Duan H., Zhang L., Chen X., Liu W., Chen G.. Direct determination of anabolic steroids in pig urine by a new SPME-GC-MS method. Talanta (2009) 78 1083-1089. DOI: 10.1016/j.talanta.2009.01.022
--- title: A practical dynamic nomogram model for predicting bone metastasis in patients with thyroid cancer authors: - Wen-Cai Liu - Meng-Pan Li - Wen-Yuan Hong - Yan-Xin Zhong - Bo-Lin Sun - Shan-Hu Huang - Zhi-Li Liu - Jia-Ming Liu journal: Frontiers in Endocrinology year: 2023 pmcid: PMC10025497 doi: 10.3389/fendo.2023.1142796 license: CC BY 4.0 --- # A practical dynamic nomogram model for predicting bone metastasis in patients with thyroid cancer ## Abstract ### Purpose The aim of this study was to established a dynamic nomogram for assessing the risk of bone metastasis in patients with thyroid cancer (TC) and assist physicians to make accurate clinical decisions. ### Methods The clinical data of patients with TC admitted to the First Affiliated hospital of Nanchang University from January 2006 to November 2016 were included in this study. Demographic and clinicopathological parameters of all patients at primary diagnosis were analyzed. Univariate and multivariate logistic regression analysis was applied to build a predictive model incorporating parameters. The discrimination, calibration, and clinical usefulness of the nomogram were evaluated using the C-index, ROC curve, calibration plot, and decision curve analysis. Internal validation was evaluated using the bootstrapping method. ### Results A total of 565 patients were enrolled in this study, of whom 25 ($4.21\%$) developed bone metastases. Based on logistic regression analysis, age (OR=1.040, $$P \leq 0.019$$), hemoglobin (HB) (OR=0.947, $P \leq 0.001$) and alkaline phosphatase (ALP) (OR=1.006, $$P \leq 0.002$$) levels were used to construct the nomogram. The model exhibited good discrimination, with a C-index of 0.825 and good calibration. A C-index value of 0.815 was achieved on interval validation analysis. Decision curve analysis showed that the nomogram was clinically useful when intervention was decided at a bone metastases possibility threshold of $1\%$. ### Conclusions This dynamic nomogram, with relatively good accuracy, incorporating age, HB, and ALP, could be conveniently used to facilitate the prediction of bone metastasis risk in patients with TC. ## Introduction Thyroid cancer (TC) is an uncommon endocrine cancer that accounts for approximately $1\%$ of all new malignancies, roughly 0 - $5\%$ of cancers in men and 1 - $5\%$ of cancers in women (1–3). However, the incidence of thyroid cancer has been increasing for more decades [4]. In the United States, the incidence of thyroid cancer tripled from 4.5 per 100,000 population in 1974 to 14.4 in 2013 [5]. Differentiated thyroid cancer (DTC) of low malignancy accounted for the highest proportion of thyroid cancer ($90\%$), including papillary carcinoma (70-$75\%$) and follicular carcinoma (15-$20\%$) [6]. Undifferentiated carcinomas, which are anaplastic malignancies, accounts for less than $5\%$ [6, 7]. Therefore, the prognosis of patients with thyroid cancer is generally good, with a 10-year survival rate of 80-$95\%$ [8]. Distant metastasis is an important risk factor for patients with thyroid cancer. Compared with simple DTC patients, the 10-year survival rate of patients with distant metastasis is decreased by about $50\%$ [3, 8]. Bone is the third most common metastasis site in patients with TC, occurring in 2-$13\%$ of DTC patients (9–11) Compared with other distant metastases, bone metastases cause bone pain, pathological fractures and spinal cord compression, which significantly impaired their quality of life [12]. Early diagnosis and intervention in such patients was important role for increasing patient survival rates [13, 14]. Bone scintigraphy and other nuclear studies, such as FDG-PET and SPECT, have high sensitivity and specificity for the early detection of bone metastases [15]. However, their use is often limited due to high cost and radiation damage to patients [16]. Thus, it is of great significance to develop a simple and feasible new method for early prediction of thyroid cancer bone metastasis. Nomograms have proven useful as models for predicting the occurrence of clinical events, and can allow visualization of incidence [17]. In this study, we aimed to developed a valid nomogram model for assessing the risk of bone metastases in patients with TC to assist physicians in making accurate clinical decision. ## Patient information This study was approved by the Ethics Committee of the First Affiliated Hospital of Nanchang University, and all participants signed written informed consent form. From January 2006 to November 2016, patients newly diagnosed with TC in our hospital were included in this study. All diagnoses were confirmed by needle biopsy or open surgical biopsy. The exclusion criteria were as follows: [1] Patients with other primary malignancies; [2] Patients with renal and/or liver insufficiency; [3] Patients with bone metabolic disorders; [4] Patients with significant hematological disease; [5] Missing critical information. The detailed screening process is shown in Figure 1. Figure 1: The study flow chart of case screening. Bone scintigraphy was used to identify possible bone metastases in patients. If necessary, magnetic resonance imaging and local computed tomography were conducted to confirm possible diagnoses of bone metastases. ## Data collection Demographic and clinicopathological parameters of all patients at primary diagnosis (before receiving clinical treatment) were collected, including age, serum concentrations of calcium, hemoglobin (HB), free triiodothyronine, free thyroxine 4, thyroid stimulating hormone, alkaline phosphatase (ALP), and common tumor markers (carcinoembryonic antigen, alpha fetoprotein, cancer antigen-125 (CA125), CA153, and CA199). Correlations between clinicopathological parameters and bone metastases were analyzed in patients with TC. ## Statistical analysis All statistical analyses were performed using SPSS (version 26) and R (version 3.6.3) software. Qualitative variables were analyzed by Chi-square test, and quantitative variables were analyzed by Student’s t-test. Univariate analyses were initially used to identify variables that may affect bone metastases, and correlated variables ($P \leq 0.05$) were included in multivariate logistic regression analysis to identify independent predictors of bone metastasis of thyroid cancer. Then, the selected independent risk factors were used to construct a dynamic nomogram for predicting bone metastases. C-index, ROC curve, calibration plot, and decision curve analysis were used to evaluate the performance and clinical usefulness of the model. The consistency index, Harrell’s C-index, was used to evaluate the predictive performance of the nomogram. And a ROC curve was established to compare the performance of nomogram model with independent predictors. Further, the nomogram was subjected to bootstrapping validation with 1,000 replications, to calculate an adjusted C-index and a calibration curve was used to judge predictive consistency. Decision curve analysis was conducted to determine the clinical usefulness of the nomogram by quantifying the net benefits at different probability thresholds. ## Demographic and pathological characteristics A total of 565 patients with TC were enrolled in this study, of whom 25 ($4.21\%$) had bone metastases and 540 ($95.79\%$) had no bone metastases at initial diagnosis. Patient clinical characteristics are detailed in Table 1. Age, ALP and HB were significantly different between the bone metastasis group and the non-bone metastasis group ($P \leq 0.05$). The average age, ALP and HB in the bone metastasis group were 53 years, 119.68U/L and 101.23g/L, respectively. The average age, ALP and HB in the non-bone metastasis group were 43.37 years, 68.83U/L and 123.07g/L, respectively. Table 1 \| Characteristics \| BM \| NBM \| P value \| \| --- \| --- \| --- \| --- \| \| Age(year) \| 53.00 ± 14.12 \| 43.37 ± 14.78 \| 0.003 \| \| Gender (%) \| \| \| 0.763 \| \| Male \| 6(24.0) \| 137(25.36) \| \| \| Female \| 19(76.0) \| 403(74.63) \| \| \| HB(g/L) \| 101.23 ± 24.75 \| 123.07 ± 17.16 \| <0.001 \| \| ALP (U/L) \| 119.68 ± 100.64 \| 68.83 ± 49.30 \| 0.028 \| \| FT3 \| 2.78± 1.33 \| 3.32± 5.63 \| 0.691 \| \| FT4 \| 1.11 ± 0.45 \| 2.25 ± 2.70 \| 0.083 \| \| TSH \| 15.31 ± 29.87 \| 7.30 ± 19.28 \| 0.288 \| \| Ca(mmol/L) \| 2.19± 0.43 \| 2.22 ± 0.28 \| 0.558 \| \| CA125 (u/ml) \| 65.04 ± 13.46 \| 38.66 ± 25.09 \| 0.658 \| \| CA153 (u/ml) \| 11.67 ± 5.33 \| 19.71 ± 29.48 \| 0.515 \| \| CA199 (u/ml) \| 21.57 ± 23.48 \| 30.09 ± 20.11 \| 0.803 \| \| CA724 (u/ml) \| 0.75 ± 0.02 \| 2.72± 1.14 \| 0.191 \| \| CEA (ng/ml) \| 55.31 ± 13.72 \| 32.43 ± 21.89 \| 0.412 \| \| Cyfra21-f \| 7.96 ± 4.04 \| 2.78+ ± 1.99 \| 0.310 \| \| NSE \| 18.60 ± 3.20 \| 23.13 ± 9.44 \| 0.839 \| \| Histopathology \| \| \| 0.661 \| \| Micro papillary carcinoma \| 17(68.0) \| 406(75.3) \| \| \| Eosinophilic follicular carcinoma \| 1(4.0) \| 11(2.0) \| \| \| Medullary carcinoma \| 4(16.0) \| 17(3.1) \| \| \| Undifferentiated carcinoma \| 0 \| 13(2.4) \| \| \| Follicular papillary carcinoma \| 1(4.0) \| 68(12.6) \| \| \| Adenocarcinoma \| 2(8.0) \| 25(4.6) \| \| \| Location of primary tumor \| \| \| 0.945 \| \| Left thyroid cancer \| 8(32.0) \| 195(36.1) \| \| \| Right thyroid cancer \| 13(52.0) \| 246(45.6) \| \| \| Thyroid cancer of isthmus \| 0 \| 5(0.9) \| \| \| Bilateral thyroid cancer \| 4(16.0) \| 94(17.4) \| \| ## Parameter selection Logistic regression analysis was used to screen parameters. Univariate logistic regression analysis showed that age, ALP and HB were significantly correlated with bone metastasis of TC ($p \leq 0.05$). Then the three related variables were included in the multivariate logistic regression analysis, and the results showed that age, ALP and HB were independent risk factors of bone metastasis of TC ($P \leq 0.05$), so these three variables were used to construct the nomogram model (Table 2). Table 2 \| Factors \| Univariate analysis \| Univariate analysis.1 \| Univariate analysis.2 \| Multivariate analysis \| Multivariate analysis.1 \| Multivariate analysis.2 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| \| β \| OR (95%CI) \| P value \| β \| OR (95%CI) \| P value \| \| Age \| 0.045 \| 1.046 (1.015-1.078) \| 0.004* \| 0.039 \| 1.040 (1.006-1.074) \| 0.019* \| \| Gender \| -0.157 \| 0.855 (0.301-2.364) \| 0.762 \| \| \| \| \| HB(g/L) \| -0.057 \| 0.944 (0.924-0.965) \| <0.001* \| -0.06 \| 0.947 (0.926-0.968) \| <0.001* \| \| ALP (U/L) \| 0.006 \| 1.007 (1.002-1.011) \| 0.007* \| 0.006 \| 1.006 (1.002-1.010) \| 0.002* \| \| FT3 \| -0.140 \| 0.870 (0.561-1.347) \| 0.532 \| \| \| \| \| FT4 \| -0.959 \| 0.383 (0.120-1.222) \| 0.105 \| \| \| \| \| TSH \| 0.013 \| 1.013 (0.997-1.030) \| 0.120 \| \| \| \| \| Ca(mmol/L) \| -0.412 \| 0.663 (0.168-2.616) \| 0.557 \| \| \| \| \| CA125 (u/ml) \| 0.001 \| 1.001 (0.997-1.005) \| 0.663 \| \| \| \| \| CA153 (u/ml) \| -0.023 \| 0.977 (0.907-1.052) \| 0.542 \| \| \| \| \| CA199 (u/ml) \| -0.003 \| 0.997 (0.980-1.015) \| 0.766 \| \| \| \| \| CEA (ng/ml) \| 0.007 \| 1.007 (0.999-1.016) \| 0.089 \| \| \| \| \| Cyfra21-f \| 0.873 \| 2.395 (0.606-9.467) \| 0.213 \| \| \| \| \| NSE \| -0.009 \| 0.991 (0.918-1.070) \| 0.822 \| \| \| \| \| Histopathology \| 0.056 \| 1.058 (0.823-1.359) \| 0.661 \| \| \| \| \| Location of primary tumor \| -0.014 \| 0.986 (0.658-1.477) \| 0.945 \| \| \| \| ## A dynamic nomogram model for predicting bone metastases The results of logistic regression analysis including age, ALP, and HB are presented in Table 2. A model that incorporated the above independent predictors was developed and presented as a nomogram (Figure 2). For patients with TC, the total points calculated using the nomogram could be visually converted to the risk of bone metastases. However, the nomogram can become more cumbersome to read and use when there are higher order interaction and smoothers items in the model. To better address this issue, we established a dynamic nomogram based on shiny to simplify the operation of users (Figure 3) [18]. The dynamic nomogram can be used to predict the probability of bone metastasis (and corresponding $95\%$ Confidence interval) for any combination of predictor values. The shiny tabs display the predicted values graphically and numerically, and the model summary presents the underlying information of the model. Everyone can simply use the dynamic nomogram by clicking on the hyperlink (https://liuwencai.shinyapps.io/thyroid/). Figure 2: Nomogram for prediction of bone metastases. Bone metastases prediction nomogram, developed based on patient age, HB, and ALP levels. Figure 3: The dynamic nomogram for prediction of bone metastases. The plot displays probability (with 95% confidence interval) of bone metastases for patients with TC. The actual explanatory values and their corresponding predictions are given in the “Numerical Summary” tab. ## Apparent performance of the nomogram The C-index for the prediction nomogram was 0.825, and bootstrapping validation confirmed a C-index of 0.815, indicating that the model has good discriminatory power. The ROC curve showed that nomogram had good predictive accuracy, which was higher than that of single independent predictor (Figure 4). The calibration curve of the nomogram for predicting of bone metastases in patients with TC demonstrated good agreement (Figure 5). The apparent performance of the nomogram showed good prediction capability. Figure 4: The ROC curve of nomogram and predictors. Figure 5: Calibration curves for the bone metastasis prediction nomogram. The x-axis represents the predicted bone metastasis risk. The y-axis represents actual diagnoses of bone metastases. The diagonal dotted line represents a perfect prediction by an ideal model. The solid line represents the performance of the nomogram, where a closer fit to the diagonal dotted line represents a better prediction. ## Clinical use Decision curve analysis for the nomogram is presented in Figure 6. It showed that if the threshold probability of a patient or doctor is > $1\%$ and < $67\%$, respectively, using the nomogram developed in the current study to predict bone metastasis risk added more benefit than either the intervention-all-patients or intervention-none scheme. Within this range, the net benefit based on the nomogram was comparable, with several overlaps. Figure 6: Decision curve analysis for the bone metastasis prediction nomogram. The dotted line represents the bone metastasis risk nomogram. The thin solid line represents the assumption that all patients have bone metastases. The thick solid line represents the assumption that no patients have bone metastases. The y-axis measures the net benefit. The net benefit was calculated by subtracting the proportion of all patients who are false positive from the proportion who are true positive, weighting by the relative harm of forgoing treatment compared with the negative consequences of an unnecessary treatment. Here, the relative harm was calculated by pt/(1-pt). “pt” (threshold probability) is where the expected benefit of treatment is equal to the expected benefit of avoiding treatment; at which time a patient will opt for treatment informs us of how a patient weighs the relative harms of false-positive results and false-negative results ([a - c]/[b -d] = [1 - pt]/pt); (a–c) is the harm from a false-negative result; (b–d) is the harm from a false positive result. a, b, c and d give, respectively, the value of true positive, false positive, false negative, and true negative (19). The decision curve showed that if the threshold probability of a patient or doctor is > 1% and < 67%, respectively, using the nomogram in the current study to predict bone metastases adds more benefit than the intervention-all-patients or intervention-none schemes. For example, if the personal threshold probability of a patient is 20% (ie, the patient would opt for treatment if his probability of bone metastasis was 20%), then the net benefit is 0.151 when using the nomogram to make the decision of whether to undergo treatment, with added benefit than the intervention-all-patients or intervention-none schemes. The net benefit was comparable, with several overlaps, on the basis of the nomogram. ## Discussion In previous studies, there were many studies related to bone metastasis of TC. Orita Y et al. observed bone metastases in 52 ($3.7\%$) of 1398 patients with DTC [20]. Choksi et al. studied the incidence of bone related events in thyroid cancer and found that the incidence of bone metastasis in TP was $3.9\%$ [21, 22]. Here, we found that the probability of bone metastasis in thyroid cancer was $4.21\%$, which was similar to previous studies. In general, the probability of bone metastasis of TC is not high, but once bone metastasis occurs, its complicated bone pain, spinal cord compression and pathological fracture will seriously affect the quality of life of patients. Bone scintigraphy is usually used to identify possible bone metastases in patients newly diagnosed with TC. However, the American Society of Clinical Oncology has reported the overuse and associated costs of BS in patients with extremely low risk of metastasis [23, 24]. In recent years, many scholars have used artificial intelligence and machine learning techniques to predict cancer metastasis (25–27), and although they have better performance, they are slightly lacking in interpretability due to the black box characteristics of complex algorithms. Currently, nomograms are widely used as prognostic devices in oncology and medicine; these instruments employ user-friendly digital interfaces to increase accuracy, and provide easily understood prognoses, which facilitates better clinical decision making [17, 28]. Therefore, based on the clinical data of 565 thyroid cancer patients, we identified independent risk factors for bone metastasis and constructed a nomogram model to predict the risk of BM in patients with newly diagnosed TC. In the present study, we found three independent risk factors associated with BM, including age, ALP and HB. More importantly, based on the three variables, we developed and validated a practical dynamic nomogram for assessing the risk of bone metastases in newly diagnosed TC patients. Incorporating demographic and clinicopathological feature risk factors into an easy-to-use dynamic nomogram could facilitate prediction of bone metastases. Internal validation in the patient cohort demonstrated good discrimination and calibration of the model. The high C-index indicated that the nomogram model has high accuracy and can be widely used [28]. Overall, the present study provides an accurate prediction tool for bone metastases in patients with TC. In previous studies, age has been demonstrated to influence the prognosis of patients with TC (29–32). Further, age has been reported as a risk factor for bone metastases in patients with lung and breast cancer, and younger patients more prone to bone metastasis (33–35). However, few studies have confirmed an association between age and bone metastases in TC. In this study, we found a statistical correlation between age and bone metastases in patients with TC, and the risk of bone metastases increased with age. This may be related to genetic variation in TC cells in patients at different ages [36]. Several studies have reported that a shortage of red blood cells containing HB was a common complication in cancer patients, affecting more than $50\%$ of cancer patients [37]. Kawai et al. [ 38] identified HB levels is related to bone metastases associated with prostate cancer, while Henke et al. [ 39] reported that HB level is a significant prognostic factor in breast cancer metastases. Moreover, Chen et al. [ 40, 41] found the HB is an independent risk factor for bone metastases of breast and renal cell cancer. In this study, HB levels were significantly lower in patients with TC who had bone metastases than those without, which could predict the probability of bone metastasis of TC. The reason why patients with low HB levels are more prone to bone metastases may be that low HB concentrations can promote tumor cell adherence to bone marrow, thereby promoting bone metastases [38]. As a bone formation marker, total serum ALP was widely used for assessment of bone metastases in breast and prostate cancers, and can effectively reflect osteogenic activity in human patients [42, 43]. Sun et al. [ 44] reported that bone ALP was a surrogate marker of bone metastasis in patients with gastric cancer, while Chen et al. [ 40] showed that ALP was a risk factor for bone metastases in breast cancer, and Huang et al. [ 45] reported that ALP was also a risk factor for bone metastases in bladder cancer. Rao et al. [ 46] showed that, tumor-derived ALP regulated epithelial plasticity, tumor growth, and disease-free survival in patients with metastatic prostate cancer. In this study, both bone-specific ALP and total ALP levels differed significantly between patients with and without bone metastases, where patients with TC and high levels of ALP were more likely to develop bone metastases. In the analysis of risk factors, age, HB, and ALP were associated with bone metastases in patients with TC. And our nomogram suggested that advanced age, lower HB, and higher ALP may be the key individual factors that determine risk of bone metastases for patients with TC. In the model, we considered the weight of each predictor according to the statistical coefficient, and finally obtained a visual prediction chart, which could combine age, HB and ALP to help predict the probability of bone metastasis in TC patients. Further, compared with the traditional nomogram, the dynamic nomogram could easily calculate the incidence of bone metastases in patients with TC for any chosen set of values of the explanatory variables when there were higher-order interaction terms and smoother in the model. Therefore, this dynamic nomogram can be used to evaluate the risk of bone metastasis of TC simply, conveniently and quickly, and assist clinicians to make accurate clinical decisions. However, there are some limitations in this study. Firstly it was based on retrospective data from a single institution, which will inevitably lead to inherent data bias. Thus, prospective and multi-center studies are required to validate our model. Secondly, the consistency of our nomogram was tested robustly using internal validation by bootstrap testing, but external validation is needed. Therefore, we will continue to collect cases to further validate the model, and we hope that other researchers could modify the prediction model with external data. Thirdly, clinical markers that may affect bone metastasis of cancer, including BRAF mutation, calcitonin, TERT mutation, and Ki-67 index, were not included in this study [9, 47, 48], and we will continue to expand our data collection to further improve our model in the future. ## Conclusion In conclusion, here we identified age, ALP, and HB as independent factors of bone metastases in patients with TC. Based on this information, we developed a practical dynamic nomogram, with a relatively good accuracy, to help clinicians predict the risk of bone metastases in patients with TC. The ability to estimate of individual risk provides clinicians and patients with more power to make decisions regarding medical interventions. Nevertheless, the nomogram requires external validation, and further research is needed. ## 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 Ethics Committee of the First Affiliated Hospital of Nanchang University, and all participants signed written informed consent form. The patients/participants provided their written informed consent to participate in this study. ## Author contributions W-CL conceived of and designed the study. W-CL, M-PL, W-YH, Y-XZ, and B-LS performed analysis and generated the figures and tables. W-CL and M-PL wrote the manuscript and S-HH, Z-LL, and J-ML critically reviewed 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. ## References 1. Roodman GD. Mechanisms of bone metastasis. New Engl J Med (2004) 350. DOI: 10.1056/NEJMra030831 2. Muresan M, Olivier P, Leclere J, Sirveaux F, Brunaud L, Klein M. Bone metastases from differentiated thyroid carcinoma. Endocrine-related canc (2008) 15 37. DOI: 10.1677/ERC-07-0229 3. Sherma SI. Thyroid carcinoma. Lancet (2003) 361. DOI: 10.1016/S0140-6736(03)12488-9 4. Davies L, Welch HG. Increasing incidence of thyroid cancer in the united states, 1973-2002. Jama (2006) 295. DOI: 10.1001/jama.295.18.2164 5. Lim H, Devesa SS, Sosa JA, Check D, Kitahara CM. Trends in thyroid cancer incidence and mortality in the united states, 1974-2013. Jama (2017) 317. DOI: 10.1001/jama.2017.2719 6. Ramadan S, Ugas MA, Berwick RJ, Notay M, Cho H, Jerjes W. Spinal metastasis in thyroid cancer. Head Neck Oncol (2012) 4 1-19. DOI: 10.1186/1758-3284-4-39 7. Runowicz CD, Leach CR, Henry NL, Henry KS, Mackey HT, Cowens-Alvarado RL. American Cancer society/American society of clinical oncology breast cancer survivorship care guideline. CA: Cancer J Clin (2016) 66 43-73. DOI: 10.3322/caac.21319 8. Eustatia-Rutten C, Romijn J, Guijt M, Vielvoye G, Van den Berg R, Corssmit E. Outcome of palliative embolization of bone metastases in differentiated thyroid carcinoma. J Clin Endocrinol Metab (2003) 88. DOI: 10.1210/jc.2003-030231 9. Fornetti J, Welm AL, Stewart SA. Understanding the bone in cancer metastasis. J Bone mineral Res: Off J Am Soc Bone Mineral Res (2018) 33. DOI: 10.1002/jbmr.3618 10. Durante C, Haddy N, Baudin E, Leboulleux S, Hartl D, Travagli J. Long-term outcome of 444 patients with distant metastases from papillary and follicular thyroid carcinoma: benefits and limits of radioiodine therapy. J Clin Endocrinol Metab (2006) 91. DOI: 10.1210/jc.2005-2838 11. Wexler JA. Approach to the thyroid cancer patient with bone metastases. J Clin Endocrinol Metab (2011) 96. DOI: 10.1210/jc.2010-1996 12. Tsuzuki S, Park SH, Eber MR, Peters CM, Shiozawa Y. Skeletal complications in cancer patients with bone metastases. Int J Urol (2016) 23. DOI: 10.1111/iju.13170 13. Bhatia R, Ravulapati S, Befeler A, Dombrowski J, Gadani S, Poddar N. Hepatocellular carcinoma with bone metastases: Incidence, prognostic significance, and management-Single-Center experience. J Gastrointest Canc (2017) 48. DOI: 10.1007/s12029-017-9998-6 14. Makino H, Nishio S, Tsubamoto H, Shimada M, Nishikawa R, Kai K. Treatment and prognosis of bone metastasis from cervical cancer (KCOG-G1202s). J Obstet Gynaecol Res (2016) 42. DOI: 10.1111/jog.12956 15. Harel R, Angelov L. Spine metastases: current treatments and future directions. Eur J Canc (2010) 46. DOI: 10.1016/j.ejca.2010.04.025 16. Wood SL, Brown JE. Skeletal metastasis in renal cell carcinoma: current and future management options. Cancer Treat Rev (2012) 38. DOI: 10.1016/j.ctrv.2011.06.011 17. Iasonos A, Schrag D, Raj GV, Panageas KS. How to build and interpret a nomogram for cancer prognosis. J Clin Oncol: Off J Am Soc Clin Oncol (2008) 26. DOI: 10.1200/JCO.2007.12.9791 18. Jalali A, Alvarez-Iglesias A, Roshan D, Newell J. Visualising statistical models using dynamic nomograms. PloS One (2019) 14. DOI: 10.1371/journal.pone.0225253 19. 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 Inf Dec Mak (2008) 8 1-17. DOI: 10.1186/1472-6947-8-53 20. Orita Y, Sugitani I, Matsuura M, Ushijima M, Tsukahara K, Fujimoto Y. Prognostic factors and the therapeutic strategy for patients with bone metastasis from differentiated thyroid carcinoma. Surgery (2010) 147. DOI: 10.1016/j.surg.2009.10.009 21. Choksi P, Papaleontiou M, Guo C, Worden F, Banerjee M, Haymart M. Skeletal complications and mortality in thyroid cancer: A population-based study. J Clin Endocrinol Metab (2017) 102. DOI: 10.1210/jc.2016-3906 22. Liu WC, Li ZQ, Luo ZW, Liao WJ, Liu ZL, Liu JM. Machine learning for the prediction of bone metastasis in patients with newly diagnosed thyroid cancer. Cancer Med (2021) 10. DOI: 10.1002/cam4.3776 23. Schnipper LE, Smith TJ, Raghavan D, Blayney DW, Ganz PA, Mulvey TM. American Society of clinical oncology identifies five key opportunities to improve care and reduce costs: the top five list for oncology. J Clin Oncol: Off J Am Soc Clin Oncol (2012) 30. DOI: 10.1200/JCO.2012.42.8375 24. Schnipper LE, Lyman GH, Blayney DW, Hoverman JR, Raghavan D, Wollins DS. American Society of clinical oncology 2013 top five list in oncology. J Clin Oncol: Off J Am Soc Clin Oncol (2013) 31. DOI: 10.1200/JCO.2013.53.3943 25. Liu W-C, Li M-X, Qian W-X, Luo Z-W, Liao W-J, Liu Z-L. Application of machine learning techniques to predict bone metastasis in patients with prostate cancer. Cancer Manage Res (2021) 13 8723. DOI: 10.2147/CMAR.S330591 26. Li W, Zhou Q, Liu W, Xu C, Tang ZR, Dong S. A machine learning-based predictive model for predicting lymph node metastasis in patients with ewing’s sarcoma. Front Med (Lausanne). (2022) 9. DOI: 10.3389/fmed.2022.832108 27. Li W, Dong Y, Liu W, Tang Z, Sun C, Lowe S. A deep belief network-based clinical decision system for patients with osteosarcoma. Front Immunol (2022) 13. DOI: 10.3389/fimmu.2022.1003347 28. Wei L, Champman S, Li X, Li X, Li S, Chen R. Beliefs about medicines and non-adherence in patients with stroke, diabetes mellitus and rheumatoid arthritis: a cross-sectional study in China. BMJ Open (2017) 7. DOI: 10.1136/bmjopen-2017-017293 29. Kazaure HS, Roman SA, Sosa JA. The impact of age on thyroid cancer staging. Curr Opin Endocrinol Diabet Obes (2018) 25. DOI: 10.1097/MED.0000000000000430 30. Mazurat A, Torroni A, Hendrickson-Rebizant J, Benning H, Nason RW, Pathak KA. The age factor in survival of a population cohort of well-differentiated thyroid cancer. Endocr Connect. (2013) 2. DOI: 10.1530/EC-13-0056 31. Kim SJ, Myong JP, Suh H, Lee KE, Youn YK. Optimal cutoff age for predicting mortality associated with differentiated thyroid cancer. PloS One (2015) 10. DOI: 10.1371/journal.pone.0130848 32. Nixon IJ, Wang LY, Migliacci JC, Eskander A, Campbell MJ, Aniss A. An international multi-institutional validation of age 55 years as a cutoff for risk stratification in the AJCC/UICC staging system for well-differentiated thyroid cancer. Thyroid: Off J Am Thyroid Assoc (2016) 26. DOI: 10.1089/thy.2015.0315 33. Purushotham A, Shamil E, Cariati M, Agbaje O, Muhidin A, Gillett C. Age at diagnosis and distant metastasis in breast cancer–a surprising inverse relationship. Eur J canc (2014) 50. DOI: 10.1016/j.ejca.2014.04.002 34. Liu W, Lee M, Wu S-N, Tong W-L, Li A-A, Sun B-L. Using machine learning methods to predict bone metastases in breast infiltrating ductal carcinoma patients. Front Public Health (2022) 10. DOI: 10.3389/fpubh.2022.922510 35. Li MP, Liu WC, Sun BL, Zhong NS, Liu ZL, Huang SH. Prediction of bone metastasis in non-small cell lung cancer based on machine learning. Front Oncol (2022) 12. DOI: 10.3389/fonc.2022.1054300 36. Kim SM, Kim SY, Park CS, Chang HS, Park KC. Impact of age-related genetic differences on the therapeutic outcome of papillary thyroid cancer. Cancers (2020) 12 448. DOI: 10.3390/cancers12020448 37. Yazdani A, Dorri S, Atashi A, Shirafkan H, Zabolinezhad H. Bone metastasis prognostic factors in breast cancer. Breast Canc: Bas Clin Res (2019) 13 1178223419830978. DOI: 10.1177/1178223419830978 38. Kawai N, Kunimatsu M, Tozawa K, Sasaki M, Kohri K. Human prostate cancer cells adhere specifically to hemoglobin: a possible role in bone-specific metastasis. Cancer letters. (2001) 171. DOI: 10.1016/S0304-3835(01)00595-X 39. Henke M, Sindlinger F, Ikenberg H, Gerds T, Schumacher M. Blood hemoglobin level and treatment outcome of early breast cancer. Strahlentherapie und Onkologie: Organ der Deutschen Rontgengesellschaft [et al]. (2004) 180 45-51. DOI: 10.1007/s00066-004-1123-7 40. Chen WZ, Shen JF, Zhou Y, Chen XY, Liu JM, Liu ZL. Clinical characteristics and risk factors for developing bone metastases in patients with breast cancer. Sci Rep (2017) 7 11325. DOI: 10.1038/s41598-017-11700-4 41. Chen XY, Lan M, Zhou Y, Chen WZ, Hu D, Liu JM. Risk factors for bone metastasis from renal cell cancer. J Bone Oncol (2017) 9 29-33. DOI: 10.1016/j.jbo.2017.10.004 42. Bishop MC, Hardy JG, Taylor MC, Wastie ML, Lemberger RJ. Bone imaging and serum phosphatases in prostatic carcinoma. Br J Urol (1985) 57. DOI: 10.1111/j.1464-410X.1985.tb06353.x 43. Du WX, Duan SF, Chen JJ, Huang JF, Yin LM, Tong PJ. Serum bone-specific alkaline phosphatase as a biomarker for osseous metastases in patients with malignant carcinomas: a systematic review and meta-analysis. J Cancer Res Ther (2014) 10 Suppl. DOI: 10.4103/0973-1482.145842 44. Lim SM, Kim YN, Park KH, Kang B, Chon HJ, Kim C. Bone alkaline phosphatase as a surrogate marker of bone metastasis in gastric cancer patients. BMC canc (2016) 16 385. DOI: 10.1186/s12885-016-2415-x 45. Huang P, Lan M, Peng AF, Yu QF, Chen WZ, Liu ZL. Serum calcium, alkaline phosphotase and hemoglobin as risk factors for bone metastases in bladder cancer. PloS One (2017) 12. DOI: 10.1371/journal.pone.0183835 46. Rao SR, Snaith AE, Marino D, Cheng X, Lwin ST, Orriss IR. Tumour-derived alkaline phosphatase regulates tumour growth, epithelial plasticity and disease-free survival in metastatic prostate cancer. Br J canc (2017) 116. DOI: 10.1038/bjc.2016.402 47. Min Y, Liu X, Hu D, Chen H, Chen J, Xiang K. Risk factors, prognostic factors, and nomogram for distant metastasis in breast cancer patients without lymph node metastasis. Front Endocrinol (2021) 12. DOI: 10.3389/fendo.2021.771226 48. Wei X, Min Y, Feng Y, He D, Zeng X, Huang Y. Development and validation of an individualized nomogram for predicting the high-volume (> 5) central lymph node metastasis in papillary thyroid microcarcinoma. J Endocrinol Invest (2022) 45 507-15. DOI: 10.1007/s40618-021-01675-5
--- title: Metabolomic and cultivation insights into the tolerance of the spacecraft-associated Acinetobacter toward Kleenol 30, a cleanroom floor detergent authors: - Rakesh Mogul - Daniel R. Miller - Brian Ramos - Sidharth J. Lalla journal: Frontiers in Microbiology year: 2023 pmcid: PMC10025500 doi: 10.3389/fmicb.2023.1090740 license: CC BY 4.0 --- # Metabolomic and cultivation insights into the tolerance of the spacecraft-associated Acinetobacter toward Kleenol 30, a cleanroom floor detergent ## Abstract ### Introduction Stringent cleaning procedures during spacecraft assembly are critical to maintaining the integrity of life-detection missions. To ensure cleanliness, NASA spacecraft are assembled in cleanroom facilities, where floors are routinely cleansed with Kleenol 30 (K30), an alkaline detergent. ### Methods Through metabolomic and cultivation approaches, we show that cultures of spacecraft-associated Acinetobacter tolerate up to $1\%$ v/v K30 and are fully inhibited at ≥$2\%$; in comparison, NASA cleanrooms are cleansed with ~0.8-$1.6\%$ K30. ### Results For A. johnsonii 2P08AA (isolated from a cleanroom floor), cultivations with $0.1\%$ v/v K30 yield [1] no changes in cell density at late-log phase, [2] modest decreases in growth rate (~$17\%$), [3] negligible lag phase times, [4] limited changes in the intracellular metabolome, and [5] increases in extracellular sugar acids, monosaccharides, organic acids, and fatty acids. For A. radioresistens 50v1 (isolated from a spacecraft surface), cultivations yield [1] ~$50\%$ survivals, [2] no changes in growth rate, [3] ~$70\%$ decreases in the lag phase time, [4] differential changes in intracellular amino acids, compatible solutes, nucleotide-related metabolites, dicarboxylic acids, and saturated fatty acids, and [5] substantial yet differential impacts to extracellular sugar acids, monosaccharides, and organic acids. ### Discussion These combined results suggest that [1] K30 manifests strain-dependent impacts on the intracellular metabolomes, cultivation kinetics, and survivals, [2] K30 influences extracellular trace element acquisition in both strains, and [3] K30 is better tolerated by the floor-associated strain. Hence, this work lends support towards the hypothesis that repeated cleansing during spacecraft assembly serve as selective pressures that promote tolerances towards the cleaning conditions. ## Introduction Maintaining low biological contamination in cleanroom-type facilities are critical components to spacecraft assembly (Rummel, 1992; NASA, 2011; Frick et al., 2014), crewed spacecraft exploration (Spry et al., 2020), delivery of healthcare facilities (Shams et al., 2016; Rutala and Weber, 2019), and the manufacturing of pharmaceuticals (Sandle, 2015). Common to these confined environments and endeavors are the routine cleansing of non-critical hard surfaces and support equipment with ≥$70\%$ v/v ethyl alcohol, $70\%$ v/v isopropyl alcohol, $3\%$ v/v hydrogen peroxide, and/or disposable disinfectant wipes containing quaternary ammonium compounds (Agalloco and Carleton, 2007; Rutala et al., 2008; Wong et al., 2013; Frick et al., 2014; Checinska Sielaff et al., 2019). For pharmaceutical facilities, and to lesser extent healthcare facilities, the floors are additionally cleansed with disinfectants and/or detergents such as benzalkonium chloride (and other quaternary ammonium compounds), hydrogen peroxide, sodium hypochlorite, and other alternatives (Murtough et al., 2001; Rutala et al., 2008; Sandle, 2012; Eissa et al., 2014; Han et al., 2021; Tembo et al., 2022). While debated in efficacy (Daschner et al., 1982; Meyer and Cookson, 2010; Suleyman et al., 2018), the surface cleaning agents are often rotated in these facilities to minimize potential microbial resistance and/or selection toward the cleaning conditions (Murtough et al., 2001; Sandle, 2012). For NASA cleanrooms (ISO class 8) where Mars and Europa spacecraft are assembled, the floors are routinely cleansed with ~0.8-$1.6\%$ v/v Kleenol 30 (Mogul et al., 2018; Danko et al., 2021), which is an alkaline detergent formulation. Due to the proprietary nature of Kleenol 30, scant information is available regarding the precise chemical composition. Available safety data sheets indicate a composition containing sodium dodecyl benzene sulfonate ($1\%$ w/w), polyethylene glycol mono-nonylphenyl ether (1–$5\%$ w/w), sodium metasilicate (1–$5\%$ w/w), ethylenediaminetetraacetic acid (EDTA; $2\%$ w/w), sodium metasilicate (1–$5\%$ w/w), potassium hydroxide (KOH; ~$2\%$ v/v; presumably prepared from concentrated KOH), and 2-butoxyethanol (10–$15\%$ w/w). When considering the mechanism of cleansing for Kleenol 30, the combined chemical ingredients imply adjustments toward alkalinity (KOH), sequestering of transition and alkaline earth metals through chelation by the metal-binding reagents (EDTA and sodium metasilicate), and sequestering, emulsification, and removal of organics (non-polar, polar, and charged organics) by detergent action (polyethylene glycol mono-nonylphenyl ether, dodecyl benzene sulfonate, and sodium metasilicate). Yet, despite these robust cleaning practices, spacecraft assembly facilities harbor a persistent yet low abundance microbial bioburden (Danko et al., 2021; Hendrickson et al., 2021). Further, molecular measures reveal the presence of a core microbiome across the spacecraft assembly facilities, International Space Station (Checinska et al., 2015; Checinska Sielaff et al., 2019), and clinical and operating facilities (Shams et al., 2016; Ellingson et al., 2020; Perry-Dow et al., 2022). In spacecraft assembly facilities, the microbiomes are reasonably diverse though generally low in total counts with measures of ~103–104 16S rRNA m−2 (intact cells), ~104–105 16S rRNA m−2 (total cells), 1–40 OTU m−2, ~102–104 colony forming units m−2, and ~ 101 spores m−2 (La Duc et al., 2012; Mahnert et al., 2015; Moissl-Eichinger et al., 2015). Among the dominant members of the core microbiome across spacecraft and healthcare facilities are the Acinetobacter (La Duc et al., 2012; Mora et al., 2016; Hendrickson et al., 2021). The Acinetobacter are a Gram-negative bacterial genus associated with desiccation tolerance (McCoy et al., 2012; Farrow et al., 2018), radiation tolerance (La Duc et al., 2003; McCoy et al., 2012), bioemulsification (Pirog et al., 2021), biofilm formation (Yeom et al., 2013; Pompilio et al., 2021), and multi-drug resistance (Peleg et al., 2008). In Mogul et al. [ 2018], we showed that Acinetobacter radioresistens 50v1, which was isolated from the surface of the pre-flight Mars Odyssey orbiter, could be cultivated in the presence of $1\%$ v/v Kleenol 30, a floor detergent, while utilizing ethanol, a surface cleaner, as a sole carbon source. Additionally, cultivations of A. radioresistens 50v1 with Kleenol 30 were shown to yield tri, penta, and octaethylene glycols, which is suggestive of partial biodegradation of polyethylene glycol mono-nonylphenyl ether (a component of the Kleenol 30 formulation) via scission of the ether linkage (White, 1993) – to yield the mixed polyethylene glycols (e.g., tri, penta, and octaethylene glycol). In addition, studies show that clinical strains of Acinetobacter baumanii tolerate benzalkonium chloride, which is a biocide and surface cleaner, through biofilm formation (Rajamohan et al., 2009) and increased expression of efflux pump genes (Fernández-Cuenca et al., 2015; Srinivasan et al., 2015). Further, survivability studies on cleanroom-associated Acinetobacter reveal extreme tolerances toward aqueous hydrogen peroxide, a surface and floor disinfectant; where the tolerances are perhaps the highest among Gram-negative bacteria (McCoy et al., 2012; Derecho et al., 2014; Mogul et al., 2018). These combined survival characteristics are likely key contributors to the Acinetobacter being among the dominant members of the spacecraft and healthcare microbiomes. Therefore, to obtain molecular and quantitative insights into the tolerances of the cleanroom-associated Acinetobacter, we measured the impacts of Kleenol 30 on the survivals, cultivation kinetics, and intracellular and extracellular metabolomes. By design, we profiled Acinetobacter strains isolated from differing sub-locations within the NASA cleanrooms for spacecraft assembly (e.g., floor and spacecraft surface) which were subjected to differing cleaning regimes (e.g., Kleenol 30 for the floors vs. isopropyl alcohol and/or ethyl alcohol for spacecraft surfaces). Hence, comparisons across these sub-environments lend support toward the hypothesis that repeated cleansing under the conditions of spacecraft assembly serve as selective pressures that favor or promote biochemical tolerances toward the local cleaning conditions. ## Materials and conditions Bacterial strains of A. johnsonii 2P08AA and A. radioresistens 50v1 were obtained through the Planetary Protection Culture Collection at the Jet Propulsion Laboratory. Stocks of ethanol (ENG Scientific) and Kleenol 30 (Mission Laboratories, Los Angeles, CA; Clovis Janitorial) were sterile filtered and stored as 200 μL aliquots at 4°C. Stock solutions of 25 mM Fe2+ were prepared using ferrous ammonium sulfate (Fe(NH4)2(SO4)2·6H2O; EM Science) in ultrapure water (18 MΩ−1), followed by sterile filtration (0.2 μm syringe filter, VWR), and storage as 200 μL aliquots at 4°C (with no visible precipitation over long term storage). Concentrated minimal media (5x MM) were prepared using 30.0 g Na2HPO4·7H2O (Sigma), 15.0 g KH2PO4 (EM Science), 2.50 g NaCl (Fisher Scientific), 5.00 g NH4Cl (EM Science) per liter of ultrapure water. Low-osmolarity media (0.2x M9) were prepared by adding 0.4929 g MgSO4·7H2O (EM Science) and 0.0147 g CaCl2·2H2O (EM Science) to 200.0 mL 5x MM and dilution to 1.00 L using ultrapure water (to yield 1x M9), followed by an additional 5-fold dilution (200 mL 1x M9 in 1.00 L) to yield 0.2x M9. Final concentrations for ions in 0.2x M9 were 20 μM Ca2+, 400 μM Mg2+, 4.4 mM K+, 10.7 mM Na+, and 3.7 mM NH4+, along with 5.4 mM Cl−, 4.5 mM H2PO4−, 4.4 mM HPO42−, and 0.4 mM SO42−. Lysogeny broth media (LB) were prepared using 10.0 g tryptone (VWR), 5.0 g yeast extract (Amresco), 5.0 g NaCl (Fisher Scientific) and 1 mL of 1 M NaOH (Sigma-Aldrich) per liter of ultrapure water. Agar plates were prepared using 1.0 L LB and 15 g agar (Amresco). Cultivations were performed in 15 mL screw cap test tubes (Pyrex, borosilicate glass; 13 × 100 mm), which were washed with tap water, rinsed with distilled water, and autoclaved between experiments. All screw cap test tubes were capped tightly and wrapped with parafilm during cultivation to prevent loss of ethanol, which served as a sole supplied carbon source. Cultivations were performed with mild agitation (200 rpm) using a New Brunswick Scientific Innova 4,200 incubator. Cell densities were monitored during cultivation by following agar plate counts and changes in optical density (OD) at 600 nm (Spectronic 20 Genesys). All media were autoclaved at 121°C and 15 psi for 45 min. ## Cultivations with Kleenol 30 Glycerol stocks of A. radioresistens 50v1 and A. johnsonii 2P08AA were separately streaked onto LB agar plates and incubated at 32°C for ~24 h. Isolated colonies were inoculated into 2.0 mL 0.2x M9 containing 25 μM Fe2+ and 0–$2.0\%$ v/v Kleenol 30 (or 0–40.0 μL). Pre-cultures were initiated by addition of 20.0 μL $90\%$ v/v ethanol to yield final concentrations of 150 mM or $1.0\%$ v/v ethanol. Pre-cultures were grown to late log phase (OD ~0.6 at ~12 h). Target cultures were prepared using fresh media (2.00 mL) which were inoculated with 20.0 μL (1:100 dilution) of the respective pre-cultures and initiated by addition of $1.0\%$ v/v ethanol. Temporal changes in OD were followed for 0–15 h. Growth kinetics were characterized by regression analysis (Microsoft Excel) using a modified version of the Gompertz equation (Begot et al., 1996), which describes a non-linear bacterial growth model, and yields the parameters of growth rate (k), lag time (L), and an estimate of the maximum change in relative biomass (log(N/N0)). All regressions were minimized by least squares analysis. Control cultures containing no inoculate showed no growth (OD ≤ 0.002), as did cultures containing inoculate but no ethanol (OD ≤ 0.002), which indicated negligible biological contamination and accumulation of abiotic particles during cultivation. Survivals of late-log phase cultures (OD ~0.4) were assessed by plating onto LB agar plates. In control experiments, plate counts using 0.2x M9 agar plates supplemented with ethanol (just prior to use) yielded irreproducible results, when compared to LB agar plate, likely due to variances in adsorption of ethanol (under our conditions). For the plate count assays, therefore, aliquots (100 μL) of the cultures (in 0.2x M9/Fe) were transferred to 2.5 mL microcentrifuge tubes, decimally diluted by 106-fold using 0.2x M9, and spread (20 μL) onto LB agar plates using sterile plastic cell spreaders. All plates were sealed with parafilm and incubated at 32°C for 24 h. Plates with ≤300 colonies were enumerated and expressed as colony forming using per mL of the parent culture (cfu mL−1). ## Metabolomics of Acinetobacter cultivations in Kleenol 30 Cultures of A. radioresistens 50v1 and A. johnsonii 2P08AA were prepared as described and harvested during early stationary phase (OD ~0.6–0.7, 7–8 h) by centrifugation (3,500 rpm) at 4°C for 15 min (Beckman Coulter Allegra 21R). After centrifugation, the cell pellets and supernatant fractions were separated and, respectively, treated. The supernatants were saved as 500 μL aliquots, dried using a DNA 110 Savant DNA SpeedVac, and stored at −80°C. The pelleted cells were washed twice with 0.2x M9 and stored at −80°C. Samples were characterized by the West Coast Metabolomics Center using gas chromatography and time-of-flight mass spectrometry (GC-TOF/MS). In brief, dried cells and culture broth were separately resuspended in 2 mL of pre-chilled (−20°C) and degassed extraction solvent (acetonitrile:isopropanol:water, 3:3:2), vortexed for 30 s, shaken for 5 min at 4°C, clarified by centrifugation (~12,000 g), and the resulting supernatant evaporated to dryness. Samples were derivatized by resuspension in 10 μL of 40 mg/mL methoxyamine hydrochloride in pyridine (30°C, 1.5 h) followed by addition of 41 μL N-methyl-N-(trimethylsilyl) trifluoroacetamide (80°C, 30 min); fatty acid methyl esters (e.g., C8-C30) were additionally added to serve as retention index markers (Barupal et al., 2019). Samples were transferred to crimp top vials and separated on an Agilent 6,890 Gas Chromatograph equipped with a Gerstel automatic liner exchange system (ALEX), multipurpose sample (MPS2) dual rail, Gerstel CIS cold injection system (Gerstel, Muehlheim, Germany), and built-in gas purifier (Airgas, Radnor PA). Chromatographic separation was afforded using a 30 m (0.25 mm i.d.) Rtx-5Sil MS column (0.25 μm $95\%$ dimethyl $5\%$ diphenyl polysiloxane film) with an additional 10 m integrated guard column (Restek, Bellefonte PA). Carrier gas was $99.9999\%$ pure Helium with a constant flow of 1 mL/min. Sample volumes of 0.5 μL were injected with a 10 μL s−1 injection speed on a spitless injector with a purge time of 25 s. Temperature profile included 1 min at 50°C for the oven temperature, an increase of 20°C min−1 to 330°C across 14 min, with a final hold for 5 min at 330°C. A temperature of 280°C was used for the transfer line between the gas chromatograph and Leco Pegasus IV time of flight mass spectrometer (single mass analyzer). The measured mass range was 85–500 Da, scan rate was at 17 spectra s−1, electron energy was 70 eV, ion source temperature was 250°C, and detector voltage was 1850 V. Spectra were acquired using the Leco ChromaTOF software vs. 2.32 (St. Joseph, MI) and processed using the BinBase database system (Fiehn et al., 2005, 2010; Trigg et al., 2019), which quantifies (signal-to-noise ratio of 5:1) and matches mass peaks to the Fiehn mass spectral library using retention index (RI) and reference mass spectral information (which are internally compiled by the West Coast Metabolomic Center). Metabolomic data (Study ID ST002380, DatatrackID: 3552) were stored at the NIH Common Fund’s National Metabolomics Data Repository (Sud et al., 2016). ## Statistical analyses Comparisons of the cultivation kinetic parameters were conducted using unpaired Student’s t-tests and one- and two-way ANOVA analyses (Microsoft Excel), where statistical relevance was accepted at $p \leq 0.05$, normal distribution of the data was supported by Shapiro–Wilk tests (Past 4.10; Hammer et al., 2001), and variances were assumed as equal. Metabolomic data were compared using univariate, multivariate, and visual approaches. Discrete changes in the metabolomes (e.g., trends for a single metabolite) were identified using unpaired Student’s t-tests (Microsoft Excel) with corrections for multiple testing using a false discovery rate (FDR) of FDR ≤ 0.20 (Benjamini and Hochberg, 1995). For the 700 metabolites in the total study (whole cell and extracellular metabolomes from 2 bacterial strains treated with and without Kleenol 30), normal distributions (Shapiro–Wilk Tests) were indicated for ~$92\%$ [646] of the metabolites. To account for potential underestimations of normality given the sample size ($$n = 3$$), metabolites exhibiting thresholds for normality of $p \leq 0.03$ (Shapiro–Wilk Tests) and significance of $p \leq 0.05$ (Student’s t-tests) were carried forward in the univariate and visual assessments of the data. Broad structural and metabolic trends were obtained by visualizing the changes in the metabolomes ($p \leq 0.05$) using MetaMapp 1 and Cytoscape 3.9.1 2 (Barupal et al., 2012), which constitute a statistical organizational approach to yield visual maps of metabolites arranged by known structural patterns and metabolic pathways. Confirmation of broad changes in the metabolomes were obtained using ChemRich 3 (Barupal and Fiehn, 2017), a statistical enrichment tool that compares groups of metabolites based on structural and biochemical classes. For this study, the standard and user-defined classifications in ChemRich included amino acids, monosaccharides, sugar acids, organic acids, fatty acids, lipid-related metabolites, nucleotide-related metabolites, and compatible solutes. Multivariate tests were conducted using canonical correspondence analyses (Past 4.10; Hammer et al., 2001) to correlate the metabolomic trends to differing cultivation growth parameters and conditions. ## Cultivation and survival in the presence of Kleenol 30 In Figure 1A we show that the cleanroom-associated Acinetobacter tested in this study, A. johnsonii 2P08AA and A. radioresistens 50v1, readily grow in the presence of ≤$1\%$ v/v Kleenol 30 (K30) under low-osmolarity aqueous conditions with ethanol serving as the sole supplied carbon source. Cultivations on $2\%$ v/v K30 did not yield measurable cell densities. Figure 1: Impacts of 0–1.0% v/v Kleenol 30 (K30) on (A) the survival and (B–D) cultivation kinetics of Acinetobacter johnsonii 2P08AA (red) and Acinetobacter radioresistens 50v1 (blue). Survival (N/N⊝) is expressed as the ratio of plate counts (cfu mL−1) from cultures grown in the absence of K30 (N⊝) and presence of K30 (N). Fitted regressions of growth curves yielded changes in (B) growth rates (h−1), (C) lag time (h), and (D) maximum relative biomass (log (N/N0)), which is assumed to represent a ratio of the total biomass (N) at stationary phase and the biomass at the start of the culture (N0). Error bars represent the standard errors (n = 4–5, growth curves; n = 3–5, plate counts). Univariate tests are represented as asterisks () for t-tests with p < 0.05, hashtags (#) for t-tests with p ≥ 0.05, and asterisks () with an underlying line for one or two-way ANOVA with p < 0.05. In $0.1\%$ v/v K30, cultures of the floor-associated A. johnsonii 2P08AA show no apparent loss in survival, as plate counts (~late log phase) effectively show no difference ($p \leq 0.05$) in the presence (2.1 ± 0.5 × 107 cfu mL−1) and absence (2.6 ± 0.1 × 107 cfu mL−1) of the detergent. These trends indicate quantitative survival of the 2P08AA strain in the presence of $0.1\%$ v/v K30. For the spacecraft surface-associated A. radioresistens 50v1, in contrast, survivability in $0.1\%$ v/v K30 readily decreases ($$p \leq 0.006$$) to 46 ± $6\%$ (propagated error) – as calculated by comparison of cultures cultivated in the presence (4.0 ± 0.4 × 107 cfu mL−1) and absence (8.8 ± 0.7 × 107 cfu mL−1) of $0.1\%$ v/v K30. These trends are indicative of ~$50\%$ survival in the presence of $0.1\%$ v/v K30. In $1.0\%$ v/v K30, survivabilities for the 2P08AA strain (26 ± $12\%$) and 50v1 strain (22 ± $8\%$) decrease to similar values – as indicated by comparison of cultures cultivated in the presence (2P08AA, 0.7 ± 0.3 × 107 cfu mL−1; 50v1, 1.9 ± 1.1 × 107 cfu mL−1) and absence (2P08AA, 2.6 ± 0.1 × 107 cfu mL−1; 50v1, 8.8 ± 0.7 × 107 cfu mL−1) of $1.0\%$ v/v K30. These trends are indicative of ~70–$80\%$ inhibition in the presence of $1.0\%$ v/v K30 across both strains. ## Impacts of Kleenol 30 on the cultivation kinetics Displayed in Figures 1B–D and Figure 2 are the cultivation kinetic parameters and associated growth curves for A. johnsonii 2P08AA and A. radioresistens 50v1. Optical measurements (Figure 2, empty blue circles) for cultures were converted to the ratiometric changes in relative biomass (N/N0) by assuming a direct relationship between optical transmittance and cell density for the initial culture media (N0) and at the time of measurement (N), as outlined in Begot et al. [ 1996]. Changes in relative biomass (Figure 2, filled red circles, fitted line) were expressed as a log function (log(N/N0)), plotted over time, and fit to a modified version of the Gompertz equation (Eq. 1), as detailed in Begot et al. [ 1996] and Mogul et al. [ 2018]. Minimized regressions yielded the parameters of growth rate (k; cell divisions h−1), estimated time in lag phase (L; h), and log of the maximum increase in relative biomass (log(N/N0)max; unitless dimension). Figure 2: Measured and fitted growth curves for A. johnsonii 2P08AA (left panel) and A. radioresistens 50v1 (right panel) cultured in the presence of 0–1.0% v/v Kleenol 30 at 28°C in 0.2x M9 media containing 25 μM Fe2+ and 150 mM ethanol (the sole supplied carbon source). Optical density measurements (blue empty circles; right-hand y-axis) were converted to log(N/N0) (red filled circles; left-hand y-axis) as described and fit by non-linear regression to yield the parameters of time in lag phase, growth rates, and maximum ratiometric and logarithmic change in biomass (log (N/N0)) at stationary phase. To account for optical scattering, the maximum increases in relative biomass (log(N/N0)max) were broadly interpreted as maximum increases in intact cells, cellular aggregation, and/or intra- and extracellular polymeric substances at stationary phase. Reported in this study are the averaged values from the minimized regressions along with standard errors ($$n = 3$$–5 biological replicates). Comparisons of the growth rates (k) show ~$18\%$ decreases for cultivations of the 2P08AA strain in $0.1\%$ v/v K30 ($$p \leq 0.029$$) – as calculated by comparison of the difference in growth rates (Figure 1B) between the absence (0.50 ± 0.01 h−1) and presence (0.41 ± 0.02 h−1) of $0.1\%$ v/v K30. The 50v1 strain, in contrast, shows no change in growth rate ($$p \leq 0.05$$) in $0.1\%$ v/v K30 – as calculated by comparison of rates (Figure 1B) in the absence (0.39 ± 0.01 h−1) and presence (0.40 ± 0.02 h−1) of $0.1\%$ v/v K30. When cultivated in $1.0\%$ v/v K30, growth rates for the 2P08AA strain (0.22 ± 0.03 h−1) and 50v1 strain (0.35 ± 0.01 h−1) decrease by ~$56\%$ ($$p \leq 0.001$$) and ~ $13\%$ ($$p \leq 0.043$$), respectively, as calculated by comparison rates (Figure 1B) in the absence (2P08AA, 0.50 ± 0.01 h−1; and 50v1, 0.39 ± 0.01 h−1) of the detergent. Comparisons by one factor ANOVA analyses confirm that cultivations in the presence of 0.1–$1.0\%$ v/v K30 significantly impact the respective growth rates for the 2P08AA ($$p \leq 0.038$$, $f = 4.256$) and 50v1 ($$p \leq 0.0002$$, $f = 4.459$) strains. Additionally, two factor ANOVA analyses confirm that cultivations in the presence of K30 differentially impact the growth rates of the 2P08AA and 50v1 strains ($$p \leq 0.015$$, $f = 5.318$). These combined trends for exponential phase behavior are indicative of higher tolerances toward K30 by the 50v1 strain at 0.1 and $1.0\%$ K30. These trends are opposite to those from plate counts at late-log phase, which indicate a higher tolerance for the 2P08AA strain. Comparisons of the lag times (L) in 0.1 and $1.0\%$ v/v K30 show decreases for both strains during cultivation (Figure 1C), which is unexpected since the addition of detergents was presumed to inhibit growth and increase the time in lag phase. For the 2P08AA strain, the native lag time of 0.77 ± 0.23 h−1 (absence of K30) reduces to negligible values (~0 h) in 0.1 and $1.0\%$ v/v K30. For the 50v1 strain, the native lag time of 2.61 ± 0.24 h−1 is ~3-fold longer than 2P08AA strain and decreases by ~$70\%$ in $0.1\%$ v/v K30 (0.76 ± 0.15 h−1) and to negligible values in $1.0\%$ v/v K30. Comparisons by one factor ANOVA analyses confirm that cultivations in the presence of 0.1–$1.0\%$ v/v K30 significantly impact the respective lag times for the 2P08AA ($$p \leq 0.00008$$, $f = 4.256$) and 50v1 ($$p \leq 0.015$$, $f = 4.459$) strains. Additionally, two factor ANOVA analyses confirm that cultivations in the presence of K30 differentially impact the lag times for the 2P08AA and 50v1 strains ($$p \leq 0.006$$, $f = 5.318$). These combined trends indicate that K30 induces accelerated entry into the exponential phase in a concentration dependent manner, where the lag phase for the 2P08AA strain is effectively eliminated at ~10-fold lower K30 abundances when compared to the 50v1 strain – under these conditions. Comparisons of the maximum change in relative biomass (log(N/N0)max) for cultivations in $0.1\%$ v/v K30 show unexpected increases for both strains (Figure 1D). For the 2P08AA strain, cultivations in $0.1\%$ v/v K30 yield ~$40\%$ increases ($$p \leq 0.026$$) in maximum apparent biomass when comparing values obtained in absence (2.30 ± 0.02) and presence of (2.44 ± 0.04) of $0.1\%$ v/v K30 and accounting for the log transformation. Similarly, for the 50v1 strain, cultivations in $0.1\%$ v/v K30 yield ~$55\%$ increases ($$p \leq 0.048$$) in maximum apparent biomass when comparing values obtained in absence (2.28 ± 0.01) and presence of (2.47 ± 0.04) of $0.1\%$ v/v K30 and accounting for the log transformation. When cultivated in $1.0\%$ v/v K30, the 2P08AA strain (1.53 ± 0.18) and 50v1 strain (2.07 ± 0.03) exhibit decreases of ~$80\%$ and ~ $40\%$ in maximum apparent biomass, respectively, which is consistent with the decreased growth rates under these conditions (~56 and ~ $13\%$, respectively). Comparisons by one factor ANOVA analyses confirm that cultivations in the presence of 0.1–$1.0\%$ v/v K30 significantly impact the respective maximum changes in relative biomass for the 2P08AA ($$p \leq 0.000003$$, $f = 4.256$) and 50v1 strains ($$p \leq 0.000004$$, $f = 4.459$). Additionally, two factor ANOVA analyses confirm that cultivations in the presence of K30 differentially impact the maximum changes in relative biomass for the 2P08AA and 50v1 strains ($$p \leq 0.007$$, $f = 5.318$). Combined, these trends suggest that $0.1\%$ v/v K30 induces changes in apparent biomass in stationary phase. ## Metabolomic considerations Metabolomes from whole cells of A. johnsonii 2P08AA and A. radioresistens 50v1 were compared after cultivation in the absence and presence of $0.1\%$ v/v Kleenol 30 (early stationary phase). Per strain, whole cell extracts yielded a total of 119 intracellular metabolites and free metabolites from the membrane (e.g., free fatty acids and monoacylglycerols). Metabolites from both strains showing important changes ($p \leq 0.05$) due to cultivation with K30 are listed in Table 1, as are changes (fold-changes) in the respective abundances; changes retaining significance ($p \leq 0.05$, FDR ≤ 0.20) after correction for multiple testing are underlined. Table 1 \| Whole cell metabolites \| Whole cell metabolites.1 \| Whole cell metabolites.2 \| Whole cell metabolites.3 \| Whole cell metabolites.4 \| 2P08AA \| 2P08AA.1 \| 50v1 \| 50v1.1 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| ID# \| Biochemical class \| CID \| RI \| m/z \| Fold-change \| Fold-change \| Fold-change \| Fold-change \| \| Amino acids \| Amino acids \| Amino acids \| Amino acids \| Amino acids \| ↓ \| ↑ \| ↓ \| ↑ \| \| 1 \| Glycine \| 750 \| 368707 \| 248 \| – \| – \| 1.7 \| \| \| 2 \| Lysine \| 5962 \| 663483 \| 156 \| – \| – \| 4.8* \| \| \| 3 \| Proline \| 145742 \| 364523 \| 142 \| – \| – \| 6.0 \| \| \| 4 \| Serine \| 951 \| 395020 \| 218 \| – \| – \| 1.7 \| \| \| 5 \| Valine \| 6287 \| 313502 \| 144 \| – \| – \| 2.2 \| \| \| 6 \| β-alanine \| 239 \| 435564 \| 248 \| – \| – \| 2.1 \| \| \| 7 \| Citrulline \| 750 \| 621404 \| 157 \| – \| – \| 2.9 \| \| \| 8 \| Homoserine \| 12647 \| 396135 \| 146 \| – \| – \| 2.4* \| \| \| 9 \| Ornithine \| 6262 \| 594051 \| 174 \| – \| – \| 2.9* \| \| \| 10 \| Alanine–alanine \| 5484352 \| 636898 \| 188 \| – \| – \| 1.7 \| \| \| Dicarboxylic acids \| Dicarboxylic acids \| Dicarboxylic acids \| Dicarboxylic acids \| Dicarboxylic acids \| ↓ \| ↑ \| ↓ \| ↑ \| \| 11 \| Fumaric acid \| 444972 \| 390016 \| 245 \| – \| – \| \| 2.2 \| \| 12 \| Malic acid \| 525 \| 463180 \| 233 \| – \| – \| \| 2.2 \| \| 13 \| Tartartic acid \| 444305 \| 534291 \| 292 \| – \| – \| \| 1.4 \| \| Sugar acids \| Sugar acids \| Sugar acids \| Sugar acids \| Sugar acids \| ↓ \| ↑ \| ↓ \| ↑ \| \| 14 \| Gluconic acid \| 6857417 \| 693148 \| 333 \| \| 5.1 \| – \| – \| \| 15 \| 3-phosphoglycerate \| 724 \| 610734 \| 227 \| – \| – \| 4.0 \| \| \| 16 \| Glyceric acid \| 439194 \| 377495 \| 189 \| – \| – \| 4.1 \| \| \| Monosaccharides \| Monosaccharides \| Monosaccharides \| Monosaccharides \| Monosaccharides \| ↓ \| ↑ \| ↓ \| ↑ \| \| 17 \| Arabinose \| 6902 \| 550621 \| 217 \| 1.8 \| \| 1.9 \| \| \| 18 \| Mannitol \| 6251 \| 663215 \| 319 \| – \| – \| 3.7* \| \| \| 19 \| Sorbitol \| 5780 \| 667922 \| 217 \| \| 4.9 \| – \| – \| \| Fatty acids and lipids \| Fatty acids and lipids \| Fatty acids and lipids \| Fatty acids and lipids \| Fatty acids and lipids \| ↓ \| ↑ \| ↓ \| ↑ \| \| 20 \| Behenic acid (22:0) \| 8215 \| 920648 \| 117 \| \| 15 \| – \| – \| \| 21 \| Ethanolamine \| 700 \| 344719 \| 174 \| \| 2.5 \| – \| – \| \| 22 \| Isopentadecanoic acid \| 151014 \| 663518 \| 117 \| – \| – \| 2.1 \| \| \| 23 \| Phosphoethanolamine \| 1015 \| 604335 \| 299 \| 2.9 \| \| – \| – \| \| 24 \| Stearic acid \| 5281 \| 787622 \| 117 \| – \| – \| \| 1.5 \| \| Nucleotide-related \| Nucleotide-related \| Nucleotide-related \| Nucleotide-related \| Nucleotide-related \| ↓ \| ↑ \| ↓ \| ↑ \| \| 25 \| 5’-AMP \| 6083 \| 1038688 \| 169 \| – \| – \| 4.5* \| \| \| 26 \| 5’-CMP \| 6131 \| 700635 \| 243 \| – \| – \| \| 1.6 \| \| 27 \| 5′-MTA \| 439176 \| 967036 \| 236 \| – \| – \| 4.0* \| \| \| 28 \| pyrophosphate \| 1023 \| 327517 \| 110 \| – \| – \| 2.1 \| \| \| Pyrimidines \| Pyrimidines \| Pyrimidines \| Pyrimidines \| Pyrimidines \| ↓ \| ↑ \| ↓ \| ↑ \| \| 29 \| Thymidine \| 5789 \| 349402 \| 170 \| – \| – \| 1.4 \| \| \| 30 \| Uracil \| 1174 \| 385735 \| 241 \| – \| – \| 2.1 \| \| \| Other \| Other \| Other \| Other \| Other \| ↓ \| ↑ \| ↓ \| ↑ \| \| 31 \| 2,5-dihydroxypyrazine \| 23368901 \| 397526 \| 241 \| – \| – \| 4.1* \| \| Displayed in Figure 3 are metabolomic maps for the 2P08AA and 50v1 strains, which readily show the changes resulting from cultivation in K30. The metabolomic maps (MetaMapp graphs) in Figure 3 are organized via networks of KEGG reactant pairs (black edges or arrows) and Tanimoto chemical similarity scores (gray edges or arrows). Detected metabolites are presented as nodes (circles). Important changes are highlighted as red nodes (increases in relative abundance), blue nodes (decreases in relative abundance), or yellow nodes (no change). Node sizes signify the relative degree of change (or fold-change). Figure 3: Maps showing the impacts of cultivation in 0.1% v/v K30 on the intracellular and extracellular metabolomes of A. johnsonii 2P08AA (A–C) and A. radioresistens 50v1 (B,D). Metabolites are represented as nodes and organized by structural and metabolic connections (MetaMapp and Cytoscape 3.9.1). Important changes in abundance (p < 0.05) are represented as red (increases), blue (decreases), and yellow (no change) nodes, where node sizes represent the degree of change. Relevant metabolites that show important (black text; p < 0.05) and significant changes (red text; p < 0.05; FDR ≤ 0.20), as listed in Tables 1, 2, are displayed using the associated identity numbers (ID#). Shaded circles represent biochemical classes (red text) that show significant changes (ChemRich); biochemical classes showing no change are listed for comparison purposes (black text). For the 2P08AA strain (Figure 3A), cultivations in $0.1\%$ v/v K30 yield no statistically discernable impacts to the whole cell metabolome. No changes are observed after accounting for multiple testing (FDR ≤ 0.20, $$n = 119$$). No changes are observed when parsing the data using ChemRich. While potential increases ($p \leq 0.05$) are observed in gluconate, analyses in ChemRich provide no support for broad changes in sugar acid content. Despite the potential ($p \leq 0.05$) decreases in arabinose and increases in sorbitol, analyses in ChemRich show no support for broad changes in monosaccharide composition. Similarly, ChemRich analyses provide no support for broad changes in lipid-related metabolites despite the potential ($p \leq 0.05$) decreases in phosphoethanolamine and increases in ethanolamine and behenic acid (22:0; long chain saturated fatty acid). In contrast, the 50v1 strain displays multiple changes in the intracellular metabolome due to cultivation in $0.1\%$ v/v K30 (Table 1). After correction for multiple testing ($$n = 119$$), significant decreases are observed for lysine, homoserine, ornithine, mannitol, adenosine-5′-monophosphate (5’-AMP), 5′-methylthioadenosine (5′-MTA), and 2,5-dihydroxypyrazine. The whole cell metabolomic maps for the 50v1 strain in Figure 3B were parsed using MetaMapp and ChemRich and expand on the univariate analyses to reveal differential changes in amino acids, compatible solutes, nucleotide-related metabolites, dicarboxylic acids, sugar acids, and saturated fatty acids. Impacts to amino acids ($$p \leq 9.1$$×10−6; $Q = 6.7$ × 10−5; ChemRich) are supported by decreases in the relative abundances of standard amino acids (glycine, lysine, proline, serine, and valine), non-standard amino acids (β-alanine, homoserine, ornithine, and citrulline), and a peptide (alanine–alanine). Dimerization of glycine under the analytical conditions is suggested by detection of 2,5-dihydroxypyrazine (Haffenden and Yaylayan, 2005) with the observed decreases in 2,5-dihydroxypyrazine in cultures with K30 being generally consistent with the decreases in glycine. Impacts to compatible solutes ($$p \leq 9.6$$×10−6; $Q = 6.7$ × 10−5; ChemRich) are supported by decreases in mannitol (Empadinhas and da Costa, 2008; Tam et al., 2022), as well as glycerate and 3-phosphoglycerate, which are involved in the biosynthesis of compatible solutes (Franceus et al., 2017). Impacts to nucleotide-related metabolites ($$p \leq 5.2$$×10−4; $Q = 2.4$ × 10−3; ChemRich) are supported by decreases in 5’-AMP, MTA, pyrophosphate, thymidine, and uracil, along with increases in cytidine-5′-monophosphate (5’-CMP). Impacts to dicarboxylic acids ($$p \leq 4.8$$×10−3; $Q = 1.7$ × 10−2; ChemRich) are supported by increases in fumarate, malate, and tartrate. No impacts to saturated fatty acids are observed ($$p \leq 0.11$$; $Q = 0.31$; ChemRich) despite the decreases in isopentadecanoic acid (branched saturated fatty acid; 13-methyl-14:0, or 15:0 iso) and increases in stearic acid (saturated fatty; 18:0). ## Metabolic considerations Cross referencing of the metabolomic changes to the KEGG database yields probable impacts to several intracellular pathways for the 50v1 strain. Provided in parentheses are the associated KEGG pathway or reaction identifiers. Inhibition or reduction in Glycine, Serine, and Threonine Metabolism (map00260) is supported by decreases in intracellular glycerate, 3-phosphoglycerate, serine, glycine, and homoserine. Inhibition or reduction in Lysine Biosynthesis (map00300) and Valine, Leucine, and Isoleucine Biosynthesis (map00290) is supported by decreases in lysine and valine; however, no other changes are noted in the respective pathways. Inhibition or reduction in proline and arginine biosynthesis through Arginine and Proline Metabolism (map00330) and Arginine Biosynthesis (map00220) are supported by decreases in ornithine, citrulline, and proline. The lack of changes in abundances for urea indicate no measurable impact on the flux of the urea cycle – despite the decreases in ornithine and citrulline. The lack of observed arginine is suggestive of conversion of arginine to ornithine under the analytical conditions (Halket et al., 2005) or the limited presence of free arginine in the cell. Inhibition or reduction in Peptidoglycan Biosynthesis (map00550) and/or peptide (protein) synthesis is suggested by decreases in alanine–alanine. Inhibition or reduction in β-alanine metabolism (map00410) is supported by decreases in β-alanine, which in turn is suggestive of reductions in arginine and uracil degradation. Differential changes in fatty acid metabolism (map01212) are suggested by the decreases in free isopentadecanoic acid (Reaction R02663) and increases in free stearic acid. Inhibition or altered flux through glycolysis (map00010) is suggested through decreases in 3-phosphoglycerate. Activation or altered flux through components of the TCA cycle are suggested by increases in fumarate and malate. Inhibition or reduced flux of selected monosaccharides through Fructose and Mannose Metabolism (map00051) and Pentose and Glucuronate Interconversions (map00040) are, respectively, suggested by decreases in mannitol and arabinose. Differential changes in Nucleotide Metabolism (map01232) are suggested by increases in 5’-CMP and 5′-MTA, along with decreases in 5’-AMP, pyrophosphate, thymidine, and uracil. Additionally, inhibition or reduction in Pyrimidine Metabolism (map00240) is suggested decreases by uracil and thymidine. ## Impacts of Kleenol 30 on the extracellular metabolomes Displayed in Figure 3 are biochemical maps that compare the extracellular metabolomes from A. johnsonii 2P08AA and A. radioresistens 50v1 (early stationary phase). Per strain, dried cultivation broth (cell free) yielded a total of 56 metabolites, which were processed and visualized as described above. Listed in Table 2 are the extracellular metabolomic changes in the 2P08AA and 50v1 strains after cultivation in the absence and presence $0.1\%$ v/v K30. Pathway analyses were not conducted for extracellular metabolites. Table 2 \| Extracellular metabolites \| Extracellular metabolites.1 \| Extracellular metabolites.2 \| Extracellular metabolites.3 \| Extracellular metabolites.4 \| 2P08AA \| 2P08AA.1 \| 50v1 \| 50v1.1 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| ID# \| Biochemical class \| CID \| RI \| m/z \| Fold-change \| Fold-change \| Fold-change \| Fold-change \| \| Sugar acids \| Sugar acids \| Sugar acids \| Sugar acids \| Sugar acids \| ↓ \| ↑ \| ↓ \| ↑ \| \| 32 \| Galactonic acid \| 128869 \| 690882 \| 292 \| – \| – \| \| 36* \| \| 14 \| Gluconic acid \| 6857417 \| 693148 \| 333 \| – \| – \| \| 46* \| \| 33 \| Gluconolactone \| 7027 \| 645815 \| 220 \| \| 22 \| \| 8.8* \| \| 16 \| Glyceric acid \| 439194 \| 377495 \| 189 \| \| 34 \| \| 9.3* \| \| 34 \| 2-Methylgyceric acid \| 560781 \| 372491 \| 219 \| \| \| \| 1.9 \| \| Monosaccharides \| Monosaccharides \| Monosaccharides \| Monosaccharides \| Monosaccharides \| ↓ \| ↑ \| ↓ \| ↑ \| \| 35 \| Fructose \| 439709 \| 639442 \| 307 \| \| 4.5 \| – \| – \| \| 36 \| Galactose \| 439357 \| 647344 \| 319 \| \| 22 \| – \| – \| \| 37 \| Mannose \| 18950 \| 645856 \| 205 \| \| 57 \| \| 20* \| \| 38 \| Ribose \| 5779 \| 553078 \| 217 \| 2.5 \| \| 1.9* \| \| \| Di and monocarboxylic acids \| Di and monocarboxylic acids \| Di and monocarboxylic acids \| Di and monocarboxylic acids \| Di and monocarboxylic acids \| ↓ \| ↑ \| ↓ \| ↑ \| \| 39 \| Adipic acid \| 196 \| 474435 \| 111 \| 1.9 \| \| – \| – \| \| 13 \| Tartaric acid \| 444305 \| 534291 \| 292 \| – \| – \| \| 2.9* \| \| 40 \| Citramalic acid \| 1081 \| 456203 \| 247 \| – \| – \| 1.6 \| \| \| 41 \| 3,4-Dihydroxybenzoic acid \| 72 \| 620200 \| 193 \| 2.0 \| \| – \| – \| \| 42 \| 4-Hydroxybutanoic acid \| 10413 \| 325027 \| 233 \| – \| – \| 2.4* \| \| \| 43 \| α-ketoglutarate \| 51 \| 507392 \| 198 \| – \| – \| 9.0* \| \| \| 44 \| Oxalic acid \| 971 \| 260513 \| 190 \| \| 5 \| – \| – \| \| Fatty acids \| Fatty acids \| Fatty acids \| Fatty acids \| Fatty acids \| ↓ \| ↑ \| ↓ \| ↑ \| \| 45 \| Pelargonic acid \| 8158 \| 399229 \| 117 \| 3.1 \| \| – \| – \| When grouped by biochemical classes, cultivations of the 2P08AA strain in K30 yield differential impacts to extracellular sugar acids, monosaccharides, and organic acids (di- and monocarboxylic acids). Yet, univariate tests show no changes after correction for multiple testing. Impacts to sugar acids ($$p \leq 1.4$$×10−5; $Q = 6.9$×10−5; ChemRich) are supported by substantial increases in extracellular gluconolactone (22-fold) and glycerate (34-fold). Impacts to di- and monocarboxylic acids, which were grouped as organic acids ($$p \leq 3.0$$×10−4; $Q = 7.6$×10−4; ChemRich), are supported by increases in oxalate and decreases in adipic acid (hexanedioic acid) and 3,4-dihydroxybenzoate (protocatechuate). Impacts to extracellular monosaccharides ($$p \leq 7.6$$×10−3; $Q = 1.3$×10−2; ChemRich) are supported by substantial increases in extracellular fructose (4.5-fold), galactose (22-fold), and mannose (57-fold) and decrease (2.5-fold) in extracellular ribose (pentose). In contrast, cultivations of the 50v1 strain in K30 – after correction for multiple testing ($$n = 56$$) – yield significant changes in the extracellular galactonate, gluconate, gluconolactone, glycerate, mannose, ribose, tartrate, 4-hydroxybutyrate, and α-ketoglutarate. Likewise, grouping by biochemical classes are reveals significant changes in the extracellular metabolome. Impacts to sugar acids ($$p \leq 1.9$$×10−8; $Q = 9.6$×10−8; ChemRich) are supported by substantial increases in extracellular galactonate (36-fold), gluconate (46-fold), gluconolactone (8.8-fold), glycerate (9.3-fold), and 2-methylglycerate (1.9-fold). Impacts to extracellular organic acids ($$p \leq 7.8$$×10−2; $Q = 0.16$; ChemRich) are suggested by the significant increases in tartrate and significant decreases in α-ketoglutatate, along with the potential decreases in citramalate and 4-hydroxybutyrate. Impacts to extracellular monosaccharides ($$p \leq 9.6$$×10−2; $Q = 0.16$; ChemRich) are suggested by significant increases in extracellular mannose (20-fold) and significant decreases in ribose (2.5-fold). ## Canonical correlation analyses In Figure 4, we use canonical correlation analyses (CCA) to, respectively, compare metabolomes from A. johnsonii 2P08AA ($$n = 3$$) and A. radioresistens 50v1 ($$n = 3$$) to differing quantitative descriptors characterizing changes in the cultivation media and growth kinetics in the absence and presence of $0.1\%$ v/v K30. Correlations for the intracellular (or whole cell) metabolomes from late log-phase cultures were assessed against growth rates at exponential phase, plate counts at late-log phase, maximum relative biomass at stationary phase, and total detergent concentrations in the media, which were assumed to be $0.025\%$ w/w in cultures containing $0.1\%$ v/v K30. Figure 4: Canonical correspondence analyses (CCA) that compare the impacts of cultivation in 0.1% v/v Kleenol 30 (K30) on (A) the intracellular metabolome of A. johnsonii 2P08AA, (B) the extracellular metabolome of A. johnsonii 2P08AA, (C) the intracellular metabolome of A. radioresistens 50v1, and (D) the extracellular metabolome of A. radioresistens 50v1 against the quantitative descriptors of growth rates at exponential phase (blue callout boxes, green vectors), plate counts at late-log phase (purple callout boxes, green vectors), maximum relative biomass at stationary phase (brown callout boxes, green vectors), and either the total concentrations of detergents (0.025% w/w; assumed) or chelators (0.02% w/w; assumed) in the cultivation media (black callout boxes, green vectors); dimension reduction of the triplicate metabolomic measures associated with 0% v/v K30 (red squares, pink circles, and red and pink diamonds) and 0.1% v/v K30 (black triangles) are provided, reduced terms are highlighted by the green and red circles and callout boxes, and metabolites are arrayed as text. The CCA plots include the total array of detected metabolites (text), listed quantitative descriptors (vector lines), and transformed data after dimension reduction (glyphs). For the 2P08AA and 50v1 strains, stronger correlations along CCA Axis 1 (61.70, $69.46\%$) are observed when compared to CCA Axis 2 (24.72, $23.70\%$), respectively. Dimension reduction for the metabolomes from the 2P08AA and 50v1 strains (Figures 4A,B) respectively show clear separations between samples prepared in the absence and presence of $0.1\%$ v/v K30. These trends suggest that the intracellular metabolomic compositions of both strains are impacted by K30 during cultivation. Described in the following sub-sections are comparisons to the quantitative descriptors. For selected metabolites, additional clarifications and/or fatty acid abbreviations are provided in parentheses. In Figure 4, we use canonical correlation analyses (CCA) to, respectively, compare the extracellular metabolomes from A. johnsonii 2P08AA ($$n = 3$$) and A. radioresistens 50v1 ($$n = 3$$) from late log-phase cultures to the quantitative descriptors of growth rates at exponential phase, plate counts at late-log phase, maximum relative biomass at stationary phase, and total concentration of EDTA and sodium metasilicate in the media, which was assumed to be $0.02\%$ w/w in cultures containing $0.1\%$ v/v K30. The CCA plots in Figures 4C,D, which compare results from 0 and $0.1\%$ v/v K30, include the total array of detected metabolites (text), listed quantitative descriptors (lines), and transformed data after dimension reduction (glyphs). For the 2P08AA and 50v1 strains, stronger correlations along CCA Axis 1 (67.18, $82.53\%$) are observed when compared to CCA Axis 2 (32.81, $11.10\%$), respectively. Dimension reduction for the metabolomes from the 2P08AA and 50v1 strains (Figures 4C,D) respectively show clear separations between samples prepared in the absence and presence of $0.1\%$ v/v K30. These trends suggest that the extracellular metabolomic compositions of both strains are impacted by K30 during cultivation. Described in the following sub-sections are comparisons to the quantitative descriptors. For selected metabolites, additional explanations and/or fatty acid abbreviations are provided in parentheses. ## Comparisons to survival When considering changes in survival (Figure 4A), the vector describing plate counts for the 2P08AA strain (late log phase) overlaps with the reduced terms associated with $0\%$ v/v K30 (filled red squares). This correlation suggests that survival trends for the 2P08AA strain (across 0 to $0.1\%$ v/v K30) are associated with limited changes in the intracellular metabolome. This assessment is consistent with the associated univariate tests which indicate no changes in plate counts ($p \leq 0.05$) or metabolite abundances for the 2P08AA strain in $0.1\%$ v/v K30 (FDR > 0.20). From the CCA plot for the 2P08AA strain (Figure 4A), metabolites trending alongside the plate counts include trehalose, trehalose-6-phosphate, fructose-6-phosphate, and potentially glucose-6-phosphate. Univariate tests infer no changes in the abundances for these metabolites ($p \leq 0.05$ or FDR > 0.20). These trends suggest that the quantitative survival for the 2P08AA strain (in part) relates to lack of changes in central metabolism and compatible solute formation. For the 50v1 strain (Figure 4B), the vector describing plate counts overlaps with the reduced terms associated with $0\%$ v/v K30 (pink circles). This correlation suggests that survival trends for the 50v1 strain (across 0 to $0.1\%$ v/v K30) are associated with limited changes in the intracellular metabolome. This assessment is contrary to the associated univariate tests that support ~$50\%$ reductions in survival ($p \leq 0.05$) and substantial changes in the metabolome ($p \leq 0.05$, FDR ≤ 0.20). From the CCA plot for the 50v1 strain (Figure 4B), metabolites trending alongside the plate counts include 4-hydroxybenzoate, several amino acids (similar to those in Table 1), ribose-5-phosphate, N-acetylglucosamine, nicotinic acid, phenylacetic acid, bisphosphoglyercol, and potentially 5’-AMP and 5′-MTA. These trends suggest that survival of the 50v1 strain is related to changes in central metabolism, amino acid metabolism, nucleotide metabolism, and cell wall metabolism (e.g., metabolism of glycosoaminoglycans and peptidoglycans, which are based on N-acetylglucosamine). Trends from the CCA plot for the 50v1 strain (Figure 4B) are additionally suggestive of changes in native benzene metabolism (e.g., phenylacetic acid and 4-hydroxybenzoate). However, univariate analyses yield no support ($p \leq 0.05$) for the apparent decreases in phenylacetic acid (~0.1-fold decreases) and 4-hydroxybenzoate (~0.9-fold decreases). Alternatively, described further in Section 3.5.3, phenylacetic acid and 4-hydroxybenzoate are potential biodegraded products of K30. Hence, trends for the 50v1 strain hint at a potential for biodegradation of K30 through benzene metabolism. ## Comparisons to maximum relative biomass When considering biomass, the vector describing the maximum relative biomass at stationary phase for the 2P08AA strain (Figure 4A) overlaps with the reduced terms associated with $0.1\%$ v/v K30 (black triangles). This correlation suggests that trends in the maximum relative biomass for the 2P08AA strain (across 0 to $0.1\%$ v/v K30) are associated with changes in the intracellular metabolome. This assessment is consistent with the associated univariate tests for the cultivation data that support increases in the maximum relative biomass ($p \leq 0.05$), yet counter to the associated univariate tests for the metabolomic data that indicate no change in the relative metabolite abundances after correction for multiple testing (FDR > 0.20). From the CCA plot for the 2P08AA strain (Figure 4A), the metabolites trending alongside the maximum relative biomass include methionine, lignoceric acid (fatty acid; 23:0), oxoproline, and glucose-1-phosphate. Univariate tests indicate no change in the relative abundances for these metabolites (FDR > 0.20). These combined trends suggest that increases in maximum relative biomass relate to minimal or no changes in glucose metabolism, amino acid metabolism, and unsaturated fatty acid abundances. For the 50v1 strain (Figure 4B), the vector describing the maximum relative biomass overlaps with the reduced terms associated with $0.1\%$ v/v K30 (black triangles). This correlation suggests that trends in the maximum relative biomass for the 50v1 strain (across 0 to $0.1\%$ v/v K30) are associated with changes in the intracellular metabolome. This assessment is consistent with the associated univariate tests on the cultivation data ($p \leq 0.05$) and metabolomic data ($p \leq 0.05$, FDR ≤ 0.20), which yield support for increases in the maximum relative biomass and differential changes in metabolite abundances. In contrast to the 2P08AA strain, these trends for 50v1 strain suggest that increases in the maximum relative biomass are associated with several intracellular metabolomic changes. From the CCA plot for the 50v1 strain (Figure 4B), metabolites trending alongside the increases in maximum relative biomass include lignoceric acid (fatty acid; 23:0), tryptophan, xanthine, stearic acid (fatty acid; 18:0), α-ketoglutarate, phosphate, 4-hydroxybutanoate (4-hydroxybutyric acid), 4-aminobutanoate (4-aminobutyric acid), O-phosphoserine, 1-monopalmitin (1-palmitoylglycerol; palmitoyl = 16:0), behenic acid (fatty acid; 22:0), and 2-monoolein (2-oleoylglycerol; oleoyl = 18:1Δ9). These trends suggest that increases in the maximum relative biomass for the 50v1 strain are related (at the minimum) to changes in butanoate metabolism (e.g., 4-hydroxybutanoate and 4-aminobutanoate), amino acid metabolism (e.g., tryptophan, α-ketoglutarate, and O-phosphoserine) and phosphate metabolism (phosphate and O-phosphoserine). The trends for the 50v1 strain (Figure 4B) also suggest a relation to changes in free fatty acids and monoacylglycerols; where the associated metabolites from the CCA plot (and respective univariate analyses in parentheses) include lignoceric acid (~1.2-fold increases; $p \leq 0.05$), stearic acid (~1.5-fold increases; $p \leq 0.05$), 1-monopalmitin (~2.2-fold increases; $p \leq 0.05$), behenic acid (~10-fold increases; $p \leq 0.05$), and 2-monoolein (~32-fold increases; $p \leq 0.05$). However, in comparison, ChemRich analyses yield no support for changes in the intracellular fatty acids. Further, trends from the CCA plot for the 50v1 strain (Figure 4B) support a correlation between the maximum relative biomass and metabolism of polyhydroxyalkanoates (PHAs), which are intracellular polymers (e.g., poly-4-hydroxybutanoate and/or other co-polymers) used for carbon and energy storage (Khanna and Srivastava, 2005). Among the metabolites trending alongside the maximum relative biomass are α-ketoglutarate, glutamate, 4-aminobutanoate, and 4-hydroxybutanoate. These combined metabolites represent a stepwise pathway toward the synthesis of PHAs (KEGG map 00250 and map00650), where oxidized carbons are acquired through the TCA cycle (α-ketoglutarate) and sequentially shuttled through amino acid metabolism (glutamate) and butanoate metabolism (4-aminobutanoate) to yield a monomeric unit (4-hydroxybutanoate) commonly found in PHAs. Together, these observations are relevant since the accumulation of PHAs can result in increases in relative optical densities (Slaninova et al., 2018), which is consistent with our growth curves and ensuing regression analyses that yield increases in the maximum relative biomass in the presence of $0.1\%$ v/v K30. Hence, for the 50v1 strain, the increases in maximum relative biomass may relate to the increased synthesis of PHAs under the stresses imposed by K30. ## Comparisons to growth rates When considering growth rates (exponential phase), the vector for the 2P08AA strain (Figure 4A) overlaps with the reduced terms associated with $0\%$ v/v K30 (red squares). This correlation suggests that growth rates trends for the 2P08AA strain (across 0 to $0.1\%$ v/v K30) are associated with limited changes in the intracellular metabolome. This assessment is consistent with the associated univariate tests for growth rates that support moderate decreases of ~$18\%$ ($p \leq 0.05$) and the associated univariate tests for the metabolomes that imply no change in relative abundances (FDR > 0.20). From the CCA plot of the 2P08AA strain, metabolites trending alongside the growth rates include palmitoleic acid (fatty acid; 16:1Δ9), 5′-MTA, ribose, several amino acids, oleic acid, N-acetylglucosamine, and 5’-AMP. These trends suggest that the decreases in growth rates are related to changes in amino acid metabolism, cell wall metabolism, and nucleotide metabolism. The trends also suggest relations to changes in free unsaturated fatty acid content through associations with palmitoleic acid (~0.7-fold decrease; $p \leq 0.05$) and oleic acid (~0.3-fold decrease; $p \leq 0.05$). However, ChemRich analyses yield no support for changes in unsaturated fatty acid metabolism. For the 50v1 strain (Figure 4B), in contrast, the vector describing growth rates overlaps more closely with the reduced terms associated with $0.1\%$ v/v K30 (black triangles). This correlation suggests that growth rate trends for the 50v1 strain (across 0 to $0.1\%$ v/v K30) are associated with changes to the intracellular metabolome. In comparison, univariate tests on the cultivation data indicate no change in the growth rates ($p \leq 0.05$), while univariate tests on the metabolomic data support changes in the abundances for multiple metabolites ($p \leq 0.05$, FDR ≤ 0.20). These combined trends are suggestive of substantial post-exponential phase changes in the metabolomes, as growth rates for the 50v1 strain during exponential phase are unaltered by – and potentially not associated with – the multiple metabolomic changes observed during late-log phase. ## Comparisons to detergent concentrations When considering the K30 formulation (Figures 4A,B), the vectors describing detergent concentrations (~0–$0.025\%$ w/w) overlap with the reduced terms associated with $0.1\%$ v/v K30 for both the 2P08AA and 50v1 strains (black triangles). These correlations suggest the metabolomes from both strains adjust in response to detergents in $0.1\%$ v/v K30. We note the trends for detergent concentrations likely overlap with the maximum relative biomass term, which plots across the same respective quadrant. For the 2P08AA strain, metabolites trending alongside the detergent concentrations include myristic acid (fatty acid; 14:0), pelargonic acid (fatty acid; 9:0), 4-hydroxybenzoate, phenylacetic acid, and behenic acid (fatty acid; 22:0). The trends for 4-hydroxybenzoate and phenylacetic acid lend support toward the potential for biodegradation of dodecyl benzene sulfonate and polyethylene glycol mono-nonylphenyl ether, the detergents from the K30 formulation. As described in Hashim et al. [ 1992], phenylacetic acid is a potential product of microbial degradation of dodecyl benzene sulfonate through oxidation (or metabolism) of the dodecyl group (to yield 4-sulfonylphenylacetic acid), followed by desulphonization (to yield phenylacetic acid). Similarly, we posit that biodegradation of polyethylene glycol mono-nonylphenyl ether could yield 4-hydroxybenzoate through scission of the polyethylene glycol units (to yield 4-nonylphenol), as described in Section 1 followed by oxidation or metabolism of the nonyl side group (to yield 4-hydroxybenzoate). In comparison, however, univariate tests yield no support for changes in the abundances for phenylacetic acid (~2.4-fold increases; $$p \leq 0.39$$) and 4-hydroxybenzoate (~1.5-fold increases; $$p \leq 0.11$$). The trends for the 2P08AA strain are also suggestive of correlations between detergent concentrations and changes in free fatty acid content. Relevant metabolites from the CCA plot (and the associated univariate analyses in parentheses) include behenic acid (~15-fold increases; $p \leq 0.05$), myristic acid (~1.2-fold increases; $p \leq 0.05$), and pelargonic acid (~1.4-fold increases; $p \leq 0.05$). We note that behenic acid levels in fluvial biofilms increase after exposure to desiccation (Serra-Compte et al., 2018). We also speculate the increases in long chain saturated fatty acid content may promote aggregation with the detergents in the intracellular and/or membrane space, which could potentially serve as a survival mechanism and/or acquisition strategy during biodegradation. Analyses by ChemRich, however, yield no support for changes in fatty acid or lipid-related metabolites. For the 50v1 strain (Figure 4B), metabolites that trend alongside detergent concentrations include tartrate, 1-monoolein (1-oleoylglycerol; oleoyl = 18:1Δ9), and octadecanol (fatty alcohol). These trends are suggestive of increases in an intracellular dicarboxylic acid (tartrate) and free lipid-related molecules (1-monoolein and octadecanol), which may, respectively, relate to metal retention in the cell and the aggregation of detergents in the intracellular or membrane space. ## Comparisons to survival and growth rates When considering survival and cultivations, the vectors for plate counts and growth rates for the 2P08AA strain (Figure 4C) overlap with the reduced terms associated with $0\%$ v/v K30 (red diamonds). These correlations suggest that survival and growth rate trends for the 2P08AA strain (across 0 to $0.1\%$ v/v K30) are associated with minimal changes in the extracellular metabolome. These assessments are roughly consistent with the associated univariate analyses that indicate quantitative survivals, ~$18\%$ decreases in growth rate, and a lack of changes in the extracellular metabolome (after correction for multiple testing). From the CCA plot, metabolites from the 2P08AA strain that trend alongside the plate counts and growth rates include 2-isopropylmalic acid, pyrophosphate, phosphate, arachidic acid (fatty acid; 20:0), lauric acid (fatty acid; 12:0), capric acid (fatty acid; 10:0), 4-hydroxyphenyllactic acid, sorbitol, and isopentadecanoic acid (13-methyl-14:0). These trends for the 2P08AA strain suggest that survival (which was not impacted) and growth rates (which minimally decreased) relate to a lack of change in the abundances of extracellular saturated fatty acids, phosphate, pyrophosphate, and α-hydroxy acids (2-isopropylmalic acid and 4-hydroxyphenyllactic acid). However, the potential increases in extracellular sorbitol (~5-fold increase, $p \leq 0.05$; Table 1) are suggestive of efflux of a compatible solute. For the 50v1 strain (Figure 4D), in contrast, the vector describing plate counts overlaps with the reduced terms associated with $0\%$ v/v K30 (pink diamonds), while the vector describing growth rates moderately overlaps with the reduced terms associated with $0.1\%$ v/v K30 (black triangles). These opposing correlations suggest that survival trends (across 0 to $0.1\%$ v/v K30) are associated with minimal changes in the extracellular metabolome, while growth rate trends (across 0 to $0.1\%$ v/v K30) are associated with more notable changes in the extracellular metabolome. Metabolites trending alongside the plate counts for the 50v1 strain include 2-isopropylmalic acid, 2-hydroxyglutarate, succinic acid, and α-ketoglutarate. These results suggest that survival for the 50v1 strain is related to changes in extracellular α-hydroxy acids (2-isopropylmalic acid, 2-hydroxyglutarate, and α-ketoglutarate) and a dicarboxylic acid (succinate); while univariate tests support ~9-fold decreases in extracellular α-ketoglutarate. Metabolites trending alongside the growth rates for the 50v1 strain include tagatose and N-acetylglucosamine, which suggests that changes in extracellular monosaccharides relate to maintaining stable growth rates. ## Comparisons to maximum relative biomass and chelator concentrations When considering changes in the maximum relative biomass (stationary phase) and chelator concentrations (EDTA and sodium metasilicate from the K30 formulation), similar trends are observed for the 2P08AA and 50v1 strains (Figures 4C,D). The vectors describing chelator concentrations for both strains overlap with the reduced terms associated with $0.1\%$ v/v K30 (black triangles). These correlations suggest the extracellular metabolomes from both strains adjust in response to the change in chelator concentration across 0 to $0.1\%$ v/v K30 (e.g., the metal sequestering agents of EDTA and sodium metasilicate). From the CCA plot for the 2P08AA strain, metabolites trending alongside changes in the maximum relative biomass and/or chelator concentration include threonate, fructose, oxalate, galactonate, gluconate, gluconolactone, mannose, galactose, and glycerate. These trends suggest that chelator concentrations influence the abundances of extracellular monosaccharides and sugar acids. While these trends are consistent with univariate tests ($p \leq 0.05$), no support for changes in the abundances of the associated metabolites (Table 2) are obtained after correction for multiple testing (FDR ≤ 0.20). For the 50v1 strain, metabolites trending alongside changes in the maximum relative biomass and/or chelator concentrations include N-acetylglucosamine, tartrate, 2-methylglycerate, gluconolactone, glycerate, mannose galactonate, and gluconate. These results suggest that the chelator concentrations influence the abundances of extracellular monosaccharides, modified monosaccharides, and sugar acids. These assessments generally are consistent with univariate analyses. ## Spacecraft-associated Acinetobacter Molecular insights into the tolerance of spacecraft-associated Acinetobacter toward Kleenol 30, a NASA cleanroom floor detergent, were obtained through cultivation, kinetic, and metabolomic approaches. Comparisons were conducted using A. johnsonii 2P08AA and A. radioresistens 50v1, which were isolated at, respectively, unique locations within the spacecraft assembly facilities, which were subjected to differing cleaning regimes. The strain A. johnsonii 2P08AA was isolated from the floor of the Mars Phoenix lander assembly facility, which was routinely cleansed with Kleenol 30. The strain A. radioresistens 50v1 was isolated from the surface of the pre-flight Mars Odyssey orbiter, which was likely cleaned with isopropyl alcohol (2-propanol, isopropanol, IPA) and never exposed to Kleenol 30. Hence, comparisons across these differing sub-environments provide initial insights into the impacts of repeated cleansing, under the conditions of spacecraft assembly, toward biochemical and cultivation selection. ## Survival of the spacecraft-associated Acinetobacter against Kleenol 30 Cultivations containing 0–$2.0\%$ v/v Kleenol 30 (K30) were conducted in a low-osmolarity minimal media (0.2x M9, 25 μM Fe) containing 150 mM ethanol, which served as the sole supplied carbon source. Cultivation media containing $2.0\%$ v/v K30 were bactericidal for the tested Acinetobacter, which suggests a biocide-type of activity for K30. In contrast, in $1.0\%$ v/v K30, appreciable cultivation tolerances were observed for the 2P08AA and 50v1 strains. Cultures containing $1.0\%$ v/v K30 exhibited [1] ~20–$30\%$ survivals, [2] ~66 and $12\%$ decreases in growth rates, and [3] eliminated lag phase times for the 2P08AA and 50v1 strains, respectively. These cultivation tolerances of <$2\%$ v/v K30 are significant since current NASA planetary protection practices utilize ~0.8-$1.6\%$ v/v K30 for floor cleansing. Our results, therefore, suggest that the spacecraft-associated Acinetobacter harbor a tolerance toward the current floor cleansing practices for NASA cleanrooms. In $0.1\%$ v/v K30, differential impacts were observed across the cultivation parameters. For the 2P08AA strain, cultures exhibited [1] no measurable changes in cell density at late-log phase (plate counts), [2] modest decreases in the growth rate (~$18\%$), and [3] negligible times in the lag phase. These results indicate that K30 impacts the 2P08AA strain by accelerating growth out of the lag phase, slightly inhibiting growth during the exponential phase, and by yielding no changes cell density after cultivation to late-log phase. For the 50v1 strain, cultures exhibited [1] ~$50\%$ survivals, [2] no changes in growth rate, and [3] ~$70\%$ decreases in the lag phase time. This is indicative of K30 impacting the 50v1 strain by mildly accelerating growth out of the lag phase, yielding no change in the growth rates during the exponential phase, and inhibiting cell density by ~$50\%$ when cultivated to late-log phase. The differential cultivation trends observed in $0.1\%$ v/v K30 are indicative of strain and growth-phase dependent responses to K30. In turn, these trends suggest that K30 reduces nutrient bioavailability in culture (e.g., alkaline earth and transition metals) and/or elicits specific biochemical responses (e.g., metabolism of K30 and/or enzymatic inhibition by K30) – which together manifest as acceleration out of the lag phase for both strains, slight inhibition in rates of cell division during the exponential phase for the 2P08AA strain, and moderate reductions in total cell density at late-log phase for the 50v1 strain. ## Metabolomic profiling of the cultivation tolerances against Kleenol 30 To obtain metabolic insights into the tolerances toward Kleenol 30 (K30), we used cultures grown in $0.1\%$ v/v K30 to obtain reliable biomasses at late-log phase for both strains (see comparison of 0.1 and $1.0\%$ w/w K30 in Figures 1D). For A. johnsonii 2P08AA, cultivations with $0.1\%$ v/v K30 exhibited [1] no significant and discreet changes in the abundances of intracellular metabolites, [2] several significant extracellular changes when considering broader classifications such as sugar acids, monosaccharides, organic acids (di- and monocarboxylic acids), and fatty acids, and [3] potential support for biodegradation and metabolism of both detergents in the K30 formulation. The combined results for the 2P08AA strain suggest that the native and intracellular metabolic state of the 2P08AA strain readily accommodates cultivation in the presence of $0.1\%$ v/v K30. The trends for the 2P08AA strain also suggest that survival is associated with adjustments in the extracellular metabolome and a potential for biodegradation of the Kleenol 30 detergents. For A. radioresistens 50v1, cultivations in $0.1\%$ v/v K30 exhibited [1] differential changes in the abundances of intracellular amino acids, compatible solutes, nucleotide-related metabolites, dicarboxylic acids, and saturated fatty acids, [2] substantial and differential changes in the abundances of extracellular sugar acids, monosaccharides, and organic acids, and [3] no clear statistical support for biodegradation and metabolism of the detergents from the K30 formulation (when excluding the potential for degradation via benzene metabolism). These results for the 50v1 strain indicate that the intracellular and extracellular metabolomes of the 50v1 strain significantly adjust to accommodate cultivation in the presence of $0.1\%$ v/v K30. These trends also suggest that survival is associated with intracellular changes to amino acid and peptide metabolism, nucleotide metabolism, central metabolism, fatty acid metabolism, and pyrimidine metabolism. For the 50v1 strain, the changes in amino acid metabolism include decreases in lysine, proline, glycine, ornithine, homoserine, valine and others listed in Table 1. This is relevant since decreases in lysine and valine are observed during biofilm formation in A. baumanni 1,656–2 (Yeom et al., 2013), while decreases in amino acid metabolism are observed after treatments of A. baumanni AB5075 with mixed antibiotics (polymyxin B and rifampicin) (Zhao et al., 2021). In E.coli UTI189, decreases in intracellular lysine and valine are associated with biofilm formation (Lu et al., 2019). In Bifodobacterium bifudum, decreases in intracellular lysine, arginine, and proline are associated with biofilm formation (Sadiq et al., 2020). These observations suggest that the Acinetobacter undergo stress-like responses toward K30 that inhibit total cell mass (e.g., survival), which could include changes in early biofilm development. Trends across nucleotide abundances and the cultivation terms are similarly suggestive of early biofilm-type development (Sarkar and Chakraborty, 2008; Castillo-Juarez et al., 2017) and/or quorum sensing related changes such as intracellular PHA synthesis (Kessler and Witholt, 2001; Xu et al., 2011). These trends include increases in the maximum relative biomass terms for the 2P08AA and 50v1 strains, decreases in 5′-MTA and 5’-AMP for the 50v1 strain, and substantial increases in extracellular monosaccharides (e.g., mannose, galactose, and fructose for the 2P08AA strain). In comparison, biofilms of E. coli UTI189 show decreases in 5′-MTA and 5’-CMP, as well increases in metal-binding siderophores, when Fe3+ concentrations in culture decrease from 10 to 1 μM (Guo et al., 2021). Lowered 5′-MTA abundances are associated with biofilms of Pseudomonas aeruginosa PAO1 (Tielen et al., 2013). In Pseudomonas fluorescens PF08, exogenously added 5’-AMP inhibits biofilm formation (Wang et al., 2021). In Streptococcus pyogenes, the import of mannose induces lysozyme resistance via a quorum sensing pathway (Chang et al., 2015); while exogenously added galactose differentially inhibits and activates biofilm development in differing species of Streptococcus (Ryu et al., 2016). ## Biochemical strategies against Kleenol 30 To obtain biochemical insights into the mechanisms of tolerance, the metabolomic trends were interpreted in context of the cultivation conditions. In this study, cultivations were conducted in low-osmolarity media (0.2x M9) supplemented with 25 μM Fe2+ (Fe(NH4)2(SO4)2), which slowly oxidized to Fe3+ under the aerobic cultivations in the capped glass tubes (per absorbance spectroscopy). However, in control experiments, cultures supplemented with [Fe(EDTA)]1− grew slower, when compared to Fe(NH4)2(SO4)2, while cultivations in acid-washed glassware yielded no growth. Therefore, under the cultivation conditions, trace elements and species such as Zn2+, Co2+, Cu2+, Mn2+, BO33− (borate; a B source), and MoO42− (molybdate; a Mo source) were likely obtained from trace contaminants in the media and/or the glassware. These details are relevant since K30 contains EDTA and sodium metasilicate, which are chelators that sequester transition and alkaline earth metals. Hence, in culture, chelation by EDTA would likely restrict the bioavailability of transition metals and inhibit growth – as observed in control studies. Limited bioavailabilities for Ca2+ and Mg2+ are also expected since sodium metasilicate (Na2SiO3) is a water softening agent that functions through ion exchange via chelation of Ca2+ or Mg2+ (and release of Na+). Under these considerations, therefore, our review of the extracellular metabolomes reveals changes in several biochemicals possessing metal-binding properties. Appraisal of the literature shows that sugar acids such as galactonate, gluconate, and gluconolactone form appreciably stable complexes with transition metals such as Fe3+, Zn2+, Co2+, Cu2+, and Mn2+ (Sawyer, 1964; Carper et al., 1989; Frutos et al., 1998; Gyurcsik and Nagy, 2000). For the alkaline earth metals of Ca2+ and Mg2+, stable complexes are formed with sugar acids such as gluconolactone and glycerate (Taga et al., 1978; Meehan et al., 1979; Tajmir-Riahi, 1990), monosaccharides such as mannose (Angyal, 1973), and organic acids such as oxalate (Chutipongtanate et al., 2013; Izatulina et al., 2018) and tartrate (Hawthorne et al., 1982). In addition, borate and Ca2+ form stable complexes with gluconate, glycerate, and tartrate (van Duin et al., 1987); while molybdate yields stable complexes with mannose and fructose (Spence and Kiang, 1963; Sauvage et al., 1996). Biochemically, therefore, these observations suggest that the 2P08AA strain targets trace Ca2+ and Mg2+ acquisition through potential increases in extracellular glycerate, gluconolactone, and oxalate, but not via gluconate and galactonate, which bind transition and alkaline earth metals but were not detected. In contrast, the 50v1 strain potentially targets both transition and alkaline earth metals given the increases ($p \leq 0.05$, FDR ≤ 0.20) in extracellular gluconate, galactonate, glycerate, gluconolactone, and tartrate. In fact, increases in glycerate export for the 50v1 strain are supported by the increases in extracellular glycerate and decreases in intracellular glycerate. Control of intracellular Ca2+ retention is also suggested by increases in extra- and intracellular tartrate. Multivariate tests support these total trends and suggest that chelator concentrations (from the K30 formulation) influence the abundances of extracellular metal-binding biochemicals. Thus, our total results suggest that up-regulations (per se) in extracellular trace element acquisition strategies are common survival features to the spacecraft-associated Acinetobacter. Future studies using concentrated and clarified cultivation media may yield spectral support for the trace metal chelation. Additional common survival features may include regulation of the bulk osmolarity, as suggested by decreases in intracellular compatible solutes for the 50v1 strain (e.g., glycerate, 3-phosphoglycerate, and mannitol), statistical associations (CCA plot) between survival and changes in intracellular compatible solutes (e.g., trehalose) for the 2P08AA strain, and increase in an extracellular compatible solute for the 2P08AA strain (e.g., sorbitol). When considering intracellular changes, differing responses to $0.1\%$ v/v K30 are observed by the 2P08AA and 50v1 strains. While limited intracellular responses are observed for the 2P08AA strain, substantial intracellular changes are observed for the 50v1 strain. Multivariate tests for the 50v1 strain support a potential for the synthesis of light scattering PHAs during stationary phase in response to $0.1\%$ v/v K30. Relatedly, univariate tests suggest decreases in the export of α-ketoglutarate and 4-hydroxybutanoate (e.g., decreases in extracellular abundances), while multivariate tests show correlations between intracellular PHA synthesis (e.g., α-ketoglutarate, 4-aminobutanoate, and 4-hydroxybutanoate) and increases in the apparent maximum relative biomass. For the 2P08AA strain, multivariate analyses highlight a potential for partial metabolism of the detergents (from the Kleenol 30 formulation) – which would serve as an additional biochemical strategy toward survival. While these exact trends were not observed for the 50v1 strain, multivariate tests do support correlations between survival and changes in the degradation of aromatic compounds (benzene metabolism). In support, GC–MS studies in Mogul et al. [ 2018] provided evidence for scission of polyethylene glycol mono-nonylphenyl ether by the 50v1 strain. Future targeted cultivation studies are required to confirm the extent of biodegradation of the detergent by these strains. When considering impacts of K30 on the lag phase, we observe a differential concentration dependence, where the time in lag phase is eliminated in the presence of 0.1 v/v K30 for the 2P08AA strain, and in $1.0\%$ v/v K30 for the 50v1 strain. We venture that metabolism of the alkyl sides chains on the detergents (via ω-oxidation of the terminal carbon and subsequent cycles of β-oxidation) may serve as activators for acceleration out of the lag phase, and that degradation of the aromatic detergent cores (via benzene metabolism) may sustain the 50v1 strain during exponential phase, where growth rates show no inhibition in the presence of $0.1\%$ v/v K30. If so, carbon and energy acquisition from the detergents could serve as an added biochemical strategy toward survival. Alternatively, the K30 detergents could potentially disperse any low abundance cellular aggregates during the lag phase, thereby yielding higher OD values and the apparent increases in biomass. Under this assumption, the observed concentration dependence across the strains would be suggestive of the cellular aggregates from both strains having differing stabilities. ## Conclusion In conclusion, our combined cultivation and metabolomic results lend supports toward the hypothesis that repetitive cleansing with Kleenol 30 presents as selective pressures toward members of the spacecraft microbiome – where the phenotypic outcomes likely include cultivation tolerances and measurable biochemical responses. Fittingly, our described results indicate that Kleenol 30 is more readily tolerated by the floor-associated A. johnsonii 2P08AA, when compared to the spacecraft surface-associated A. radioresistens 50v1. However, both strains display reasonable tolerances toward $1\%$ v/v Kleenol 30, which is higher than the current formulation (~0.8-$1.6\%$ v/v) used in NASA planetary protection practices. These results indicate that the spacecraft-associated Acinetobacter may harbor tolerances toward the floor cleaning conditions. Such tolerances are likely important considerations for lower bioburden planetary missions, such as orbiter missions to Europa (Planetary Protection Category III), life detection missions to Mars (Planetary Protection Category IVb) or Europa (Planetary Protection Category IV), and investigations of Mars Special Regions (Planetary Protection Category IVc). In this context, our results suggest that higher floor cleanliness levels may be achieved with higher concentrations of Kleenol 30 (e.g., 2–$3\%$ v/v) or rotations with differing concentrations of Kleenol 30 (e.g., 1 and $3\%$ v/v). Alternatively, rotation of differing floor cleansers could assist in achieving lower floor bioburdens (e.g., Kleenol 30 and a quaternary ammonium compound), with the caveat that the detergents and/or biocides retain compatibility with the spacecraft assembly procedures and spacecraft materials. ## Data availability statement Metabolomic data (Study ID ST002380, DatatrackID: 3552) are available through the Metabolomics Workbench (https://www.metabolomicsworkbench.org/) at the NIH Common Fund’s National Metabolomics Data Repository (NMDR) and can be accessed directly via the project DOI: http://dx.doi.org/10.21228/M8XH73. ## Author contributions RM, DM, SL, and BR contributed to analysis of the data, assisted in drafting of the report, approved of submission of the manuscript. RM analyzed the metabolomic data, performed the statistical analyses, prepared the manuscript, and is the corresponding author. DM and BR conducted the survival assays and cultivation kinetics. SL prepared cultures for metabolomics analysis. All authors contributed to the article and approved the submitted version. ## Funding This work was supported by a grant (NNH18ZDA001N) from the Planetary Protection Research component of the National Aeronautics and Space Administration (NASA) Research Opportunities in Space and Earth Sciences program. ## 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. Agalloco J. P., Carleton F. J.. Validation of Pharmaceutical Processes (2007) 2. Angyal S., Doane W. M.. Complex formation between sugars and metal ions. Carbohydrate Chemistry–VI. (1973) 131-146 3. Barupal D. K., Fiehn O.. Chemical similarity enrichment analysis (ChemRICH) as alternative to biochemical pathway mapping for metabolomic datasets. Sci. Rep. (2017) 7 14567. DOI: 10.1038/s41598-017-15231-w 4. Barupal D. K., Haldiya P. K., Wohlgemuth G., Kind T., Kothari S. L., Pinkerton K. E.. MetaMapp: mapping and visualizing metabolomic data by integrating information from biochemical pathways and chemical and mass spectral similarity. BMC Bioinform. (2012) 13 1-15. DOI: 10.1186/1471-2105-13-99 5. Barupal D. K., Zhang Y., Shen T., Fan S., Roberts B. S., Fitzgerald P.. A comprehensive plasma metabolomics dataset for a cohort of mouse knockouts within the international mouse phenotyping consortium. Meta (2019) 9 101. DOI: 10.3390/metabo9050101 6. Begot C., Desnier I., Daudin J. D., Labadie J. C., Lebert A.. Recommendations for calculating growth parameters by optical density measurements. J. Microbiol. Methods (1996) 25 225-232. DOI: 10.1016/0167-7012(95)00090-9 7. Benjamini Y., Hochberg Y.. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J. R. Stat. Soc. B (1995) 57 289-300. DOI: 10.1111/j.2517-6161.1995.tb02031.x 8. Carper W., Coffin D., Addis J.. Studies of magnesium and manganese interactions with gluconate and 1, 5-gluconolactone. Spectrochim. Acta Part A (1989) 45 391-392. DOI: 10.1016/0584-8539(89)80106-0 9. Castillo-Juarez I., Lopez-Jacome L. E., Soberón-Chávez G., Tomás M., Lee J., Castañeda-Tamez P.. Exploiting quorum sensing inhibition for the control of. Curr. Top. Med. Chem. (2017) 17 1915-1927. DOI: 10.2174/1568026617666170105144104 10. Chang J. C., Jimenez J. C., Federle M. J.. Induction of a quorum sensing pathway by environmental signals enhances group a streptococcal resistance to lysozyme. Mol. Microbiol. (2015) 97 1097-1113. DOI: 10.1111/mmi.13088 11. Checinska A., Probst A. J., Vaishampayan P., White J. R., Kumar D., Stepanov V. G.. Microbiomes of the dust particles collected from the international Space Station and spacecraft assembly facilities. Microbiome (2015) 3 50. DOI: 10.1186/s40168-015-0116-3 12. Checinska Sielaff A., Urbaniak C., Mohan G. B. M., Stepanov V. G., Tran Q., Wood J. M.. Characterization of the total and viable bacterial and fungal communities associated with the international Space Station surfaces. Microbiome (2019) 7 50. DOI: 10.1186/s40168-019-0666-x 13. Chutipongtanate S., Sutthimethakorn S., Chiangjong W., Thongboonkerd V.. Bacteria can promote calcium oxalate crystal growth and aggregation. J. Biol. Inorg. Chem. (2013) 18 299-308. DOI: 10.1007/s00775-012-0974-0 14. Danko D. C., Sierra M. A., Benardini J. N., Guan L., Wood J. M., Singh N.. A comprehensive metagenomics framework to characterize organisms relevant for planetary protection. Microbiome (2021) 9 1-15. DOI: 10.1186/s40168-021-01020-1 15. Daschner F., Fahlberg W. J., Mallison G. F.. Floor disinfection in the United States. Infect. Control Hosp. Epidemiol. (1982) 3 281-284. DOI: 10.1017/S0195941700056289 16. Derecho I., Mccoy K. B., Vaishampayan P., Venkateswaran K., Mogul R.. Characterization of hydrogen peroxide–resistant. Astrobiology (2014) 14 837-847. DOI: 10.1089/ast.2014.1193 17. Eissa M. E., Kamel M., Beshir M.. Study of antimicrobial power of amphoteric disinfectants of Tego series used in pharmaceutical industry. Egypt. J. Biol. Sci. G (2014) 6 17-27. DOI: 10.21608/eajbsg.2014.16626 18. Ellingson K. D., Pogreba-Brown K., Gerba C. P., Elliott S. P.. Impact of a novel antimicrobial surface coating on health care–associated infections and environmental bioburden at 2 urban hospitals. Clin. Infect. Dis. (2020) 71 1807-1813. DOI: 10.1093/cid/ciz1077 19. Empadinhas N., Da Costa M. S.. Osmoadaptation mechanisms in prokaryotes: distribution of compatible solutes. Int. Microbiol. (2008) 11 151-161. PMID: 18843593 20. Farrow J. M., Wells G., Pesci E. C.. Desiccation tolerance in. PLoS One (2018) 13 e0205638. DOI: 10.1371/journal.pone.0205638 21. Fernández-Cuenca F., Tomás M., Caballero-Moyano F.-J., Bou G., Martínez-Martínez L., Vila J.. Reduced susceptibility to biocides in. J. Antimicrob. Chemother. (2015) 70 3222-3229. DOI: 10.1093/jac/dkv262 22. Fiehn O., Garvey W. T., Newman J. W., Lok K. H., Hoppel C. L., Adams S. H.. Plasma metabolomic profiles reflective of glucose homeostasis in non-diabetic and type 2 diabetic obese African-American women. PLoS One (2010) 5 e15234. DOI: 10.1371/journal.pone.0015234 23. Fiehn O., Wohlgemuth G., Scholz M., Doane W. M.. Setup and annotation of metabolomic experiments by integrating biological and mass spectrometric metadata. International Workshop on Data Integration in the Life Sciences (2005) 224-239 24. Franceus J., Pinel D., Desmet T.. Glucosylglycerate phosphorylase, an enzyme with novel specificity involved in compatible solute metabolism. Appl. Environ. Microbiol. (2017) 83 e01434. DOI: 10.1128/AEM.01434-17 25. Frick A., Mogul R., Stabekis P., Conley C. A., Ehrenfreund P.. Overview of current capabilities and research and technology developments for planetary protection. Adv. Space Res. (2014) 54 221-240. DOI: 10.1016/j.asr.2014.02.016 26. Frutos A. A., Signorella S., Sala L. F., Escandar G. M., Peregrin J. M. S., Moreno V.. Interaction of divalent metal ions with D-aldonic acids in the solid phase-II. Structural information fn1. Polyhedron (1998) 17 3371-3378. DOI: 10.1016/S0277-5387(98)00119-3 27. Guo R., Luo X., Liu J., Lu H.. Mass spectrometry based targeted metabolomics precisely characterized new functional metabolites that regulate biofilm formation in. Anal. Chim. Acta (2021) 1145 26-36. DOI: 10.1016/j.aca.2020.12.021 28. Gyurcsik B., Nagy L.. Carbohydrates as ligands: coordination equilibria and structure of the metal complexes. Coord. Chem. Rev. (2000) 203 81-149. DOI: 10.1016/S0010-8545(99)00183-6 29. Haffenden L. J., Yaylayan V. A.. Mechanism of formation of redox-active hydroxylated benzenes and pyrazine in. J. Agric. Food Chem (2005) 53 9742-9746. DOI: 10.1021/jf051610r 30. Halket J. M., Waterman D., Przyborowska A. M., Patel R. K., Fraser P. D., Bramley P. M.. Chemical derivatization and mass spectral libraries in metabolic profiling by GC/MS and LC/MS/MS. J. Exp. Biol. (2005) 56 219-243. DOI: 10.1093/jxb/eri069 31. Hammer Ø., Harper D. A., Ryan P. D.. PAST: paleontological statistics software package for education and data analysis. Palaeontol. Electron. (2001) 4 9 32. Han Z., Pappas E., Simmons A., Fox J., Donskey C. J., Deshpande A.. Environmental cleaning and disinfection of hospital rooms: a nationwide survey. Am. J. Infect. Cont. (2021) 49 34-39. DOI: 10.1016/j.ajic.2020.08.008 33. Hashim M., Kulandai J., Hassan R.. Biodegradability of branched alkylbenzene sulphonates. J. Chem. Technol. Biotechnol. (1992) 54 207-214. DOI: 10.1002/jctb.280540302 34. Hawthorne F., Borys I., Ferguson R.. Structure of calcium tartrate tetrahydrate. Acta Crystallogr. B (1982) 38 2461-2463. DOI: 10.1107/S0567740882009042 35. Hendrickson R., Urbaniak C., Minich J. J., Aronson H. S., Martino C., Stepanauskas R.. Clean room microbiome complexity impacts planetary protection bioburden. Microbiome (2021) 9 238. DOI: 10.1186/s40168-021-01159-x 36. Izatulina A. R., Gurzhiy V. V., Krzhizhanovskaya M. G., Kuz’mina M. A., Leoni M., Frank-Kamenetskaya O. V.. Hydrated calcium oxalates: crystal structures, thermal stability, and phase evolution. Cryst. Growth Des. (2018) 18 5465-5478. DOI: 10.1021/acs.cgd.8b00826 37. Kessler B., Witholt B.. Factors involved in the regulatory network of polyhydroxyalkanoate metabolism. J. Biotechnol. (2001) 86 97-104. DOI: 10.1016/S0168-1656(00)00404-1 38. Khanna S., Srivastava A. K.. Recent advances in microbial polyhydroxyalkanoates. Process Biochem. (2005) 40 607-619. DOI: 10.1016/j.procbio.2004.01.053 39. La Duc M. T., Nicholson W., Kern R., Venkateswaran K.. Microbial characterization of the Mars odyssey spacecraft and its encapsulation facility. Environ. Microbiol. (2003) 5 977-985. DOI: 10.1046/j.1462-2920.2003.00496.x 40. La Duc M. T., Vaishampayan P., Nilsson H. R., Torok T., Venkateswaran K.. Pyrosequencing-derived bacterial, archaeal, and fungal diversity of spacecraft hardware destined for Mars. Appl. Environ. Microbiol. (2012) 78 5912-5922. DOI: 10.1128/AEM.01435-12 41. Lu H., Que Y., Wu X., Guan T., Guo H.. Metabolomics deciphered metabolic reprogramming required for biofilm formation. Sci. Rep. (2019) 9 13160. DOI: 10.1038/s41598-019-49603-1 42. Mahnert A., Vaishampayan P., Probst A. J., Auerbach A., Moissl-Eichinger C., Venkateswaran K.. Cleanroom maintenance significantly reduces abundance but not diversity of indoor microbiomes. PLoS One (2015) 10 e0134848. DOI: 10.1371/journal.pone.0134848 43. McCoy K., Derecho I., Wong T., Tran H., Huynh T., La Duc M.. Insights into the extremotolerance of. Astrobiology (2012) 12 854-862. DOI: 10.1089/ast.2012.0835 44. Meehan E., Einspahr H., Bugg C. E.. Calcium binding to hydroxycarboxylic acids: crystal structure of calcium di-dl-glycerate dihydrate. Acta Crystallogr. B (1979) 35 828-832. DOI: 10.1107/S0567740879004994 45. Meyer B., Cookson B.. Does microbial resistance or adaptation to biocides create a hazard in infection prevention and control?. J. Hosp. Infect. (2010) 76 200-205. DOI: 10.1016/j.jhin.2010.05.020 46. Mogul R., Barding G. A., Lalla S., Lee S., Madrid S., Baki R.. Metabolism and biodegradation of spacecraft cleaning reagents by strains of spacecraft-associated. Astrobiology (2018) 18 1517-1527. DOI: 10.1089/ast.2017.1814 47. Moissl-Eichinger C., Auerbach A. K., Probst A. J., Mahnert A., Tom L., Piceno Y.. Quo vadis? Microbial profiling revealed strong effects of cleanroom maintenance and routes of contamination in indoor environments. Sci. Rep. (2015) 5 9156. DOI: 10.1038/srep09156 48. Mora M., Mahnert A., Koskinen K., Pausan M. R., Oberauner-Wappis L., Krause R.. Microorganisms in confined habitats: microbial monitoring and control of intensive care units, operating rooms, cleanrooms and the international Space Station. Front. Microbiol. (2016) 7 1573. DOI: 10.3389/fmicb.2016.01573 49. Murtough S., Hiom S., Palmer M., Russell A.. Biocide rotation in the healthcare setting: is there a case for policy implementation?. J. Hosp. Infect. (2001) 48 1-6. DOI: 10.1053/jhin.2001.0950 50. NASA. (2011). Planetary Protection Provisions for Robotic Extraterrestrial Missions. Washington, DC: National Aeronautics and Space Administration.. Planetary Protection Provisions for Robotic Extraterrestrial Missions (2011) 51. Peleg A. Y., Seifert H., Paterson D. L.. Clin. Microbiol. Rev. (2008) 21 538-582. DOI: 10.1128/CMR.00058-07 52. Perry-Dow K. A., De Man T. J., Halpin A. L., Shams A. M., Rose L. J., Noble-Wang J. A.. The effect of disinfectants on the microbial community on environmental healthcare surfaces using next generation sequencing. Am. J. Infect. Cont. (2022) 50 54-60. DOI: 10.1016/j.ajic.2021.08.027 53. Pirog T., Lutsai D., Muchnyk F.. Biotechnological potential of the. Mikrobiol. Zh. (2021) 83 92-109. DOI: 10.15407/microbiolj83.03.092 54. Pompilio A., Scribano D., Sarshar M., Di Bonaventura G., Palamara A. T., Ambrosi C.. Gram-negative bacteria holding together in a biofilm: the. Microorganisms (2021) 9 1353. DOI: 10.3390/microorganisms9071353 55. Rajamohan G., Srinivasan V., Gebreyes W. A.. Biocide-tolerant multidrug-resistant. J. Hosp. Infect. (2009) 73 287-289. DOI: 10.1016/j.jhin.2009.07.015 56. Rummel J. D.. Planetary protection policy (U.S.A.). Adv. Space Res. (1992) 12 129-131. DOI: 10.1016/0273-1177(92)90166-U 57. Rutala W. A., Weber D. J.. Best practices for disinfection of noncritical environmental surfaces and equipment in health care facilities: a bundle approach. Am. J. Infect. Cont. (2019) 47 A96-A105. DOI: 10.1016/j.ajic.2019.01.014 58. Rutala W. A., Weber D. J.. Guideline for Disinfection and Sterilization in Healthcare Facilities (2008) 59. Ryu E.-J., Sim J., Sim J., Lee J., Choi B.-K.. D-galactose as an autoinducer 2 inhibitor to control the biofilm formation of periodontopathogens. J. Microbiol. (2016) 54 632-637. DOI: 10.1007/s12275-016-6345-8 60. Sadiq F. A., Yan B., Zhao J., Zhang H., Chen W.. Untargeted metabolomics reveals metabolic state of. LWT (2020) 118 108772. DOI: 10.1016/j.lwt.2019.108772 61. Sandle T.. The CDC Handbook-a Guide to Cleaning and Disinfecting Clean Rooms (2012) 62. Sandle T.. Pharmaceutical Microbiology: Essentials for Quality Assurance and Quality Control (2015) 63. Sarkar S., Chakraborty R.. Quorum sensing in metal tolerance of. FEMS Microbiol. Lett. (2008) 282 160-165. DOI: 10.1111/j.1574-6968.2008.01080.x 64. Sauvage J.-P., Verchère J.-F., Chapelle S.. A multinuclear NMR spectroscopy study of the tungstate and molybdate complexes of d-fructose and l-sorbose. Carbohydr. Res. (1996) 286 67-76. DOI: 10.1016/0008-6215(96)00048-1 65. Sawyer D. T.. Metal-gluconate complexes. Chem. Rev. (1964) 64 633-643. DOI: 10.1021/cr60232a003 66. Serra-Compte A., Corcoll N., Huerta B., Rodríguez-Mozaz S., Sabater S., Barceló D.. Fluvial biofilms exposed to desiccation and pharmaceutical pollution: new insights using metabolomics. Sci. Total Environ. (2018) 618 1382-1388. DOI: 10.1016/j.scitotenv.2017.09.258 67. Shams A. M., Rose L. J., Edwards J. R., Cali S., Harris A. D., Jacob J. T.. Assessment of the overall and multidrug-resistant organism bioburden on environmental surfaces in healthcare facilities. Infect. Control Hosp. Epidemiol. (2016) 37 1426-1432. DOI: 10.1017/ice.2016.198 68. Slaninova E., Sedlacek P., Mravec F., Mullerova L., Samek O., Koller M.. Light scattering on PHA granules protects bacterial cells against the harmful effects of UV radiation. App. Microbiol. Biotechnol. (2018) 102 1923-1931. DOI: 10.1007/s00253-018-8760-8 69. Spence J., Kiang S.-C.. Complexes of sugars with molybdate. J. Org. Chem. (1963) 28 244-245. DOI: 10.1021/jo01036a522 70. Spry J. A., Siegel B., Race M., Kminek G., Pratt L., Coustenis A.. (2020) 71. Srinivasan V. B., Venkataramaiah M., Mondal A., Rajamohan G.. Functional characterization of AbeD, an RND-type membrane transporter in antimicrobial resistance in. PLoS One (2015) 10 e0141314. DOI: 10.1371/journal.pone.0141314 72. Sud M., Fahy E., Cotter D., Azam K., Vadivelu I., Burant C.. Metabolomics workbench: an international repository for metabolomics data and metadata, metabolite standards, protocols, tutorials and training, and analysis tools. Nucleic Acids Res. (2016) 44 D463-D470. DOI: 10.1093/nar/gkv1042 73. Suleyman G., Alangaden G., Bardossy A. C.. The role of environmental contamination in the transmission of nosocomial pathogens and healthcare-associated infections. Cur. Infect. Dis. Rep. (2018) 20 1-11. DOI: 10.1007/s11908-018-0620-2 74. Taga T., Ohashi M., Osaki K.. Interactions of calcium ions with carbohydrates: crystal structure of calcium DL-Glycerate Dihydrate. Bull. Chem. Soc. Japan (1978) 51 1697-1700 75. Tajmir-Riahi H.. Carbohydrate complexes with alkaline earth metal ions. Interaction of D-glucono-1,5-lactone with the mg(II), ca(II), Sr(II), and Ba(II) cations in the crystalline solid and aqueous solution. J. Inorg. Biochem. (1990) 39 33-41. DOI: 10.1016/0162-0134(90)80013-N 76. Tam H.-K., König P., Himpich S., Ngu N. D., Abele R., Müller V.. Unidirectional mannitol synthesis of. Proc. Natl. Acad. Sci. U. S. A. (2022) 119 e2107994119. DOI: 10.1073/pnas.2107994119 77. Tembo G. M., Chaggar G. K., Li X., Teska P. J., Oliver H. F.. Evaluation of automated floor cleaning, disinfection, and application methods against. Am. J. Infect. Cont. (2022) 19 S0196-6553(22)00535-1. DOI: 10.1016/j.ajic.2022.07.011 78. Tielen P., Rosin N., Meyer A.-K., Dohnt K., Haddad I., Jänsch L.. Regulatory and metabolic networks for the adaptation of. PLoS One (2013) 8 e71845. DOI: 10.1371/journal.pone.0071845 79. Trigg S. A., Mcelhany P., Maher M., Perez D., Busch D. S., Nichols K. M.. Uncovering mechanisms of global ocean change effects on the Dungeness crab (. Sci. Rep. (2019) 9 1-12. DOI: 10.1038/s41598-019-46947-6 80. Van Duin M., Peters J. A., Kieboom A. P., Van Bekkum H.. Synergic coordination of calcium in borate-polyhydroxycarboxylate systems. Carbohydr. Res. (1987) 162 65-78. DOI: 10.1016/0008-6215(87)80201-X 81. Wang Y., Wang Y., Chen J., Koseki S., Yang Q., Yu H.. Screening and preservation application of quorum sensing inhibitors of. LWT (2021) 149 111749. DOI: 10.1016/j.lwt.2021.111749 82. White G. F.. Bacterial biodegradation of ethoxylated surfactants. Pest. Sci. (1993) 37 159-166. DOI: 10.1002/ps.2780370209 83. Wong W. C., Zalesak S., Yoets A., Capriotti J., Smith M. J., Castro V. A.. Effectiveness of HEPA filter vacuum in removing transient microbial contaminants on cargo bags destined for the international Space Station. J. Occup. Environ. Hyg. (2013) 10 D71-D75. DOI: 10.1080/15459624.2013.784179 84. Xu C., Li M., Huang Y., Zhang Z., Bian Z., Song S.. Involvement of quorum-sensing in biosynthesis of polyhydroxyalkanoates in. Acta Microbiolog. Sin. (2011) 51 769-775. PMID: 21866701 85. Yeom J., Shin J.-H., Yang J.-Y., Kim J., Hwang G.-S.. 1H NMR-based metabolite profiling of planktonic and biofilm cells in. PLoS One (2013) 8 e57730. DOI: 10.1371/journal.pone.0057730 86. Zhao J., Han M.-L., Zhu Y., Lin Y.-W., Wang Y.-W., Lu J.. Comparative metabolomics reveals key pathways associated with the synergistic activity of polymyxin B and rifampicin combination against multidrug-resistant. Biochem. Pharmacol. (2021) 184 114400. DOI: 10.1016/j.bcp.2020.114400
--- title: Involvement of p38 MAPK in Leydig cell aging and age-related decline in testosterone authors: - Dandan Luo - Xiangyu Qi - Xiaoqin Xu - Leilei Yang - Chunxiao Yu - Qingbo Guan journal: Frontiers in Endocrinology year: 2023 pmcid: PMC10025507 doi: 10.3389/fendo.2023.1088249 license: CC BY 4.0 --- # Involvement of p38 MAPK in Leydig cell aging and age-related decline in testosterone ## Abstract ### Introduction Age-related decline in testosterone is associated with Leydig cell aging with impaired testosterone synthesis in aging. Obesity accelerates the age-related decline in testosterone. However, the mechanisms underlying the Leydig cell aging and the effects of obesity on Leydig cell aging remain unclear. ### Method Natural aging mice and diet-induced obese mice were used to assess the process of testicular Leydig cell senescence with age or obesity. Bioinformatic analysis of the young and aged human testes was used to explore key genes related Leydig cell aging. Leydig cell-specific p38 MAPK knockout (p38LCKO) mice were used to further analyze the roles of p38 MAPK in Leydig cell aging. The levels of testosterone and steroidogenic enzymes, activity of p38 MAPK, aging status of Leydig cells, and oxidative stress and inflammation of testes or Leydig cells were detected by ELISA, immunoblotting, immunofluorescence, and senescence-associated β-galactosidase (SA-β-Gal) staining analysis, respectively. ### Result The serum testosterone level was significantly reduced in aged mice compared with young mice. In the testis of aged mice, the reduced mRNA and protein levels of LHCGR, SRB1, StAR, CYP11A1, and CYP17A1 and the elevated oxidative stress and inflammation were observed. KEGG analysis showed that MAPK pathway was changed in aged Leydig cells, and immunoblotting displayed that p38 MAPK was activated in aged Leydig cells. The intensity of SA-β-Gal staining on Leydig cells and the number of p21-postive Leydig cells in aged mice were more than those of young mice. Similar to aged mice, the testosterone-related indexes decreased, and the age-related indexes increased in the testicular Leydig cells of high fat diet (HFD) mice. Aged p38LCKO mice had higher levels of testosterone and steroidogenic enzymes than those of age-matched wild-type (WT) littermates, with reduced the intensity of SA-β-Gal staining and the expression of p21 protein. ### Conclusion Our study suggested that obesity was an important risk factor for Leydig cell aging. p38 MAPK was involved in Leydig cell aging induced by age and obesity. The inhibition of p38 MAPK could delay Leydig cell aging and alleviate decline in testosterone. ## Introduction In males, serum testosterone decreases with age after the age 40 years, known as age-related decline in testosterone [1]. Decreased testosterone causes infertility, sexual dysfunction, and age-related symptoms in older male populations known as late-onset hypogonadism [2]. Population aging remarkably increases the incidence of this disease [3]. Men with testosterone deficiency benefit a lot from testosterone replacement therapy (TRT), including improved bone density and strength, body composition, sexual function, and mood [4, 5]. However, the long-term effects of TRT on major cardiovascular events and prostate cancer risk in older men remain controversial. Recent clinical trials in older men have shown inconsistent effects of TRT on cardiovascular events, with some showing positive effects, a reduced risk of atrial fibrillation and myocardial infarction (6–8), some showing negative effects, an increased the risk of death, myocardial infarction, and stroke (9–11), and others showing no effects [12]. And the controversial observations also exist in the effects of TRT on prostate cancer (13–16). The reason for these results might be related to the subject characteristics and the length of treatment. Neither the clinical benefits nor the long-term safety of TRT has been fully established in older men with age-related decline in testosterone, thus, elucidating the risk factors and molecular mechanisms of age-related decline in testosterone will contribute to the prevention and treatment of this disease. Obesity is considered a major culprit of cell aging and aging-related diseases, such as neurodegenerative disorders, cardiovascular diseases, and cancer (17–19). Cellular alterations caused by obesity, such as oxidative stress, inflammation, and DNA damage accumulation are the potential driving factors for the aging process [20]. Epidemiologic studies show that obese men have lower testosterone compared to lean men in diverse ethnic population (21–24). Recent studies suggest that obesity is related to age-related decline in testosterone. In the European male aging study, obesity is the greatest determinant of the variance in serum testosterone with age [25]. Serum testosterone decline with age in obese patients is more pronounced than in lean individuals [26]. Weight control in obese older men is associated with an increase in total testosterone and free testosterone [27]. All the above studies suggest that obesity epidemic has the potential to exacerbate the age-related decline in testosterone. Serum testosterone in males is synthesized by testicular Leydig cells under the control of luteinizing hormone (LH). Steroidogenic acute regulatory protein (StAR) and a series of steroidogenic enzymes (CYP11A1, 3β-HSD, and CYP17A1) are responsible for testosterone synthesis. Accumulating evidence suggest that age- or obesity-related decline in testosterone is correlated with impaired testosterone biosynthesis of Leydig cells. The level of LH is unchanged or slightly increased in most males [28, 29]. Decreased expressions of StAR and steroidogenic enzymes account for impaired function of Leydig cells [27, 30, 31]. Whether the impaired function of Leydig cells in aged males follows cell aging, and whether the obesity accelerates Leydig cell aging remains unknown. Few works have been done to investigate the Leydig cell aging and the underly mechanisms. The p38 mitogen-activated protein kinase (p38 MAPK) belongs to the family of MAPKs, which is involved in a variety of cell stress response [3]. Oxidative stress and chronic inflammatory state, the common pathological features underlying aging and obesity, are strong inducer of p38 MAPK activation [32]. It has been demonstrated that p38 MAPK is implicated in the obesity- and age-related diseases [33, 34]. Recent reports showed that p38 MAPK is activated in Leydig cells under oxidative stress and might be involved in reduced expression of StAR [35]. Whether the p38 MAPK is involved in the Leydig cell aging with age or obesity in still unknow. In the present study, we found that obesity was a risk factor for Leydig cell aging. p38 MAPK was implicated in Leydig cell aging and age- and obesity-related decline in testosterone. ## Materials Testosterone and LH enzyme-linked immunosorbent assay (ELISA) kit was obtained from CUSABIO (Wuhan, China). Mouse TNFα, IL-1β, and IL-6 ELISA were obtained from Lianke Biotech (Hangzhou, China). SOD and MDA detection kit and senescence β-galactosidase (SA-β-gal) staining kit were purchased from Biyuntian (Shanghai, China). The bicinchoninic acid (BCA) protein assay kit was purchased from Shenneng Bocai (Shanghai, China). RNAiso reagent and reverse transcription kit were purchased from Takara (Tokyo, Japan). SYBR green PCR master mix was purchased from Yeasen (Shanghai, China). Antibodies against phosphorylated p38α MAPK (p-p38 MAPK, Thr180/Tyr182), p38α MAPK, phospho-extracellular signal-regulated kinase $\frac{1}{2}$ (anti-p-ERK$\frac{1}{2}$, Thr202/Tyr204), anti-ERK, anti-p-c-Jun N-terminal kinase (p-JNK, hr183/Tyr185), anti-JNK, StAR and CYP11A1 were purchased from CST (Boston, MA, USA). Antibodies against CYP17A1, Hsp90, and Tubulin were from Proteintech (Wuhan, China). Mouse monoclonal antibodies against p21 and p16 were purchased from Santa Cruz Biotechnology (Santa Cruz, CA, USA). Antibody against p-p38 MAPK (Thr180) for IHC was obtained from ZEN-BIOSCIENCE (Chengdu, China). All other chemicals used in this study were of analytical grade and obtained from Sigma (St. Louis, MO, USA) unless otherwise stated. ## Animals The experimental design and animal treatment protocol were approved by the Animal Ethics Committee of Shandong Provincial Hospital. Animals were raised in controlled environmental conditions (22 ± 2°C; 12h light/dark cycle) with food and water ad libitum. The mice were sacrificed at different times according to the experimental plan. Serum samples were collected from eyeball blood in anesthetized mice. Blood samples centrifuged at 3000rpm for 15 min and stored at −80°C until assayed. One testis of each mouse was stored at −80°C for subsequent experiments. Another one was fixed in mDF (37–$40\%$ formaldehyde, absolute ethanol, glacial acetic acid, distilled water volume ratio of 3:1.5:0.5:5) for 24h for subsequent analysis. ## Assessment of serum lipid levels To evaluate the effect of HFD on serum lipid in mice, serum total cholesterol (TC), triglyceride (TG), high-density lipoprotein-cholesterol (HDL-C) and low-density lipoprotein-cholesterol (LDL-C) were measured directly using automatic biochemical analyzer (Beckmen, AU5831, USA) at the clinical laboratory of Shandong Provincial Hospital. ## Assessment of serum hormone and testicular testosterone levels The mouse serum testosterone and LH levels were measured by an ELISA kit according to the manufacturer’s protocols. For intratesticular testosterone assay, testis tissues (10 mg) were homogenized by a tissue homogenizer in 100 μl phosphate buffered solution. Then the homogenates were lysed by three times freeze-thaw and centrifuged at 12,000 rpm for 10 min to obtain the supernatant. Testosterone concentrations were detected with ELISA kit for testosterone according to the kit was expressed as ng/mg tissue weight ## Detection of antioxidant enzymes in testis tissue Testis tissues (10 mg) was homogenized with 100μl assay buffer and centrifuged at 12,000 rpm for 10 min to obtain the supernatant. Then the contents of MDA and SOD were obtained by using respective assay kit according to the manufacture’s protocols. Then they were normalized to protein concentrations, which is examined by BCA protein assay kit. ## Detection of inflammatory factors in testis tissue The testis tissue was also prepared for inflammation factors analysis by repeated freeze-thaw treatments. The levels of TNFα, IL-1β, and IL-6 were detected with ELISA kits according to the kit manufacturer’s protocol and then were normalized to protein concentrations. ## Senescence-associated β-galactosidase staining SA-β-Gal staining of Leydig cells was performed utilizing a SA-β-Gal staining kit. The frozen sections (10μm) from mouse testis were fixed fixative solution and then incubated with fresh β-galactosidase staining solution at 37°C for at least 12 h. Aged cells displayed a blue color in the cytoplasm and the intensity of staining indicated the cell senescence status. Images were acquired using a light microscope (Imager A2, Zeiss, Germany). ## Measurement of lipid content using oil red O staining For examination of lipid content, the frozen sections (10μm) from mouse testis were fixed with $4\%$ formalin for 30 min, washed with PBS, then stained with Oil Red O for 10 min. The nucleus was counterstained with hematoxylin. Images were acquired using a light microscope. ## Single-cell sequencing analysis Single-cell RNA sequencing data of young and old human testis (GSE182786) were downloaded from https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE182786. The count data was imported into the Seurat (version 4.1.1) R package for quality control. Genes detected in < 3 cells and cells where < 1000 genes had nonzero counts were excluded. The low-quality cells that had > $15\%$ mitochondrial genes were also discarded. Library size normalization was performed in Seurat on the filtered matrix to obtain the normalized count. The count matrix for the respective gene in each sample were gained by using pseudobulks method. Then, we obtained the expression matrix and group information of the data. Subsequently, the differences between the two groups were analyzed by DESeq2 (version 1.34.0) R package. ## RNA isolation and quantitative real-time PCR RNAiso Reagent was used to extract total mRNA from testicular tissues. mRNA samples were reverse transcribed by reverse transcription kit according to the manufacturer’s instructions. The cDNA obtained by reverse transcription was used as a template for Quantitative real time PCR (Q‐PCR) conducted with the LightCycler480 (Roche). Use the 2−ΔΔCt method to calculate the relative expression of mRNA. The sequences of the PCR primers (synthesized by Qingdaoqingke Biotech Co., Ltd., China) of Lhcgr (5′- GATGCACAGTGGCACCTTC and 5′- TCAGCGTGGCAACCAGTAG), Star (5′-GGAGCA GAGTGGTGTCATCAGA and 5′-AGGTGGTTGGCG AACTCTATCT), Cyp11a1 (5′-TGCTTGAGAGGCTGGAAGTTGA and 5′-CGG ATTGCGGAGCTGGAGAT), Cyp17a1 (5′-GTACCC AGGCGAAGAGAATAGA and 5′-GCCCAAGTCAAA GACACCTAAT), 3β-hsd (5′-AGCTCTGGACAAAGT ATTCCGA and 5′-GCCTCCAATAGGTTCTGGGT) and β-actin (5′-GGCTGTATTCCCCTCCATCG and 5′-CCA GTTGGTAACAATGCCATGT). ## Immunoblotting analysis Tissue extracts were prepared by protein lysis buffer (RIPA: PMSF = 100:1) with a phosphatase inhibitor cocktail, and concentration was subsequently determined. Protein sample (60µg) was separated by 10 or $12\%$ separating gel and electrophoretically transferred to polyvinylidene difluoride membranes (PVDF membrane, Millipore Corporate). The membranes were blocked with $5\%$ milk in PBS for 1 h and incubated with primary antibodies overnight at 4°C. The membranes were subsequently incubated with secondary antibodies. Finally, the bands were visualized by chemiluminescence and quantified by Photoshop (Adobe Software). The relative expression levels of tested proteins were normalized to the corresponding tubulin. ## Hematoxylin and eosin staining To verify the histopathological changes of the testis, testicular tissue samples were dehydrated in a graded series of ethanol solutions and embedded in paraffin. Sections (5 μm) were de-paraffinized, rehydrated, and histologically stained with hematoxylin-eosin (HE) according to standard protocols, then imaged with a light microscope. ## Immunofluorescence and immunohistochemistry For immunofluorescence (IF), fixed tissues were heat treated in Tris‐EDTA buffer for antigen retrieval. After washed by PBS, sections were blocked with $5\%$ goat serum albumin, then incubated primary antibody overnight at 4°C. After that, sections were incubated with the FITC‐conjugated second antibody and DAPI for 1 h at room temperature. Each slice was observed with a confocal microscope (Leica, Germany). For Immunohistochemistry (IHC), after antigen retrieval, the tissue sections were incubated with $3\%$ peroxide blocking solution for 20 min at room temperature before adding primary antibody and were incubated overnight at 4°C. The slides were then labeled with horseradish peroxides-conjugated streptavidin. The chromogenic reaction that was developed using Liquid DAB Substrate according to the manufacturer’s instructions. Images were acquired using a light microscope. ## Statistical analyses Statistical analyses were conducted using GraphPad Prism (v.9, GraphPad Software Inc, CA). All samples were determined by an independent‐sample t test. All data were expressed as mean ± SD, and the difference was considered statistically significant if $p \leq 0.05.$ ## A decline in serum testosterone and testosterone synthesis of aged male mice The naturally aged mice were used to detect the age-related decline in serum testosterone and testosterone synthesis. We evaluated the general features and sex hormone levels of young mice (3 and 6 months) and aged mice (18 months). Body weight of aged mice was greater than that of 3-months old mice (Figure 1A, $p \leq 0.01$) and 6-months old mice (Figure 1A, $p \leq 0.05$). There was no change in mouse testis weight of all age (Figure 1B). The ratio of testis to body weight gradually decreased with age. The ratio of 3-months old mice was higher than that of 6-months old mice (Figure 1C, $p \leq 0.05$) and 18-months old mice (Figure 1C, $p \leq 0.01$). The serum testosterone level in aged mice was reduced compared with 3-and 6-months old mice (Figure 1D, $p \leq 0.05$). There was no difference in LH level among three groups of mice (Figure 1E). Then, we selected 6-months old (young) and 18-months old (aged) mice to further evaluate the ability of Leydig cell to produce testosterone. Figure 1: General features and testosterone synthesis of young and aged mice. (A) Body weight. (B) Testis weight. (C) Ratio of testis to body weight. (D, E) The levels of serum testosterone and serum luteinizing hormone (LH) were analyzed by ELISA. (F) Testis morphology was examined by HE staining. (G) The mRNA levels of proteins associated with testosterone synthesis were analyzed by Quantitative real time PCR. (H) Immunoblot assessed the protein levels of proteins related to testosterone synthesis. (I) Quantitative analyses for the protein level. (J) Immunofluorescence measured testosterone synthesis-related proteins. (K) Quantitative analyses of immunofluorescence. p<0.05, *p<0.01. HE showed that the area of Leydig cells appeared normal across two groups of mice. Compared to the young mice, aged mice had lower height of the seminiferous epithelium (Figure 1F). The mRNA levels of the proteins associated with testosterone synthesis in Leydig cells of aged mice showed a decreasing trend, among which the difference in Star ($p \leq 0.05$), Cyp11a1 ($p \leq 0.05$), and Cyp17a1 ($p \leq 0.01$) had statistically significances (Figure 1G). The protein levels of testosterone synthesis-related proteins were lower in aged mice than those in young mice (Figures 1H, I, $p \leq 0.05$). The fluorescence intensity of SRB1, StAR, CYP11A1, and CYP17A1 were weaker in Leydig cells of aged mice than those in young mice with statistically significances (Figures 1J, K, $p \leq 0.05$). The above results confirmed that the ability of Leydig cell to produce testosterone in aged mice was decreased. ## p38 MAPK was activated in Leydig cells of aged mice To understand the alterations in physiological state of testis or Leydig cells with age, we evaluated related indexes of aging status, oxidative stress, and inflammation. Accumulation of SA-β-Gal and increased levels of p21 and p16 are the most common markers of cell aging. Compared with young mice, SA-β-Gal staining was deeper in testicular interstitium of aged mice (Figure 2A). And the protein level of p21 was also higher in Leydig cells of aged mice than that of young mice, detected by immunoblot and IHC staining (Figures 2B–D, $p \leq 0.05$). SOD and MDA are two indices of oxidative stress. The levels of MDA were elevated in testis of aged mice compared with young mice (Figure 2E, $p \leq 0.05$), and there was no difference in SOD (Figure 2F). The expressions of pro-inflammatory (IL-6 and IL-1β) were significantly increased in testis of aged mice (Figure 2G, $p \leq 0.05$). The above results suggested the existence of oxidative stress and inflammation in aged mouse testis. Figure 2: Age-related alterations in testes and Leydig cells. (A) Senescence-associated β-galactosidase staining in testes of young and aged mice. (B) The p21 and p16 protein expression in testis of two group mice were analyzed by immunoblot. (C) Quantitative analysis for the protein levels of p21 and p16. (D) The p21 expression and location was analyzed by immunohistochemistry. The levels of MDA (E), SOD (F), and inflammation factors (G) in the testis tissue of two groups of mice were detected by assay kits. The top 15 KEGG terms (H) and GO terms (I) enriched in the differentially expressed genes in Leydig cells from young and old human testis single-cell RNA sequencing were listed with p value and gene numbers. The p-p38 MAPK protein level in testis of two groups of mice were assessed by immunoblot (J) and IHC staining (L). (K) Quantitative analysis for protein levels of p-p38 MAPK. p< 0.05, *p< 0.01. To mine molecular mechanisms related to Leydig cell aging, we obtained and further enriched the differentially expressed genes (DEGs) between young and aged Leydig cells from single-cell RNA sequencing results of young and old human testes. The KEGG pathway analysis revealed the MAPK pathway as one of the most significantly affected pathway (Figure 2H). The MAPK pathway was also enriched by GO analysis (Figure 2I). In addition, KEGG analysis also identified a significant enrichment of genes in the p53 signaling pathway, which is important for p21 transcriptional activation (Figure 2H). Then, MAPK family, including ERK$\frac{1}{2}$, JNK and p38 MAPK were evaluated in young and aged Leydig cells. Immunoblot results showed that the levels of p-p38 MAPK in testis of aged mice were higher than that in young mice with statistically significance (Figures 2J, K, $p \leq 0.01$). Meanwhile, the protein levels of p-JNK and p-ERK$\frac{1}{2}$ were not changed (Supplementary Figure 1). IHC staining of p-p38 MAPK showed that increased level of p-p38 MAPK in aged Leydig cells (Figure 2L). The above results suggested that activiated-p38 MAPK was involved in Leydig cell aging. ## p38 MAPK was involved in obesity induced Leydig cell aging To explore whether obesity could accelerate the Leydig cell aging and the underly mechanisms, we used a high fat diet (HFD) to induce obesity in the mice. The 8-weeks male mice developed diet-induced obesity with the $60\%$ high-fat diet for 24 weeks, as shown by increased body weight (Figure 3A, $p \leq 0.01$). There was no difference in testis weight between the two groups of mice (Figure 3B). Serum LDL-C ($p \leq 0.05$) and TC ($p \leq 0.01$) were significantly elevated in mice fed with HFD compared with mice fed with ND (Figure 3C). The results of Oil Red O staining showed that obvious lipid deposited in the testis interstitium of mice fed with HFD, which was more than that of mice fed with ND (Figure 3D). The mRNA levels of the proteins associated with testosterone synthesis in Leydig cells of mice fed with HFD showed a decreasing trend. The protein levels of testosterone synthesis-related proteins, including SRB1, StAR, and CYP17A1 were lower in mice fed with HFD than those in control mice (Supplementary Figure 2, $p \leq 0.05$). Figure 3: The aging status and p-p38 MAPK level of Leydig cells in mice fed with HFD or ND for 24 weeks. Body weight (A), testis weight (B), and serum lipid levels (C) were evaluated for mice. Lipid deposition in testicular tissues was detected by oil red O staining (D). Aging status of Leydig cells was assessed by SA-β-Gal staining (E). Immunoblot was used to detect the levels of p16, p21, and p-p38 MAPK (F). Quantitative analysis for protein level (G, H). IHC staining of p21 (I) and p-p38 MAPK (J) in testis tissue. p<0.05, *p<0.01. We then evaluated the effects of β-galactosidase staining was deeper in testis interstitium of mice fed with HFD (Figure 3E). Immunoblot and IHC staining showed that the expression of p21 protein was higher in mice fed with HFD than that in mice fed with ND (Figures 3F, G, J), which indicated that obesity promoted Leydig cell premature aging. The levels of p-p38 MAPK detected by immunoblot and IHC staining in mice fed with HFD were higher compared with mice fed with ND (Figures 3F, H, I). All above results suggested that obesity was an important risk factor for Leydig cell aging, and p38 MAPK might be involved in obesity induced Leydig cell aging. ## Leydig cell-specific p38 MAPK knockout alleviated age-related decline in testosterone We successfully generated Leydig cell-specific p38 MAPK knockout (p38LCKO) mice. As shown in Figures 4A, B, the levels of p-p38 MAPK and total p38 MAPK were decreased significantly in testis tissues of p38LCKO mice ($p \leq 0.01$). Specificity of Cyp17a1-icre recombinase expression was determined by crossing with tdTomato reporter mice. Significant red fluorescence was only observed in the testis interstitium of Cyp17a1-iCre: tdTomato mice, while no fluorescence was observed in WT mice (Figure 4C). Figure 4: General features and serum testosterone levels of p38LCKO and WT mice. (A) The protein levels of t-p38 MAPK and p-p38 MAPK were analyzed by immunoblot. (B) Quantitative analyses for the protein levels. (C) Red fluorescence of Tdtomato in testis. The body weight (D), testis weight (E), ratio of testis to body weight (F), and testicular testosterone (G) of p38LCKO and WT mic at 3-,6-, and 18- months of age. p<0.05, p<0.01, p<0.001. Then we evaluated the features of p38LCKO mice at 3, 6, and 18 months of age. Body weight of p38LCKO mice at different ages were not different from those of age matched WT littermates (Figure 4D). Similarly, there were no differences in testis weight and the ratio of testis to body weight (Figures 4E, F) between two groups of mice. The above results illustrated that specific p38 MAPK knockout in Leydig cells did not affect the general features of mice. There were no differences in serum and testicular testosterone level between two groups of mice at 3 and 6 months of age. At 18 months of age, serum and testicular testosterone levels were higher in p38LCKO mice than that of WT mice (Figure 4G; Supplementary Figure 3A, $p \leq 0.05$). These results suggested that specific p38 MAPK knockout in Leydig cells could alleviate age-related decline in testosterone. ## Specific p38 MAPK knockout improved the testosterone synthesis in aged Leydig cells The effects of Leydig cell-specific p38 MAPK knockout on steroidogenesis synthesis in aged mice were further assessed. The morphology of testis did not differ between both groups of mice at 3-and 6-month of age (Supplementary Figure 3B) and at 18-month of age (Figure 5A). The mRNA levels of Lhcgr, Star, Cyp11a1, and Cyp17a1 were higher in 18-month-old p38LCKO mice than those in WT mice (Figure 5B, $p \leq 0.05$). The levels of these proteins in Leydig cells were higher in p38LCKO mice at 18-month of age (Figures 5C, D, $p \leq 0.05$). There were no differences in the mRNA and protein levels of those proteins in mice at age of 3 and 6 months between WT and p38LCKO mice (Supplementary Figures 3C–F). IF showed that, compared with aged WT mice, fluorescence intensity of the SRB1, StAR, CYP11A1, and CYP17A1 were stronger in p38LCKO mice (Figures 5E, F, $p \leq 0.05$). Figure 5: Assessment of ability of Leydig cell to produce testosterone in p38LCKO mice and WT mice at 18-month of age. (A) Testis morphology was examined by HE staining. (B) The mRNA levels of proteins related to were assessed by Q-PCR. (C) The protein levels of proteins associated with testosterone synthesis were analyzed by immunoblot. (D) Quantitative analysis for the protein level. (E) IF assessed the levels of testosterone synthesis-related proteins. (F) Quantitative analysis of fluorescence intensity. p<0.05, *p<0.01. ## Specific p38 MAPK knockout delayed Leydig cell aging in aged mice To determine whether the Leydig cell-specific p38 MAPK knockout could delay cell aging, we assess the aging status of Leydig cell in p38LCKO mice at 18-month of age. Compared with WT mice, there was a weaker staining of SA-β-Gal in the testicular interstitium of p38LCKO mice (Figure 6A). The protein level of p21 was lower in p38LCKO mice than that in WT mice (Figures 6B, C). IHC staining of p21 showed that the number of p21-positive Leydig cells in p38LCKO mice was less than that in WT mice (Figure 6D). The above results showed that Leydig cell-specific p38 MAPK knockout could delay cell aging. Figure 6: Assessment aging status of Leydig cell in p38LCKO mice and WT mice at 18-month of age. (A) SA-β-Gal staining in testes of two groups of mice was assessed by using staining kit. (B) The protein levels of p21 and p16 were detected by immunoblot. (C) Quantitative analysis for protein levels of p21 and p16. (D) IHC staining of p21 in testis tissue. p<0.01. ## Discussion Age-related decline in testosterone leads to testosterone deficiency in males, which can seriously affect male fertility and quality of life [1, 2]. Elucidating mechanisms of aging and developing interventions are vital to reduced age-associated damage and functional decline. Aging is an irreversible natural process in human life which is influenced by many exogenous factors, such as living environment and diseases [38]. Obesity is closely related to aging and accelerates the development of age-related diseases [20]. It is unclear whether obesity promotes age-related decline in testosterone via accelerating Leydig cell aging. In this study, we demonstrated that obesity was an important factor to promote Leydig cell aging. p38 MAPK was involved in Leydig cell aging and age- and obesity-related decline in testosterone. In present study, we found reduced serum testosterone in 18-month-old mice, illustrating a suitable model to explore the mechanisms under age-related decline in testosterone. In the adult testis, Leydig cell testosterone production depends upon the pulsatile secretion of LH by binding to LHCGR to maintain of optimal levels of steroidogenic enzymes and to mobilize and transport of cholesterol into the inner mitochondrial membrane via StAR. In this study, we found that the expressions of nearly all proteins associated testosterone synthesis were reduced, without changes in LH levels in aged mice. The above results were consistent with previous studies that age-related changes in Leydig cell steroidogenesis occurred at the gonadal level rather than secondary to hypothalamic-pituitary changes [39]. Despite cellular functional decline is the characteristic of cell aging, the aging status of Leydig cell in aged males is not directly evaluated. In this study, cell senescence markers, represented as expression of p21 and p16 protein and staining for SA-β-Gal were evaluated. The SA-β-Gal staining was present in testicular interstitium in aged mice. The level of testicular p21 protein was increased, and IHC showed that increased p21 focused on Leydig cells in aged mice. The above results suggested that Leydig cells were the prime target for cell aging in testes. And Leydig cell senescence was accompanied by resultant reduced testosterone synthesis. To explore molecular related to Leydig cell aging, we obtained and further enriched the DEGs between Leydig cells from single-cell RNA sequencing results of young and old human testes. Both KEGG and GO terms enriched in MAPK pathway, a family of serine/threonine kinases consisted of three family members: JNK, ERK$\frac{1}{2}$, and p38 MAPK [32]. In the present study, p38 MAPK, rather than JNK, ERK$\frac{1}{2}$ was activated in aged Leydig cells. The JNK and ERK$\frac{1}{2}$, together with p38 MAPK belong to the MAPK family. ERK$\frac{1}{2}$ is activated by growth factors and cytokines and plays a central role in the control of cell proliferation and differentiation. In our study, there was no change in the expression of the p-ERK$\frac{1}{2}$ in testis of aged mice, which was compatible with previous perspectives of ERK$\frac{1}{2.}$ The JNK and p38 MAPK have similar function and are known as stress-activated protein kinase and are strongly activated by various environmental stresses and inflammatory cytokines [40]. Cellular Cell aging is referred to as stress-induced premature aging. Thus, the activation of JNK and p38 MAPK is closely associated with cell aging. In addition, in previous studies, when exposed to external harmful stimuli, p38 MAPK is majorly activated in Leydig cells (41–44), which may be due to cell type-specific effects. All these implied that p38 MAPK might be important regulator of Leydig cell aging. Obesity is a well-recognized risk factor for cell aging [20]. Clinical studies suggested that obesity could accelerate age-related decline in testosterone, but the underly mechanisms remain unclear. In our previous study, we demonstrated that obesity impaired testosterone synthesis in Leydig cells [29]. In the present study, we found that the levels of p21 protein and the intensity of SA-β-Gal staining were higher in Leydig cells of obese mice, suggesting that obesity promoted Leydig cell premature aging. At the same time, we found that p38 MAPK was activated in Leydig cells of obese mice. This result indicated that p38 MAPK might be the key molecule linked obesity and Leydig cell aging. To test these hypothesizes, a mouse model with p38 MAPK knockout in Leydig cells was established. In aged p38LCKO mice, Leydig cells underwent less senescence compared with aged WT mice. Accompanied by the remission of cell aging, the levels of testosterone and steroidogenic enzymes were increased. The resulted confirmed that reduced testosterone synthesis of aged Leydig cells was the consequence of cell aging, and p38 MAPK-mediated Leydig cell aging is an intrinsic mechanism of age-related decline in testosterone. It has been reported that p38 MAPK may promote cell aging the following points. p38 MAPK facilitates the transcription of p21 and p16, enhances the transcription of senescence-associated secretory phenotype (SASP) genes including IL-6, IL-8, and GM-CSF, and inhibits senescent cell apoptosis for pro-survival (45–48). In present study, the level of p21, not p16 protein was increased in aged Leydig cells. Meanwhile, KEGG analysis revealed p53 signaling pathway was activated in aged human Leydig cells. Based on these, we speculated that p38 MAPK promoted Leydig cell aging by p53/p21 signaling pathway. Future work could elucidate the detailed mechanisms. It well known that both oxidative stress and inflammatory factors can activate p38 MAPK pathways to provoke cell adaptive response. Oxidative stress and inflammation are also the main factors inducing cell senescence. Aged Leydig cells produced significantly more reactive oxygen species (ROS) and had reduced expressions of key enzymatic and non-enzymatic antioxidants, leading cell oxidative stress (49–51). Our results showed that the levels of oxidative stress and pro-inflammatory factors were elevated in testis of aged mice, which suggested that inflammation and oxidative stress might be involved in the activation of p38 MAPK during Leydig cell aging. Oxidative stress and inflammation are important mechanisms in the pathogenesis of obesity. In our previous studies, elevated levels of oxidative stress and inflammation in testis and Leydig cells of diet-induced obese mice [29, 52], suggesting that obesity promoted Leydig cell premature aging via enhanced oxidative stress and inflammation. The results from aged and obese mice implied that oxidative stress and inflammation are common factors of Leydig cell aging caused by age and obesity by activating the p38 MAPK pathway. As previously mentioned, obesity accelerates aging by shortening telomere length and increasing epigenetic aging and DNA damage (53–55). Whether obesity accelerates Leydig cell aging through the above mechanisms requires further research to confirm. In generally, we demonstrated that p38 MAPK was involved in Leydig cell aging and aged-related decline in testosterone. Obesity was an important risk factor for aged-related decline in testosterone by promoting Leydig cell aging via activating p38 MAPK. The results provide a new insight to explore the mechanisms and look for intervention of age-related decline in testosterone. ## 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 animal study was reviewed and approved by Animal Ethics Committee of Shandong Provincial Hospital. ## Author contributions DL conducted most of the experiments and wrote the manuscript. XX analysed data. XQ and LY performed animal husbandry. CY and QG conceived of and designed the study. 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.2023.1088249/full#supplementary-material ## References 1. Harman SM, Metter EJ, Tobin JD, Pearson J. Longitudinal effects of aging on serum total and free testosterone levels in healthy men. J Clin Endocrinol Metab (2001) 86. DOI: 10.1210/jcem.86.2.7219 2. Nieschlag E. Late-onset hypogonadism: A concept comes of age. Andrology (2020) 8. DOI: 10.1111/andr.12719 3. Kim EK, Choi EJ. Compromised MAPK signaling in human diseases: An update. Arch Toxicol (2015) 89. DOI: 10.1007/s00204-015-1472-2 4. Yabluchanskiy A, Tsitouras PD. Is testosterone replacement therapy in older men effective and safe. Drugs Aging (2019) 36. DOI: 10.1007/s40266-019-00716-2 5. Dos Santos MR, Bhasin S. Benefits and risks of testosterone treatment in men with age-related decline in testosterone. Annu Rev Med (2021) 72 75-91. DOI: 10.1146/annurev-med-050219-034711 6. Cole AP, Hanske J, Jiang W, Kwon NK, Lipsitz SR, Kathrins M. Impact of testosterone replacement therapy on thromboembolism, heart disease and obstructive sleep apnoea in men. BJU Int (2018) 121. DOI: 10.1111/bju.14149 7. Sharma R, Oni OA, Gupta K, Chen G, Sharma M, Dawn B. Normalization of testosterone level is associated with reduced incidence of myocardial infarction and mortality in men. Eur Heart J (2015) 36. DOI: 10.1093/eurheartj/ehv346 8. Baillargeon J, Urban RJ, Kuo YF, Ottenbacher KJ, Raji MA, Du F. Risk of myocardial infarction in older men receiving testosterone therapy. Ann Pharmacother (2014) 48. DOI: 10.1177/1060028014539918 9. Finkle WD, Greenland S, Ridgeway GK, Adams JL, Frasco MA, Cook MB. Increased risk of non-fatal myocardial infarction following testosterone therapy prescription in men. PloS One (2014) 9 1-7. DOI: 10.1371/journal.pone.0085805 10. Papazian L, Forel J-M, Gacouin A, Penot-Ragon C, Perrin G, Loundou A. Adverse events associated with testosterone administration. N Engl J Med (2010) 363. DOI: 10.1056/NEJMoa1000485 11. Vigen R, O’Donnell CI, Barón AE, Grunwald GK, Maddox TM, Bradley SM. Association of testosterone therapy with mortality, myocardial infarction, and stroke in men with low testosterone levels. Jama (2013) 310. DOI: 10.1001/jama.2013.280386 12. Etminan M, Skeldon SC, Goldenberg SL, Carleton B, Brophy JM. Testosterone therapy and risk of myocardial infarction: A pharmacoepidemiologic study. Pharmacotherapy (2015) 35. DOI: 10.1002/phar.1534 13. Warburton D, Hobaugh C, Wang G, Lin H, Wang R. Testosterone replacement therapy and the risk of prostate cancer. Asian J Androl (2015) 17. DOI: 10.4103/1008-682X.150841 14. Cui Y, Zong H, Yan H, Zhang Y. The effect of testosterone replacement therapy on prostate cancer: A systematic review and meta-analysis. Prostate Cancer Prostatic Dis (2014) 17. DOI: 10.1038/pcan.2013.60 15. Saad F, Caliber M, Doros G, Haider KS, Haider A. Long-term treatment with testosterone undecanoate injections in men with hypogonadism alleviates erectile dysfunction and reduces risk of major adverse cardiovascular events, prostate cancer, and mortality. Aging Male (2020) 23 81-92. DOI: 10.1080/13685538.2019.1575354 16. Santella C, Renoux C, Yin H, Yu OHY, Azoulay L. Testosterone replacement therapy and the risk of prostate cancer in men with late-onset hypogonadism. Am J Epidemiol (2019) 188. DOI: 10.1093/aje/kwz138 17. Pilleron S, Sarfati D, Janssen-Heijnen M, Vignat J, Ferlay J, Bray F. Global cancer incidence in older adults, 2012 and 2035: A population-based study. Int J Cancer (2019) 144 49-58. DOI: 10.1002/ijc.31664 18. Chuang YF, An Y, Bilgel M, Wong DF, Troncoso JC, O'Brien RJ. Midlife adiposity predicts earlier onset of alzheimer’s dementia, neuropathology and presymptomatic cerebral amyloid accumulation. Mol Psychiatry (2016) 21. DOI: 10.1038/mp.2015.129 19. Koliaki C, Liatis S, Kokkinos A. Obesity and cardiovascular disease: Revisiting an old relationship. Metabolism (2019) 92 98-107. DOI: 10.1016/j.metabol.2018.10.011 20. Tam BT, Morais JA, Santosa S. Obesity and ageing: Two sides of the same coin. Obes Rev (2020) 21 1-21. DOI: 10.1111/obr.12991 21. Tajar A, Forti G, O’Neill TW, Lee DM, Silman AJ, Finn JD. Characteristics of secondary, primary, and compensated hypogonadism in aging men: Evidence from the European male ageing study. J Clin Endocrinol Metab (2010) 95. DOI: 10.1210/jc.2009-1796 22. Cao J, Chen TM, Hao WJ, Li J, Liu L, Zhu BP. Correlation between sex hormone levels and obesity in the elderly male. Aging Male (2012) 15. DOI: 10.3109/13685538.2012.666585 23. Mancini M, Pecori Giraldi F, Andreassi A, Mantellassi G, Salvioni M, Berra CC. Obesity is strongly associated with low testosterone and reduced penis growth during development. J Clin Endocrinol Metab (2022) 106. DOI: 10.1210/clinem/dgab535 24. Kurniawan LB, Adnan E, Mulyono B. Insulin resistance and testosterone level in Indonesian young adult males. Rom J Intern Med (2020) 58. DOI: 10.2478/rjim-2020-0004 25. Wu FCW, Tajar A, Pye SR, Silman AJ, Finn JD, O'Neill TW. Hypothalamic-pituitary-testicular axis disruptions in older men are differentially linked to age and modifiable risk factors: The European male aging study. J Clin Endocrinol Metab (2008) 93. DOI: 10.1210/jc.2007-1972 26. Iacona R, Bonomo V, Di Piazza M, Sansone A, Usala M, Novo S. Five-year prospective study on cardiovascular events, in patients with erectile dysfunction and hypotestosterone. Arch Ital di Urol e Androl (2017) 89. DOI: 10.4081/aiua.2017.4.313 27. Camacho EM, Huhtaniemi IT, O’Neill TW, Finn JD, Pye SR, Lee DM. Age-associated changes in hypothalamic-pituitary-testicular function in middle-aged and older men are modified by weight change and lifestyle factors: longitudinal results from the European Male ageing study. Eur J Endocrinol (2013) 168. DOI: 10.1530/EJE-12-0890 28. Chen H, Hardy MP, Huhtaniemi I, Zirkin BR. Age-related decreased leydig cell testosterone production in the brown Norway rat. J Androl (1994) 15 29. Su X, Lin D, Luo D, Sun M, Wang X, Ye J. Cyclophilin d participates in the inhibitory effect of high-fat diet on the expression of steroidogenic acute regulatory protein. J Cell Mol Med (2019) 23. DOI: 10.1111/jcmm.14569 30. Luo L, Chen H, Zirkin BR. Leydig cell aging: Steroidogenic acute regulatory protein (StAR) and cholesterol side-chain cleavage enzyme. J Androl (2001) 22. DOI: 10.1002/j.1939-4640.2001.tb02165.x 31. Chen H, Irizarry RA, Luo L, Zirkin BR. Leydig cell gene expression: Effects of age and caloric restriction. Exp Gerontol (2004) 39 31-43. DOI: 10.1016/j.exger.2003.09.021 32. Kyriakis JM, Avruch J. Mammalian MAPK signal transduction pathways activated by stress and inflammation: A 10-year update. Physiol Rev (2012) 92 689-737. DOI: 10.1152/physrev.00028.2011 33. Yang W, Zhu L, Lai S, Ding Q, Xu T, Guo R. Cimifugin ameliorates lipotoxicity-induced hepatocyte damage and steatosis through TLR4/p38 MAPK- and SIRT1-involved pathways. Oxid Med Cell Longev (2022) 2022. DOI: 10.1155/2022/4557532 34. He D, Wu H, Xiang J, Ruan X, Peng P, Ruan Y. Gut stem cell aging is driven by mTORC1. Nat Commun (2020) 11 1-13. DOI: 10.1038/s41467-019-13911-x 35. Zhao Y, Liu X, Qu Y, Wang L, Geng D, Chen W. The roles of p38 MAPK → COX2 and NF-κB → COX2 signal pathways in age-related testosterone reduction. Sci Rep (2019) 9 1-11. DOI: 10.1038/s41598-019-46794-5 36. Bridges PJ, Koo Y, Kang DW, Hudgins-Spivey S, Lan ZJ, Xu X. Generation of Cyp17iCre transgenic mice and their application to conditionally delete estrogen receptor alpha (Esr1) from the ovary and testis. Genesis (2008) 46 499-505. DOI: 10.1002/dvg.20428 37. Engel FB, Schebesta M, Duong MT, Lu G, Ren S, Madwed JB. p38 MAP kinase inhibition enables proliferation of adult mammalian cardiomyocytes. Genes Dev (2005) 19. DOI: 10.1101/gad.1306705 38. López-Otín C, Blasco MA, Partridge L, Serrano M, Kroemer G. The hallmarks of aging. Cell (2013) 153. DOI: 10.1016/j.cell.2013.05.039 39. Wang Y, Chen F, Ye L, Zirkin B, Chen H. Steroidogenesis in leydig cells: Effects of aging and environmental factors. Reproduction (2017) 154. DOI: 10.1530/REP-17-0064 40. Cargnello M, Roux PP. Activation and function of the MAPKs and their substrates, the MAPK-activated protein kinases. Microbiol Mol Biol Rev (2011) 75 50-83. DOI: 10.1128/mmbr.00031-10 41. Chen Y, Wang J, Chen X, Li D, Han X. Microcystin-leucine arginine mediates apoptosis and engulfment of leydig cell by testicular macrophages resulting in reduced serum testosterone levels. Aquat Toxicol (2018) 199. DOI: 10.1016/j.aquatox.2018.03.018 42. Xiong X, Zhang L, Fan M, Han L, Wu Q, Liu S. β-endorphin induction by psychological stress promotes leydig cell apoptosis through p38 MAPK pathway in Male rats. Cells (2019) 8 1265. DOI: 10.3390/cells8101265 43. Kang FC, Wang SC, So EC, Chang MM, Wong KL, Cheng KS. Propofol may increase caspase and MAPK pathways, and suppress the akt pathway to induce apoptosis in MA−10 mouse leydig tumor cells. Oncol Rep (2019) 41. DOI: 10.3892/or.2019.7129 44. Liu J, Chang F, Li F, Fu H, Wang J, Zhang S. Palmitate promotes autophagy and apoptosis through ROS-dependent JNK and p38 MAPK. Biochem Biophys Res Commun (2015) 463. DOI: 10.1016/j.bbrc.2015.05.042 45. Xu Y, Li N, Xiang R, Sun P. Emerging roles of the p38 MAPK and PI3K/AKT/mTOR pathways in. Trends Biochem Sci (2014) 39. DOI: 10.1016/j.tibs.2014.04.004.Emerging 46. Wang W, Pan K, Chen Y, Huang C, Zhang X. The acetylation of transcription factor HBP1 by p300/CBP enhances p16 INK4A expression. Nucleic Acids Res (2012) 40. DOI: 10.1093/nar/gkr818 47. Alimbetov D, Davis T, Brook AJ, Cox LS, Faragher RG, Nurgozhin T. Suppression of the senescence-associated secretory phenotype (SASP) in human fibroblasts using small molecule inhibitors of p38 MAP kinase and MK2. Biogerontology (2016) 17. DOI: 10.1007/s10522-015-9610-z 48. Slobodnyuk K, Radic N, Ivanova S, Llado A, Trempolec N, Zorzano A. Autophagy-induced senescence is regulated by p38α signaling. Cell Death Dis (2019) 10 376. DOI: 10.1038/s41419-019-1607-0 49. Finkel T, Holbrook NJ. Oxidants, oxidative stress and the biology of ageing. Nature (2000) 408. DOI: 10.1038/35041687 50. Chen H, Cangello D, Benson S, Folmer J, Zhu H, Trush MA. Age-related increase in mitochondrial superoxide generation in the testosterone-producing cells of brown Norway rat testes: Relationship to reduced steroidogenic function. Exp Gerontol (2001) 36. DOI: 10.1016/S0531-5565(01)00118-8 51. Cao L, Leers-Sucheta S, Azhar S. Aging alters the functional expression of enzymatic and non-enzymatic anti-oxidant defense systems in testicular rat leydig cells. J Steroid Biochem Mol Biol (2004) 88. DOI: 10.1016/j.jsbmb.2003.10.007 52. Ye J, Luo D, Xu X, Sun M, Su X, Tian Z. Metformin improves fertility in obese males by alleviating oxidative stress-induced blood-testis barrier damage. Oxid Med Cell Longev (2019) 537. DOI: 10.1155/2019/9151067 53. Kim S, Parks CG, DeRoo LA, Chen H, Taylor JA, Cawthon RM. Obesity and weight gain in adulthood and telomere length. Cancer Epidemiol Biomarkers Prev (2009) 18. DOI: 10.1158/1055-9965.EPI-08-0935 54. Horvath S, Erhart W, Brosch M, Ammerpohl O, von Schönfels W, Ahrens M. Obesity accelerates epigenetic aging of human liver. Proc Natl Acad Sci U.S.A. (2014) 111. DOI: 10.1073/pnas.1412759111 55. Schumacher B, Pothof J, Vijg J, Hoeijmakers JHJ. The central role of DNA damage in the ageing process. Nature (2021) 592 695-703. DOI: 10.1038/s41586-021-03307-7
--- title: Key genes associated with non-alcoholic fatty liver disease and hepatocellular carcinoma with metabolic risk factors authors: - Fan Yang - Beibei Ni - Qinghai Lian - Xiusheng Qiu - Yizhan He - Qi Zhang - Xiaoguang Zou - Fangping He - Wenjie Chen journal: Frontiers in Genetics year: 2023 pmcid: PMC10025510 doi: 10.3389/fgene.2023.1066410 license: CC BY 4.0 --- # Key genes associated with non-alcoholic fatty liver disease and hepatocellular carcinoma with metabolic risk factors ## Abstract Background: Hepatocellular carcinoma (HCC) has become the world’s primary cause of cancer death. Obesity, hyperglycemia, and dyslipidemia are all illnesses that are part of the metabolic syndrome. In recent years, this risk factor has become increasingly recognized as a contributing factor to HCC. Around the world, non-alcoholic fatty liver disease (NAFLD) is on the rise, especially in western countries. In the past, the exact pathogenesis of NAFLD that progressed to metabolic risk factors (MFRs)-associated HCC has not been fully understood. Methods: Two groups of the GEO dataset (including normal/NAFLD and HCC with MFRs) were used to analyze differential expression. Differentially expressed genes of HCC were verified by overlapping in TCGA. In addition, functional enrichment analysis, modular analysis, Receiver Operating Characteristic (ROC) analysis, LASSO analysis, and Genes with key survival characteristics were analyzed. Results: We identified six hub genes (FABP5, SCD, CCL20, AGPAT9(GPAT3), PLIN1, and IL1RN) that may be closely related to NAFLD and HCC with MFRs. We constructed survival and prognosis gene markers based on FABP5, CCL20, AGPAT9(GPAT3), PLIN1, and IL1RN.This gene signature has shown good diagnostic accuracy in both NAFLD and HCC and in predicting HCC overall survival rates. Conclusion: As a result of the findings of this study, there is some guiding significance for the diagnosis and treatment of liver disease associated with NAFLD progression. ## 1 Introduction Hepatocellular carcinoma (HCC) accounts for about $90\%$ of all liver cancers, making it the second leading cause of cancer death worldwide (Llovet et al., 2021, M. A; Morse et al., 2019). The development of HCC usually follows a background of chronic low-grade inflammation characterised by chronic liver damage followed by inflammation, hepatocellular necrosis, and regeneration. HCCs are predominantly caused by Hepatitis B (HBV), Hepatitis C (HCV) infection, alcohol consumption, as well as metabolic perturbations leading to non-alcoholic fatty liver disease (NAFLD) (R. E. Ericksen et al., 2019, C; Trierweiler et al., 2016). Due to universal vaccination and antiviral therapy, viral HCC prevalence is decreasing. It will be necessary to modify strategies for cancer prevention, prediction, and surveillance for HCC (Awosika and Sohal, 2022, S. F; Huang et al., 2018). The metabolic syndrome refers to a group of disorders that include dyslipidemia, hyperglycemia, and obesity, has received increasing attention as a novel risk factor for HCC (K. Akahoshi et al., 2016). There was a greater than 100-fold increase in risk for HBV or HCV carriers who also had diabetes or obesity, which suggests synergistic effects of metabolic factors and hepatitis (C. L. Chen et al., 2008). Patients with metabolic risk factors (MRFs) may be at greater risk of developing hepatocarcinogenesis when FABP4 is overexpressed in HSCs (N. Chiyonobu et al., 2018). According to statistics, 25 per cent of the population worldwide suffers from NAFLD (A. Lonardo et al., 2016). NAFLD is becoming more prevalent worldwide, especially in western countries (Mancina et al., 2016). Consequently, NAFLD has become an economic and health concern worldwide. There is no doubt that NAFLD is a hepatic manifestation of metabolic syndrome (MS) and is often associated with dyslipidemia, obesity, and T2DM (J. Wattacheril, 2020). As a result of liver lipid accumulation, NAFLD can cause inflammation and damage to the hepatocytes (J. Wattacheril, 2020). The liver biopsy usually shows milder forms (steatosis) to severe conditions (non-alcoholic steatohepatitis (NASH), advanced fibrosis, cirrhosis) (Pouwels et al., 2022). In the past, the exact pathogenesis of NAFLD that progressed to MFRs-associated HCC has not been fully understood. High-throughput gene chips and transcriptome sequencing have entirely changed the previous systematic analysis methods for disease research (M. Bustoros et al., 2020). RNA sequencing and high-throughput microarrays help to identify reliable biological markers, classify diseases, and reveal mechanisms of disease development. The discovery of new biomarkers can be helpful in predicting risk and determining which treatment is most suitable for an individual patient. Thus, the prediction of candidate genes may also be based on NAFLD-HCC with MRFs pathogenesis. This study aims to identify the key genes involved in NAFLD and HCC with MRFs and to provide a reference for further study of the transformation of MFRs-associated HCC and a molecular-targeted approach to cancer treatment. In this study, we analyzed microarray data comprehensively, selecting normal tissues and NAFLD samples and microarray data of MFRs-associated HCC and adjacent normal tissues, and separately analysed the differentially expressed genes (DEGs) in both groups of chips. Combining the GEO DEG data of human HCC with MFRs and normal liver tissue with chip data to determine key DEGs that directly affect the diagnosis and treatment of NAFLD. Afterwards, further functional enrichment analysis was conducted to determine how DEGs regulate the main biological functions. Furthermore, by using protein-protein interaction (PPI) networks and survival analysis of patient data, key genes are identified that affect the diagnosis, treatment, and prognosis of patients with NAFLD. ## 2.1 Profiles of gene expression GSE63067, GSE89632, and GSE102079 datasets were downloaded from Gene Expression Omnibus (GEO), an open-access database that provides gene expression profiles. GSE63067 (Frades et al., 2015) and GSE102079 (N. Chiyonobu, S. Shimada, Y. Akiyama, K. Mogushi, M. Itoh, K. Akahoshi, S. Matsumura, K. Ogawa, H. Ono, Y. Mitsunori, D. Ban, A. Kudo, S. Arii, T. Suganami, S. Yamaoka, Y. Ogawa, M. Tanabe and S. Tanaka, 2018) are both based on the GPL570 [(HG-U133_Plus_2) Affymetrix Human Genome U133 Plus 2.0 Array]. GSE89632 (B. M. Arendt et al., 2015) is based on [(GPL14951) Illumina HumanHT-12 WG-DASL V4.0 R2 expression bead chip]. The title of the GSE63067 data set is “Expression data from human non-alcoholic fatty liver disease stages”. The data contained the gene expression profiles of 11 NAFLD patients and seven non-NAFLD controls. The title of the GSE102079 data set is “FABP4 overexpressed in intratumoral hepatic stellate cells within hepatocellular carcinoma with metabolic risk factors”. Between 2006 and 2011, 152 patients who underwent curative hepatic resection for HCC at Tokyo Medical and Dental University Hospital participated in an integrated gene expression microarray study. In the control group, 14 adjacent liver tissues were obtained from patients with metastases of colorectal cancer without chemotherapy. The validation data set was from GSE89632 and The Cancer Genome Atlas (TCGA) data set. The title of the GSE89632 data set is “Genome-wide analysis of hepatic gene expression in patients with non-alcoholic fatty liver disease and healthy donors with hepatic fatty acid composition and other nutritional factors”. A cross-sectional study included 20 patients with simple steatosis (SS), 19 non-alcoholic steatohepatitis (NASH), and 24 healthy liver donors. The TCGA database of liver hepatocellular carcinoma (LIHC) contains RNA-Seq data for 374 HCC patients and 50 normal tissues (https://portal.gdc.cancer.gov/) for gene expression and immune system infiltrates. ## 2.2 Analysis of differentially expressed genes (DEGs) in NAFLD and HCC with MRFs A comparison of DEGs between NAFLD and normal controls, HCC patients with MRFs, and corresponding controls was performed using the limma R package“complexheatmap” and “ggplot2” to generate heat maps and volcano maps, respectively, which is an efficient analysis method in bioinformatics (M. E. Ritchie et al., 2015). In NAFLD datasets, the selected criteria were p-value <0.05 and \|log2FC\|>1. In HCC datasets, the selected criteria were p-value <0.05 and \|log2FC\|>1. Additionally, the overlapping DEGs between NAFLD and HCC with MRFs were determined by Venn diagrams using the Venn platform (http://bioinformatics.psb.ugent.be/webtools/Venn/). A subsequent analysis was performed on these overlapping DEGs. ## 2.3 Functional classification and pathway enrichment for DEGs GO function enrichment analyses were conducted on the above overlapping DEGs. It consisted of biological process (BP), cellular component (CC), and molecular function (MF) [2006]. The analysis of KEGG signaling pathway enrichment using a package called “clusterProfiler” (M. Kanehisa et al., 2016). GO terms and KEGG pathways enriched with adjusted p-value of 0.05 were selected for analysis. ## 2.4 Establishment of protein-protein interactions and identification of hub genes In order to further investigate the interactions between the above-mentioned common genes, a search tool called the Search Tool for Retrieval of Interacting Genes (STRING) has been developed for PPI network construction (D. Szklarczyk et al., 2015). Interaction scores of at least 0.4 were considered significant. Subsequently, PPI network visualisation was conducted with Cytoscape software. Then, the Maximal Clique Centrality (MCC), Density of Maximum Neighborhood Component (DMNC), Maximum Neighborhood Component (MNC), Degree, and Edge Percolated Component (EPC), algorithms in the cytoHubba plug-in (http://hub.iis.sinica.edu.tw/cytohubba/) was applied to identify PPI hub genes with high connectivity. ## 2.5 Comparing the hub gene expression degree and analysing the prognosis Based on the TCGA database, the six hub genes expression in HCC normal tissues and tumor tissues was investigated. In LIHC, 374 HCC specimens with normal adjacent tissues and HCC tissue (50 each) were compared with neighbouring normal tissues. GEPIA was used to investigate the prognostic significance of hub genes (http://gepia.cancer-pku.cn/index.html) (Z. Tang et al., 2017). Survival analyses were considered significant when log-rank $p \leq 0.05$ was used. ## 2.6 Developing signatures and evaluating their reliability Based on the training dataset, hub genes associated with prognosis were identified and assessed against other datasets for their predictive performance. Half of TCGA is set as the training set. The other half of TCGA is set as a validation set. The entire TCGA cohort is a verification set. Using univariate Cox proportional hazard regression analysis, it was evaluated whether hub genes are associated with overall survival (OS) in the training set. In the “glmnet” package, the Latent Selection Operator penalised Cox proportional hazard regression using Cox proportional hazards models. A prediction formula for gene characteristics was devised. The formula for the model is as follows: risk score = gene1×β1 (gene one expression level) + gene 2×β2 (gene two expression level) +…gene n×βn (gene n expression level). In this formula, genes are combined with gene expression values and regression coefficients from multiple Cox proportional hazards regression models (George et al., 2014; Zhang J. et al., 2021). Using the Kaplan–Meier (K–M) survival curves, survival comparisons were performed between low- and high-risk groups via the R package “survival”. Furthermore, a time-dependent receiver operating characteristic (ROC) analysis (including 1-, 3-, and 5-year survival) was conducted to evaluate hub gene sensitivity and specificity using the R package “survival ROC” (P. J. Heagerty et al., 2000). It is critical to consider the area of the AUC curve when trying to predict clinical outcomes. Prognosis is better when AUC >0.5; the closer AUC is to 1, the better. ## 2.7 The expression of hub genes is correlated with the presence of immune cells in tumor Tumor contains a large number of immune cells, and the prognosis of high-grade HCC patients with high subtype of dominant immunity is obviously better (Y. Kurebayashi et al., 2018). To examine whether the expression of hub genes is correlated with the presence of immune cells in HCC, we examined the correlation between hub gene mRNA expression and tumor-infiltrating immune cells. The web tool TIMER was used (https://cistrome.shinyapps.io/timer/) (T. Li et al., 2017). Six tumor-infiltrating cell subsets were analysed, such as B cells, CD8+ T cells, CD4+ T cells, macrophages, neutrophils, and dendritic cells. ## 2.8 Statistical analysis ROC curves for hub genes were constructed using the pROC package (X. Robin et al., 2011). To measure the effectiveness of the model, we calculated the area under the curves (AUC). These results showed the usefulness of genes for diagnostic purposes. p-values <0.05 were considered statistically significant. ## 3.1 Identification of DEGs in NAFLD and HCC The series GSE63067 dataset about NAFLD and the series GSE102079 dataset about HCC from the NCBI GEO database was downloaded. Based on a p-value of 0.05 and \|log2FC \| of >1.0, 125 DEGs were identified in GSE63067, and 726 DEGs were identified in GSE102079 using the “limma” package in R software. Volcano plots and heatmaps were used to visualise the DEGs of the two data sets shown in Figures 1A,B and Figures 2A,B, respectively. Using the Venn Diagram online tool, 26 common genes of two diseases were identified and are shown in Figure 1C. FIGURE 1: Differentially expressed genes (DEGs) shown in a volcano plot and Venn diagram (A) An analysis of the differential genes in GSE63067 using a volcano map. (B) An analysis of the differential genes in GSE102079 using a volcano map. (C) Venn diagram of DEGs in GSE63067 and GSE102079 data sets. Abbreviations: DEGs, differentially expressed genes; NAFLD, non-alcoholic fatty liver; HCC, Hepatocellular Carcinoma, MFRs, metabolic risk factors. FIGURE 2: Differentially expressed genes heatmaps (A) Heat map of DEGs in GSE63067 (NAFLD) and (B) Heatmap of DEGs in GSE102079(HCC with MFRs). DEGs in red indicate upregulation, DEGs in blue indicate downregulation, and DEGs in white indicate no significant changes. Abbreviations: NAFLD, non-alcoholic fatty liver; HCC, Hepatocellular Carcinoma. ## 3.2 Analysis of pathways and functional roles associated with overlapping DEGs Functional enrichment and KEGG pathway analyses of 26 common NAFLD and HCC genes were performed at a threshold of p-value <0.05. The results showed that DEGs were enriched in biological processes, including cellular response to environmental stimuli, cellular response to abiotic stimuli, cellular response to ionising radiation, unsaturated fatty acid biosynthetic process, and response to zinc ions (Figure 3A). Regarding molecular function, DEGs were principally associated with receptor ligand activity, cytokine activity, monocarboxylic acid binding, and fatty acid binding Figure 3B. The KEGG pathways of DEGs were enriched in the PPAR signalling pathway (Figure 3C). FIGURE 3: Analyses of functional enrichment between two groups of DEGs. (A) Enrichment results for GO biological processes; (B) Enrichment results for GO molecular function processes; (C) Enrichment results for KEGG pathways A bubble’s size represents the number of genes associated with each term. A term’s bubble size represents how many genes are associated with it. Each bubble’s color indicates the adjusted p-value abbreviations: GO, Gene Ontology; BP, biological process; MF, molecular function; KEGG, Kyoto Encyclopedia of Genes and Genomes. ## 3.3 Analysing the PPI network and selecting hub genes The PPI network was first performed based on the STRING database to investigate how DEGs interact with one another. Afterwards, the results were imported into Cytoscape to be analysed (Figure 4A). A Cytoscape plug-in, Cytohubba, was used to analyse the PPI network and identify hub genes. We got top 10 genes from protein-protein network ranked by five different algorithms of cytohubba including MCC, DMNC, MNC, Degree, and EPC (Table1). In this study, the genes with the top six values were considered as hub genes. Based on the five algorithms, the top six genes were determined to be hub genes: FABP5, SCD, CCL20, AGPAT9, PLIN1, IL1RN (Figure 4B). FIGURE 4: The PPI network analysis highlights the most significant modules related to DEGs. (A) 26 DEGs were used in the construction of this PPI network. (B) The most significant module of the PPI network includes 6 hub genes(yellow circles). DEGs differentially expressed genes; PPI, Protein-Protein interaction. TABLE_PLACEHOLDER:TABLE 1 ## 3.4 The diagnostic value of hub genes has been validated To evaluate the diagnostic value of the top six hub genes obtained from the above analysis, ROC curves were constructed and their corresponding area under the curve (AUC) was calculated. Figure 5A shows the result of NAFLD. The AUC for FABP5, SCD, CCL20, AGPAT9, PLIN1, and IL1RN in NAFLD patients and normal controls were 0.828, 0.818, 0.883, 0.857, 0.961, 0.818 at NAFLD GSE63067 dataset. Figure 5B shows the ROC curves in HCC patients and normal controls. The AUC for FABP5, SCD, CCL20, AGPAT9, PLIN1, and IL1RN in HCC and the normal controls were 0.717, 0.785, 0.776, 0.821, 0.901, 0.839 at HCC with MRFs GSE102079 dataset. For validation, The AUC for FABP5, SCD, CCL20, AGPAT9, and IL1RN were 0.561, 0.746, 0.858, 0.981, 0.907 in NAFLD based on GSE89632 (Figure 5C). The AUC for FABP5, SCD, CCL20, AGPAT9, PLIN1, and IL1RN in HCC and the normal controls were 0.900, 0.614, 0.807, 0.589, 0.725, 0.769 at HCC TGCA dataset Figure 5D). FIGURE 5: The diagnostic value of the top six hub genes with ROC curves in NAFLD and HCC. (A) The diagnostic value of the top six hub genes on ROC curves in NAFLD is based on the GSE63067 data set. (B) The diagnostic value of the top six hub genes on ROC curves in HCC with metabolic risk factors is based on the GSE102079 data set. (C) The diagnostic value of the top six hub genes on ROC curves in NAFLD is based on GSE89632 for validation. (D) The diagnostic value of the top six hub genes on ROC curves in HCC is based on TCGA data. Abbreviations: TPR, True Positive Rate: FPR, False Positive Rate. ## 3.5 An evaluation of the expression patterns and survival analysis of six hub genes Consistent with the results in GEO datasets, the mRNA expression of AGPAT9, PLIN1, and IL1RN was significantly downregulated. At the same time, that of FABP5, SCD, and CCL20 were upregulated considerably in TCGA HCC compared with non-tumor tissues (Figures 6A–L). According to the GEPIA web tool, FABP5 and PLIN1 mRNA expression are significantly linked to overall survival (OS) (Figures 6M–Q). FIGURE 6: TCGA database, has been used to validate the expression patterns of six hub genes. (A–F) The different expressions of six hub genes in paired HCC and normal controls in TCGA-LIHC. (G–L) The different expressions of six hub genes between HCC and the normal group. (M–R) This Kaplan-Meier plot shows how the hub genes were significant prognostic factors. ## 3.6 An analysis of the correlation between hub gene expression levels The correlation of expression levels of hub genes was captured using GEPIA. An analysis of correlation was performed on any two of FABP5, SCD, CCL20, AGPAT9(GPAT3), PLIN1, and IL1RN and six hub genes. The above data indicate that upregulation of one of them will decrease the high expression of other genes. IL1RN and GPAT3 (Figure 7A), CCL20 and SCD (Figure 7B), CCL20 and FABP5 (Figure 7C), SCD and PLIN1 (Figure 7D), IL1RN and CCL20 (Figure 7E), and IL1RN and PLIN1 (Figure 7F) are all positively related to each other. This may indicate that there is a common transcription factor as well as epigenetic modifications controlling them all. FIGURE 7: A correlation analysis was conducted on six key genes. (A) IL1RN-GPAT3 (B) CCL20-SCD (C) CCL20-FABP5 (D) SCD-PLIN1 (E) IL1RN-CCL20 (F) IL1RN-PLIN1. ## 3.7 The construction of the hub gene prognostic signature To avoid overfitting by LASSO regression, lambda. Min was selected, resulting in a more accurate prediction rate. We used the multivariate Cox proportional hazards regression analysis. Five prognostic genes were developed, including ABP5, CCL20, GPAT3, PLIN1, and IL1RNIn order to calculate the risk score for each patient, the following formula was used: risk score = (0.170124209 ×FABP5) + (0.073621309 ×CCL20) + (0.011005683 × GPAT3) + (−0.056212587 ×PLIN1)+ (−0.100588077 ×IL1RN). The LASSO coefficient for SCD is equal to 0. Hub gene risk scores were used to determine whether HCC patients were low-risk or high-risk (Figure 8A). A significantly worse OS was observed in high-risk patients compared to low-risk patients (Figure 8B, training set $$p \leq 0.008$$, validation set $$p \leq 0.0026$$, entire TCGA set $$p \leq 0.001$$). The reliability of hub genes was subsequently assessed using time-dependent ROC curves (Figure 8C). As a result, the area under the curve (AUC) was 0.738, 0.612, and 0.695 for 1-year, 3-year, and 5-year survival, respectively for the training set. The AUC was 0.611, 0.633, and 0.664 for 1-year, 3-year, and 5-year survival, respectively, for the validation set. These curves were also applied in the entire TCGA set. The AUC was 0.696, 0.634, and 0.673 for 1-year, 3-year, and 5-year survival, respectively. FIGURE 8: An analysis of the five-gene signature model in the TCGA cohort for prognosis. Half of TCGA is set as the training set. The other half of TCGA is designated as a validation set. The entire TCGA cohort is a verification set. (A) A comparison of risk score distribution, survival rates, and gene expression between patients in low- and high-risk groups in TCGA training set and TCGA validation set, entire TCGA cohort. (B) The Kaplan-Meier curves of OS for high-risk and low-risk groups in TCGA training set, TCGA validation set, and the entire TCGA cohort. (C) Time-dependent ROC curve AUCs from the TCGA training set, TCGA validation set, and entire TCGA cohort. ## 3.8 Correlation analysis of hub gene mRNA levels with tuours-infiltrating immune cells There are three kinds of cells in the tumor microenvironment: tumor cells, stromal cells, and immune cells that infiltrate the tumor. The TIMER web tool showed that the expression of all six hub genes was associated with infiltrating immune subsets, and the expression of NAFLD and HCC showed the most significant correlation with them. For B-cells, CD4+ T-cells, CD8+ T-cells, neutrophils, macrophages, and dendritic cells, the expression of FABP5 showed the most significant correlation with them (Figures 9A–F). FIGURE 9: Correlation analysis of hub gene mRNA levels with tumor-infiltrating immune cells. (A–F) The correlation of FABP5, SCD, CCL20, AGPAT9, PLIN1, IL1RN mRNA with tumor-infiltrating immune cells. TIMER is the database used for the data (https://cistrome.shinyapps.io/timer/). ## 4 Discussion In recent years, increasingly studies have confirmed the link between NAFLD and HCC. A higher risk of HCC has been associated with metabolic syndrome (Agosti et al., 2018, Y. P; Lin et al., 2022, Y; Tan et al., 2019). In the clinic, development and transformation of NAFLD are governed by common law, and its transformation process is also typical of HCC transformation. So far, the mechanism linking NAFLD and HCC remains unclear. Therefore, exploring the molecular mechanisms between NAFLD and other diseases and early identifying and intervening are likely to have significant clinical significance. Bioinformatics analyses comprehensively concentrate primarily on DEGs screening, the development of related protein interaction networks, the screening of genes, and the study of gene associations. In this study, through searching the datasets of NAFLD and HCC with MRFs from the GEO database, we found 26 common DEGs between these diseases. The results of GO enrichment analysis indicated that the DEGs were mainly enriched in receptor-ligand activity, cytokine activity, monocarboxylic acid binding, and fatty acid binding. Based on the KEGG pathway enrichment analysis results, overlapping differential genes are mainly involved in the PPAR signalling pathway. As members of the nuclear receptor superfamily, PPARs can regulate multiple metabolic pathways and are effective targets in the treatment of many metabolic disorders, including NAFLD (Wu et al., 2021). The PPAR signalling pathway is critical to the progression of non-alcoholic steatohepatitis (Zhang Y. et al., 2021). It is possible to predict HCC prognosis using the PPAR signaling pathway effectively, independently, and usefully (Xu et al., 2021). As a result of the PPI network and module analysis, we identified six key genes, including FABP5, SCD, CCL20, AGPAT9(GPAT3), PLIN1, and IL1RN. The six genes were all changed in both NAFLD patients and HCC patients with MRFs, suggesting that they may play an essential role in NAFLD and HCC with MRFs. An analysis of ROC curves was performed to validate the diagnostic value of NAFLD and HCC. This gene signature has shown good diagnostic accuracy in both NAFLD and HCC. The expression of FABP5 in NAFLD correlates with histological progression and the loss of hepatic fat during cirrhosis progression in NASH (K. Enooku et al., 2020). Several studies have shown that (fatty acid binding protein 5, FABP5) is highly expressed in HCC. It has been shown that FABP5 promotes angiogenesis and activates the IL6/STAT3/VEGFA pathway in HCCs (F. Liu et al., 2020). Overall survival time for HCC patients was negatively correlated with FABP5 levels in monocytes. The FABP5 protein promotes immune tolerance in patients with HCC by regulating monocytes and tumor-associated monocytes’ fatty acid oxidation process via suppressing the PPARα pathway (J. Liu et al., 2022). Our results indicated that FABP5 expression is significantly linked to the overall survival of HCC patients. FABP5 showed the most significant correlation with tumor-infiltrating immune subsets, such as B-cells, CD4+ T-cells, CD8+ T-cells, neutrophils, macrophages, and dendritic cells. The close association between certain genes, especially FATP5, and the presence of immune subsets that infiltrate tumor may indicate their importance in immune dysregulation in HCC. The SCD gene encodes an enzyme involved in the biosynthesis of fatty acids, primarily oleic acid. Cancer cells are resistant to chemotherapy-induced apoptosis partly because of the expression of SCD, which is mediated by phosphatidylinositol three kinase/c-Jun N-terminal kinases activation (Bansal et al., 2014). NAFLD fibrosis is known to be associated with an increase in CCL20, an essential inflammatory mediator (Chu et al., 2018). A poor prognosis is related to CCL20 expression in hepatocellular carcinomas after curative resection of cancer (X. Ding et al., 2012). The GPAT3 (AGPAT9) gene encodes a lysophosphatidic acid acyltransferase family member. The protein encoded by this gene catalyses the conversion of glycerol-3-phosphate to lysophosphatidic acid in triacylglycerol synthesis (J. Cao et al., 2006). Mice with severe congenital generalised lipodystrophies exhibit insulin resistance and hepatic steatosis when GPAT3 is deficient (Gao et al., 2020). It was found that knocking down GPAT3 effectively inhibited HCC cell growth, induced cell apoptosis, and blocked mTOR signalling in HCC cells. IL1RN encodes an antagonist protein (IL1RA) that binds to IL-1 as a natural antagonist. IL1RN is involved in developing NAFLD features (M. G. Wolfs et al., 2015). A serum level of L-1RA is associated with inflammation of the liver and higher levels of ALT regardless of obesity, alcohol consumption, or insulin resistance. There is potential for IL-1RA to be used as a non-invasive indicator of NASH inflammatory responses (Pihlajamäki et al., 2012). PLIN1, an adipocyte-specific protein encoded by this gene, coats lipid storage droplets to protect them until hormone-sensitive lipases can break them down. In adipocytes, PLIN1 is the major cAMP-dependent protein kinase substrate, and it may inhibit lipolysis when unphosphorylated (J. H. Sohn et al., 2018). NAFLD (non-alcoholic steatohepatitis, NASH) leads to an upregulation of PLIN1. However, it impairs glucose homeostasis and may be protective against lipotoxicity33 (Carr and Ahima, 2016). Our study indicated that PLIN1 mRNA expression is positively linked to overall survival. ## 5 Conclusion Generally, by utilising biological information research methods, we have identified six key genes for diagnosing NAFLD and HCC with MRFs. Moreover, five key genes were identified for the prognosis of HCC changes and the created gene marker composed of these genes was FABP5, CCL20, and GPAT3 may be the critical dangerous prognostic genes of HCC. PLIN1 and IL1RN are protective prognostic genes of HCC. Nevertheless, since our research is based on data analysis, further experiments would be required to confirm our findings. Nevertheless, we hope that our research findings will contribute to improving the diagnosis and treatment of liver disease associated with NAFLD and HCC progression. ## 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 XZ, FH, and WC conceived and designed the study. BN, QL, XQ, and YH acquired and analysed the data. FY performed data analysis and prepared the figures and tables. FY wrote the manuscript. QZ revised the manuscript. 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. Agosti P., Carlo S., Mazzocca A.. Emerging metabolic risk factors in hepatocellular carcinoma and their influence on the liver microenvironment. Biochim. Biophys. Acta. Mol. Basis Dis. (2018) 1864 607-617. DOI: 10.1016/j.bbadis.2017.11.026 2. Akahoshi K., Tanaka S., Mogushi K., Shimada S., Matsumura S., Akiyama Y.. Expression of connective tissue growth factor in the livers of non-viral hepatocellular carcinoma patients with metabolic risk factors. J. Gastroenterol. (2016) 51 910-922. DOI: 10.1007/s00535-015-1159-8 3. Arendt B. M., Comelli E. M., Ma D. W., Lou W., Teterina A., Kim T.. Altered hepatic gene expression in nonalcoholic fatty liver disease is associated with lower hepatic n-3 and n-6 polyunsaturated fatty acids. Hepatology (2015) 61 1565-1578. DOI: 10.1002/hep.27695 4. Awosika Joy, Sohal Davendra. A narrative review of systemic treatment options for hepatocellular carcinoma: State of the art review. J. Gastrointest. Oncol. (2022) 13 426-437. DOI: 10.21037/jgo-21-274 5. Bansal S., Berk M., Alkhouri N., DavidPartrick A., JohnFungAriel Feldstein J., Feldstein A.. Stearoyl-CoA desaturase plays an important role in proliferation and chemoresistance in human hepatocellular carcinoma. J. Surg. Res. (2014) 186 29-38. DOI: 10.1016/j.jss.2013.07.001 6. Bustoros M., Sklavenitis-Pistofidis R., Park J., Redd R., Zhitomirsky B., Dunford A. J.. Genomic profiling of smoldering multiple myeloma identifies patients at a high risk of disease progression. J. Clin. Oncol. (2020) 38 2380-2389. DOI: 10.1200/JCO.20.00437 7. Cao J., Li J. L., Li D., Tobin J. F., Gimeno R. E.. Molecular identification of microsomal acyl-CoA:glycerol-3-phosphate acyltransferase, a key enzyme in de novo triacylglycerol synthesis. Proc. Natl. Acad. Sci. U. S. A. (2006) 103 19695-19700. DOI: 10.1073/pnas.0609140103 8. Carr R M., Ahima R S.. Pathophysiology of lipid droplet proteins in liver diseases. Exp. Cell Res. (2016) 340 187-192. DOI: 10.1016/j.yexcr.2015.10.021 9. Chen C. L., Yang H. I., Yang W. S., Liu C. J., Chen P. J., You S. L.. Metabolic factors and risk of hepatocellular carcinoma by chronic Hepatitis B/C infection: A follow-up study in taiwan. Gastroenterology (2008) 135 111-121. DOI: 10.1053/j.gastro.2008.03.073 10. Chiyonobu N., Shimada S., Akiyama Y., Mogushi K., Itoh M., Akahoshi K.. "Fatty acid binding protein 4 (Fabp4) overexpression in intratumoral hepatic stellate cells within hepatocellular carcinoma with metabolic risk factors.". Am. J. Pathol. (2018) 188 1213-1224. DOI: 10.1016/j.ajpath.2018.01.012 11. Chu X., Jin Q., Chen H., Craig Wood G., Petrick A., Strodel W.. CCL20 is up-regulated in non-alcoholic fatty liver disease fibrosis and is produced by hepatic stellate cells in response to fatty acid loading. J. Transl. Med. (2018) 16 108. DOI: 10.1186/s12967-018-1490-y 12. Ding X., Wang K., Wang H., Zhang G., Liu Y., Yang Q.. High expression of CCL20 is associated with poor prognosis in patients with hepatocellular carcinoma after curative resection. J. Gastrointest. Surg. (2012) 16 828-836. DOI: 10.1007/s11605-011-1775-4 13. Enooku K., Tsutsumi T., Kondo M., Fujiwara N., Sasako T., Shibahara J.. Hepatic FATP5 expression is associated with histological progression and loss of hepatic fat in NAFLD patients. J. Gastroenterol. (2020) 55 227-243. DOI: 10.1007/s00535-019-01633-2 14. Ericksen R. E., Lim S. L., McDonnell E., Shuen W. H., Vadiveloo M., White P. J.. Loss of BCAA catabolism during carcinogenesis enhances mTORC1 activity and promotes tumor development and progression. Cell Metab. (2019) 29 1151-1165. DOI: 10.1016/j.cmet.2018.12.020 15. Frades I., Andreasson E., Mato J. M., Alexandersson E., Matthiesen R., Martínez-Chantar M. L.. Integrative Genomic Signatures of Hepatocellular Carcinoma Derived from Nonalcoholic Fatty Liver Disease. PLoS ONE (2015) 10 e0124544. PMID: 25993042 16. Gao M, Liu Lin, Wang X, Yin Mak H, Liu G, Yang H,. GPAT3 deficiency alleviates insulin resistance and hepatic steatosis in a mouse model of severe congenital generalized lipodystrophy. Hum. Mol. Genet. (2020) 29 432-443. DOI: 10.1093/hmg/ddz300 17. The gene Ontology (go) project in 2006. Nucleic Acids Res (2006) 34 D322-D326. DOI: 10.1093/nar/gkj021 18. George B., Seals S., Aban I.. Survival analysis and regression models. J. Nucl. Cardiol. (2014) 21 686-694. DOI: 10.1007/s12350-014-9908-2 19. Heagerty P. J., Lumley T., Pepe M. S.. Time-dependent ROC curves for censored survival data and a diagnostic marker. Biometrics (2000) 56 337-344. DOI: 10.1111/j.0006-341x.2000.00337.x 20. Huang S. F., Chang I. C., Hong C. C., Yen T. C., Chen C. L., Wu C. C.. Metabolic risk factors are associated with non-Hepatitis B non-hepatitis C hepatocellular carcinoma in Taiwan, an endemic area of chronic Hepatitis B. Hepatol. Commun. (2018) 2 747-759. DOI: 10.1002/hep4.1182 21. Kanehisa M., Sato Y., Kawashima M., Furumichi M., Tanabe M.. KEGG as a reference resource for gene and protein annotation. Nucleic Acids Res. (2016) 44 D457-D462. DOI: 10.1093/nar/gkv1070 22. Kurebayashi Y., Ojima H., Tsujikawa H., Kubota N., Maehara J., Abe Y.. Landscape of immune microenvironment in hepatocellular carcinoma and its additional impact on histological and molecular classification. Hepatology (2018) 68 1025-1041. DOI: 10.1002/hep.29904 23. Li T., Fan J., Wang B., Traugh N., Chen Q., Liu J. S.. Timer: A web server for comprehensive analysis of tumor-infiltrating immune cells. Cancer Res. (2017) 77 e108-e110. DOI: 10.1158/0008-5472.CAN-17-0307 24. Lin Y. P., Wang P. M., Chuang C. H., Yong C. C., Liu Y. W., Huang P. Y.. Metabolic risks are increasing in non-B non-C early-stage hepatocellular carcinoma: A 10-year follow-up study. Front. Oncol. (2022) 12 816472. DOI: 10.3389/fonc.2022.816472 25. Liu F., Liu W., Zhou S., Yang C., Tian M., Jia G.. Identification of Fabp5 as an immunometabolic marker in human hepatocellular carcinoma. J immunother cancer (2020) 8 26. Liu J., Sun B., Guo K., Yang Z., Zhao Y., Gao M.. "Lipid-Related Fabp5 activation of tumor-associated monocytes fosters immune privilege via Pd-L1 expression on treg cells in hepatocellular carcinoma.". cancer gene ther (2022) 27. Llovet J. M., Kate Kelley R., Villanueva A., AmitSingal G., Pikarsky E., Roayaie S.. Hepatocellular carcinoma. Nat. Rev. Dis. Prim. (2021) 7 6. DOI: 10.1038/s41572-020-00240-3 28. Lonardo A., Byrne C. D., Caldwell S. H., Cortez-Pinto H., Targher G.. Global epidemiology of nonalcoholic fatty liver disease: Meta-analytic assessment of prevalence, incidence, and outcomes. Hepatology (2016) 64 1388-1389. DOI: 10.1002/hep.28584 29. Mancina R., Margherita P., Dongiovanni S., Petta P., Pingitore M.. The MBOAT7-TMC4 variant rs641738 increases risk of nonalcoholic fatty liver disease in individuals of European descent. Gastroenterology (2016) 150 1219. DOI: 10.1053/j.gastro.2016.01.032 30. Morse M. A., Sun W., Kim R., He A. R., Abada P. B., Mynderse M.. The role of angiogenesis in hepatocellular carcinoma. Clin. Cancer Res. (2019) 25 912-920. DOI: 10.1158/1078-0432.CCR-18-1254 31. Pihlajamäki J., Kuulasmaa T., Kaminska D., Simonen M., Kärjä V., Grönlund S.. Serum interleukin 1 receptor antagonist as an independent marker of non-alcoholic steatohepatitis in humans. J. Hepatol. (2012) 56 663-670. DOI: 10.1016/j.jhep.2011.10.005 32. Pouwels S., Sakran N., Graham Y., Leal A., Pintar T., Yang W.. Non-alcoholic fatty liver disease (NAFLD): A review of pathophysiology, clinical management and effects of weight loss. BMC Endocr. Disord. (2022) 22 63. DOI: 10.1186/s12902-022-00980-1 33. 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 34. Robin X., Turck N., Hainard A., Tiberti N., Lisacek F., Sanchez J. C.. pROC: an open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinforma. (2011) 12 77. DOI: 10.1186/1471-2105-12-77 35. Sohn J. H., Lee Y. K., Han J. S., Jeon Y. G., Kim J. I., Choe S. S.. "Perilipin 1 (Plin1) deficiency promotes inflammatory responses in lean adipose tissue through lipid dysregulation.". J. Biol. Chem. (2018) 293 13974-13988. DOI: 10.1074/jbc.RA118.003541 36. 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 37. Tan Y., Zhang X., Zhang W., Tang L., Yang H., Yan K.. The influence of metabolic syndrome on the risk of hepatocellular carcinoma in patients with chronic hepatitis B infection in mainland China. Cancer Epidemiol. Biomarkers Prev. (2019) 28 2038-2046. DOI: 10.1158/1055-9965.EPI-19-0303 38. Tang Z., Li C., Kang B., Gao G., Li C., Zhang Z.. Gepia: A web server for cancer and normal gene expression profiling and interactive analyses. Nucleic Acids Res. (2017) 45 W98-w102. DOI: 10.1093/nar/gkx247 39. Trierweiler C., Hockenjos B., Zatloukal K., Thimme R., Blum H. E., Wagner E. F.. The transcription factor c-JUN/AP-1 promotes HBV-related liver tumorigenesis in mice. Cell Death Differ. (2016) 23 576-582. DOI: 10.1038/cdd.2015.121 40. Wattacheril J.. Extrahepatic manifestations of nonalcoholic fatty liver disease. Gastroenterol. Clin. North Am. (2020) 49 141-149. DOI: 10.1016/j.gtc.2019.10.002 41. Wolfs M. G., Gruben N., Rensen S. S., Verdam F. J., Greve J. W., Driessen A.. Determining the association between adipokine expression in multiple tissues and phenotypic features of non-alcoholic fatty liver disease in obesity. Nutr. Diabetes (2015) 5 e146. DOI: 10.1038/nutd.2014.43 42. Wu L, Li J, Feng J, Ji J, Yu Q, Li Y. Crosstalk between PPARs and gut microbiota in NAFLD. (2021) 136 111255. DOI: 10.1016/j.biopha.2021.111255 43. Xu W, Chen Z, Liu G, Dai Y, Xu X, Ma D. Identification of a potential PPAR-related multigene signature predicting prognosis of patients with hepatocellular carcinoma. PPAR Res. (2021) 2021 6642939. DOI: 10.1155/2021/6642939 44. Zhang J., Liu X., Zhou W., Lu S., Wu C., Wu Z.. Identification of key genes associated with the process of hepatitis B inflammation and cancer transformation by integrated bioinformatics analysis. Front. Genet. (2021a) 12 654517. DOI: 10.3389/fgene.2021.654517 45. Zhang Y., Ge X., Li Y., Zhang B., Wang P., Hao M.. TWIST2 and the PPAR signaling pathway are important in the progression of nonalcoholic steatohepatitis. Lipids Health Dis. (2021b) 20 39. DOI: 10.1186/s12944-021-01458-0
--- title: 'The association between sleep duration, quality, and nonalcoholic fatty liver disease: A cross-sectional study' authors: - Huiwei Liu - Shiliang Huang - Mengdan Xu - Dan Zhao - Xinxue Wang - Liangshun Zhang - Dahua Chen - Jinman Du - Rongbin Yu - Hong Li - Hua Ye journal: Open Medicine year: 2023 pmcid: PMC10025511 doi: 10.1515/med-2023-0670 license: CC BY 4.0 --- # The association between sleep duration, quality, and nonalcoholic fatty liver disease: A cross-sectional study ## Abstract Sleep can affect nonalcoholic fatty liver disease (NAFLD). We investigated the association between sleep duration, sleep quality, and NAFLD. From January to December 2018, 1,073 patients (age: 37.94 ± 10.88, Body Mass Index (BMI): 22.85 ± 3.27) were enrolled. Pittsburgh Sleep Quality Index Questionnaire and Munich Chronotype Questionnaire were used to assess sleep duration, quality, and habits. Ultrasonography was used to diagnose NAFLD. Multivariate logistic regression models were used to calculate the odds ratio (OR) and $95\%$ confidence interval (CI) of the risk of NAFLD by different types of sleep duration and sleep quality. No significant differences in sleep time, sleep quality, and sleep habits between the NAFLD and the non-NAFLD groups were observed ($P \leq 0.05$). There was no correlation between sleep duration and NAFLD in the whole cohort. After adjusting for age, exercise, fasting plasma glucose, and BMI, the group with long sleep duration showed a decreased risk of NAFLD in men (OR = 0.01, $95\%$ CI: 0.001–0.27, $$P \leq 0.032$$). However, in all four adjusted models, no correlation between sleep duration, quality, and NAFLD was found in women. In conclusion, sleep duration was significantly and negatively associated with NAFLD in men but not women. Prospective studies are required to confirm this association. ## Introduction Nonalcoholic fatty liver disease (NAFLD) is defined as the accumulation of triglycerides in the liver of individuals without significant alcohol consumption or viral hepatitis infection [1]. NAFLD has become one of the most common liver diseases worldwide, with a global prevalence of approximately $25\%$ [2]. Unhealthy lifestyle habits, such as excess nutritional intake, less physical activity, or insufficient sleep are thought to be associated with obesity and diabetes. NAFLD is strongly associated with metabolic diseases such as obesity, diabetes, or dyslipidemia which are risk factors for nonalcoholic steatohepatitis, the severe form of NAFLD [3]. There are currently no approved medical therapies for NAFLD [4]. The first-line treatment is lifestyle interventions, such as diet modification and exercise [5,6]. However, compared with nutrition or exercise, sleep improvement has drawn less attention. Sleep disturbance and deprivation are common medical complaints in modern society [7]. Insufficient sleep may cause metabolic disorders such as insulin sensitivity, obesity, and type 2 diabetes mellitus, and may therefore contribute to the development of NAFLD [8,9]. The associations between sleep duration or quality and the prevalence of NAFLD have been reported in several studies [8–11]. However, the relationships between sleep duration, quality, and NAFLD remain controversial [7–12]. In a community-based cohort study, long sleep duration was associated with the elevation of NAFLD scores in Korean middle-aged adults [12]. In another larger middle-aged Korean population study, short sleep duration and poor sleep quality were significantly associated with an increased risk of NAFLD [10]. A meta-analysis pooled data from six studies and found a small but significantly increased risk of NAFLD among short sleep duration subjects [13]. In the study, we investigated the associations between sleep duration, quality, and NAFLD, as determined by ultrasonography which is a common clinical diagnostic tool for detecting fatty liver [10]. Besides, we examined the gender differences in sleep duration, quality, and their effects on the risk of NAFLD. ## Study population The subjects in this study were from Lihuili Hospital of Ningbo Medical Center and the Physical Examination Center of Lihuili East Hospital of Ningbo. The study population was restricted to individuals who underwent a health screening examination with information on sleep duration and sleep quality from January 2018 to December 2018 ($$n = 1$$,714). All subjects received a questionnaire survey and physical examination. The exclusion criteria are incomplete medical examination data, uncompleted sleep questionnaire, lack of hepatic ultrasonography, lack of BMI information, lack of triglycerides data, lack of fasting blood glucose data, history of liver disease, and pregnancy and breastfeeding. A total of 641 subjects met one or more of the exclusion criteria at baseline (Figure 1). The total number of eligible subjects for the study was 1,073. The protocol conformed to the ethical guidelines of the 1975 Declaration of Helsinki and was approved by the ethics committee of the Lihuili Hospital of Ningbo (ethics review number: 2018036). All the subjects knew the purpose of the questionnaire survey and signed the informed consent before entering the study. Figure 1: Flowchart of the included subjects. ## Data collection Sleep duration and quality were assessed using the validated Pittsburgh Sleep Quality Index (PSQI), a self-administered questionnaire. PSQI, comprised of 19 items, generates seven component scores that reflect subjective sleep quality, sleep latency, sleep duration, habitual sleep efficiency, sleep disturbances, use of sleeping medication, and daytime function [14,15]. Poor sleep quality is defined as having a sum of scores greater than 5 for these seven components. Sleep duration was estimated using component 3, which asked for the number of hours of actual night-time sleep during the past month. Sleep duration means the sum of all sleep time in a day. According to the 2015 National Sleep Foundation recommendations for sleep disorders, we classified sleep duration by age group [16]. During health examinations, information such as medical history, medication use, health-related behavior, physical measurements, and serum biochemical measurements is collected. There were questions regarding the amount and frequency of alcohol consumed weekly and daily. We also collected information on smoking history (duration and daily consumption of cigarettes) using questionnaires. Sleep habits were assessed using the Munich Chronotype Questionnaire (MCTQ) [17]. The assessment of daily life includes the distribution and assessment of daily behaviors, as well as a reminder of working days and rest days [18]. We define 0 as a very early bedtime, 1 as an appropriate early bedtime, 2 as an earlier bedtime, 3 as a normal bedtime, 4 as a slightly late bedtime, 5 as an appropriate late bedtime, and 6 as an extremely late bedtime. Any type of exercise which lasts longer than 20 min, every time more than 3 times a week, is classified as physical exercise. Depending on the eating habits in the past month, three types of eating habits have been identified: meat-based, vegetable-based, and meat and vegetable equivalent. For Body Mass Index (BMI), we defined BMI < 18.5 kg/m2 as lean, 18.5 kg/m2 ≤ BMI < 24 kg/m2 as normal, 24 kg/m2 ≤ BMI < 28 kg/m2 as overweight; 28 kg/m2 ≤ BMI as obesity. According to the classification by the American Diabetes Association, fasting plasma glucose (FPG) levels were divided into three grades: normal (FPG < 5.6 mmol/L), impaired (5.6 mmol/L ≤ FPG < 7.0 mmol/L), and diabetes mellitus (FPG ≥ 7.0 mmol/L) [19]. To make the questionnaire reliable, the survey was designed according to the basic situation of the Chinese population preliminary design and was discussed, evaluated, and revised by epidemiologists and clinical experts before large-scale application. Before the beginning of the survey, all personnel participating in the survey were given unified training, the questions in the questionnaire were explained one by one, and the relevant controversial issues were standardized. The questionnaire was completed after face-to-face, one-to-one guidance between the investigators and the respondents. ## Statistical analysis The subject characteristics were expressed as mean ± standard deviation for continuous variables or as percentages for categorical variables. The significance of differences in gender, age, smoking, and drinking between NAFLD and non-NAFLD groups was determined by the chi-square test. Continuous variables such as age and BMI were analyzed by a two-sided t-test or by analysis of variance. A univariate and multivariate logistic regression model was used to analyze the contributions of different variables. The odds ratio (OR) and $95\%$ confidence interval ($95\%$ CI) were used to evaluate the effect of sleep-related variables on NAFLD. The multivariate model was controlled for potential covariates such as age, smoking, drinking, physical exercise, eating habits, and BMI. All statistical analysis was performed by SPSS 18.0 software. All reported P values are two-tailed, and the statistically significant threshold was set at 0.05. ## General characteristics of subjects We enrolled 1,073 valid participants comprising 282 NAFLD and 791 non-NAFLD controls. The demographic and clinical characteristics of the study population are shown in Table 1. The average age of the NAFLD group (41.30 ± 10.04 years) and the non-NAFLD group (36.74 ± 10.92 years) was significantly different ($t = 6.153$, $P \leq 0.001$). Participants were further divided into six age groups and the differences in age distribution were also significant between NAFLD and non-NAFLD groups ($P \leq 0.001$). The average BMI of the NAFLD group (25.86 ± 2.76 kg/m2) and non-NAFLD group (21.7 ± 2.72 kg/m2) was significantly different (t = −2.102, $$P \leq 0.036$$). The BMI segment distribution was also significantly different (χ 2 = 319.772, $P \leq 0.001$). The composition of smokers between NAFLD and non-NAFLD groups was different (χ 2 = 4.906, $$P \leq 0.027$$). The eating habits between the NAFLD group and the non-NAFLD group were also different (χ 2 = 9.444, $$P \leq 0.009$$). The two groups also have significant differences in marital status (χ 2 = 4.544, $$P \leq 0.033$$). However, there was no significant difference in the composition of drinkers (χ 2 = 0.064, $$P \leq 0.800$$) or exercisers (χ 2 = 2.082, $$P \leq 0.149$$) or occupations (χ 2 = 0.284, $$P \leq 0.594$$) between the two groups. Table 1 \| Variables \| NAFLD (n = 282) \| NAFLD (n = 282).1 \| Non-NAFLD (n = 791) \| Non-NAFLD (n = 791).1 \| P \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| Variables \| N \| Percentage \| N \| Percentage \| P \| \| Age \| 41.30 ± 10.04 \| 41.30 ± 10.04 \| 36.74 ± 10.92 \| 36.74 ± 10.92 \| <0.001a \| \| Age \| 41.30 ± 10.04 \| 41.30 ± 10.04 \| 36.74 ± 10.92 \| 36.74 ± 10.92 \| <0.001b \| \| <20 \| 0 \| 0 \| 4 \| 0.51 \| \| \| 20–30 \| 34 \| 12.06 \| 241 \| 30.47 \| \| \| 30–40 \| 102 \| 36.17 \| 268 \| 33.88 \| \| \| 40–50 \| 75 \| 26.60 \| 152 \| 19.22 \| \| \| 50–60 \| 64 \| 22.70 \| 108 \| 13.65 \| \| \| ≥60 \| 7 \| 2.47 \| 18 \| 2.27 \| \| \| Gender \| \| \| \| \| 0.001b \| \| Men \| 253 \| 89.72 \| 459 \| 58.03 \| \| \| Women \| 29 \| 10.28 \| 332 \| 41.97 \| \| \| BMI (kg/m2) \| 25.86 ± 2.76 \| 25.86 ± 2.76 \| 21.7 ± 2.72 \| 21.7 ± 2.72 \| 0.036a \| \| BMI (kg/m2) \| 25.86 ± 2.76 \| 25.86 ± 2.76 \| 21.7 ± 2.72 \| 21.7 ± 2.72 \| <0.001b \| \| <18.5 \| 0 \| 0 \| 84 \| 10.62 \| \| \| 18.5–24 \| 63 \| 22.34 \| 545 \| 68.90 \| \| \| 24–28 \| 163 \| 57.80 \| 146 \| 18.46 \| \| \| ≥28 \| 56 \| 19.85 \| 16 \| 2.03 \| \| \| Smoking history \| Smoking history \| Smoking history \| \| \| 0.027b \| \| Smoking now \| 65 \| 23.05 \| 135 \| 17.07 \| \| \| No smoking \| 217 \| 76.95 \| 656 \| 82.93 \| \| \| Drinking history \| Drinking history \| Drinking history \| \| \| 0.800b \| \| Drinking now \| 50 \| 17.73 \| 135 \| 17.07 \| \| \| No drinking \| 232 \| 82.27 \| 656 \| 82.93 \| \| \| Eating habits \| \| \| \| \| 0.009b \| \| Meat \| 73 \| 25.89 \| 148 \| 18.71 \| \| \| Vegetable \| 35 \| 12.41 \| 144 \| 18.20 \| \| \| Equal \| 174 \| 61.70 \| 499 \| 63.08 \| \| \| Physical exercise \| Physical exercise \| Physical exercise \| \| \| 0.149b \| \| Yes \| 108 \| 38.30 \| 342 \| 43.24 \| \| \| No \| 174 \| 61.70 \| 449 \| 56.76 \| \| \| Marriage \| \| \| \| \| 0.033b \| \| Married \| 231 \| 81.91 \| 599 \| 75.73 \| \| \| Singlec \| 51 \| 18.09 \| 192 \| 24.27 \| \| \| Occupation \| \| \| \| \| 0.594b \| \| Yes \| 265 \| 93.97 \| 736 \| 93.05 \| \| \| No \| 17 \| 6.03 \| 55 \| 6.95 \| \| ## Association between sleep duration, quality, habits, and NAFLD We further examined the associations between sleep duration, sleep quality, sleep habits, and NAFLD in the whole cohort. No significant associations were observed between sleep and NAFLD (Table 2). Multivariate logistic regression analysis for independent variables in NAFLD identified physical exercise as a protective factor (OR: 1.59, $95\%$ CI: 1.11–2.28), while age (OR: 0.97, $95\%$ CI: 0.96–0.99), gender (OR: 0.36, $95\%$ CI: 0.22–0.60), BMI (OR: 0.61, $95\%$ CI: 0.57–0.66), and FPG (OR: 0.63, $95\%$ CI: 0.48–0.81) as risk factors for NAFLD with $P \leq 0.05.$ However, sleep duration, sleep quality, and sleep habits showed no association with NAFLD in the whole cohort (Table 3). To test if the associations exist in different genders, we performed logistic regression analysis for NAFLD and sleep in men and women separately. We found no correlation between sleep habits and NAFLD in both men and women subjects, whether the covariate was adjusted or not. However, we found that the risk of NAFLD was significantly lower in men with recommended or longer sleep duration than in those with too short sleep duration after adjustment for age, exercise, FPG, and BMI (OR = 0.01, $95\%$ CI: 0.001–0.27, $$P \leq 0.032$$). No associations between sleep duration, sleep quality, and NAFLD was observed in women (Table 4). ## Discussion There are few studies about associations of sleep and NAFLD in Chinese population. In the study, the associations of NAFLD with sleep duration, sleep quality, and sleep habits were evaluated. Sleep duration, sleep quality, and sleep habits showed no statistically significant association with NAFLD in the whole cohort when the model was adjusted for other parameters. It was found that the risk of NAFLD was significantly lower in men with long sleep duration than in those with short sleep duration. However, sleep quality and sleep habits in both genders were not associated with NAFLD. Previous studies have suggested that sleep quality was associated with NAFLD, and there were sex differences [6]. However, results showed no associations of NAFLD with the global PSQI score, subjective sleep quality score, sleep duration score, and sleep disturbance score [6]. A recent study also showed that a mean sleep time of 7 h or more had a significant negative relationship with NAFLD [20]. Short sleep duration was associated with an increased risk of prevalent NAFLD in Chinese, two South Korean, and American populations [8–11]. Different results were also reported in some South Korean populations. Poor sleep quality but not sleep duration was associated with a lower risk of NAFLD in men. In women, the association of sleep quality and duration with the risk of NAFLD was insignificant [7]. A relationship between long sleep duration and the elevation of NAFLD scores was found after adjusting for several confounding factors in Korean middle-aged adults [12]. Thus, these findings are controversial, although a recent study pooled six datasets and meta-analysis demonstrated a small but significantly increased risk of NAFLD among participants who had short sleep duration [13]. Many reasons may account for the above phenomenon. NAFLD is the most common chronic liver disease worldwide. The disease is a heterogeneous group of liver diseases characterized by the accumulation of fat in the liver [21]. The pathogenesis of NAFLD is not yet fully understood. Many factors may contribute to NAFLD, including diet, medications, genetic predisposition, and gut microbiota [22]. Although this study had a large enough sample size to conclude, it had some limitations. First, the PSQI assessment is a self-reported questionnaire and is subjective. Thus, the associations between sleep quality and NAFLD need to be confirmed by objective methods, such as polysomnography. Second, the study was cross-sectional. The collected data reflect the recent sleep characteristics as well as NAFLD status. Therefore, longitudinal evaluation is essential to reveal associations between sleep quality, duration, and NAFLD in the future. Third, the influences of nutritional factors, genetic background, and gut microbiota are hard to control in the study [22]. Hypercaloric nutrition, including the effects of saturated fat and fructose, as well as adipose tissue dysfunction and intestinal dysbiosis, may contribute to the incidence of NAFLD [23]. In the future study, these factors can be recorded and incorporated into the disease model. Finally, more large-scale cohorts are needed to draw an explicit conclusion. The multi-center study can be carried out to increase statistical power for associations between sleep and NAFLD. Our result suggests that men with short sleep duration should be cautious about their sleep habits as they have a higher probability to have NAFLD. It has been reported that individuals with NAFLD were more likely to be men and had a higher prevalence of sleep disorders [24]. Thus, the differences in gender should be considered when designing large-scale epidemiologic studies. The molecular mechanisms of chronic sleep deprivation have been extensively studied. Transcriptome analysis identified genes affected by insufficient sleep were associated with circadian rhythms, sleep homeostasis, oxidative stress, and metabolism [25]. Short sleep may increase diabetes risk through three pathways that are alterations in glucose metabolism, upregulation of appetite, and decreased energy expenditure. Type 2 diabetes mellitus is a risk factor often linked with NAFLD [26]. ## Conclusions According to the study, we found that sleep duration is an independent influencing factor of NAFLD in men, and the risk of NAFLD decreases with the increase in sleep duration, but there are no significant associations in women. Thus, men with short sleep duration should be cautious about their sleep habits as they have a higher probability to have NAFLD. ## References 1. Jullian-Desayes I, Trzepizur W, Boursier J, Joyeux-Faure M, Bailly S, Benmerad M. Obstructive sleep apnea, chronic obstructive pulmonary disease and NAFLD: an individual participant data meta-analysis. Sleep Med (2021) 77 357-64. DOI: 10.1016/j.sleep.2020.04.004 2. Loomba R, Sanyal AJ. The global NAFLD epidemic. Nat Rev Gastroenterol Hepat (2013) 10 686-90. DOI: 10.1038/nrgastro.2013.171 3. Kasper P, Martin A, Lang S, Kütting F, Goeser T, Demir M. NAFLD and cardiovascular diseases: a clinical review. Clin Res Cardiol (2021) 110 921-37. DOI: 10.1007/s00392-020-01709-7 4. Chalasani N, Younossi Z, Lavine JE, Charlton M, Cusi K, Rinella M. 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. DOI: 10.1002/hep.29367 5. Caylak E. The genetics of sleep disorders in humans: narcolepsy, restless legs syndrome, and obstructive sleep apnea syndrome. Am J Med Genet A (2009) 149A 2612-26. DOI: 10.1002/ajmg.a.33087 6. Takahashi A, Anzai Y, Kuroda M, Kokubun M, Kondo Y, Ogata T. Effects of sleep quality on non-alcoholic fatty liver disease: a cross-sectional survey. BMJ Open (2020) 10 e039947. DOI: 10.1136/bmjopen-2020-039947 7. Chou YT, Cheng HJ, Wu JS, Yang YC, Chou CY, Chang CJ. The association of sleep duration and sleep quality with non-alcoholic fatty liver disease in a Taiwanese population. Obes Res Clin Pract (2018) 12 500-5. DOI: 10.1016/j.orcp.2018.05.002 8. Peng K, Lin L, Wang Z, Ding L, Huang Y, Wang P. Short sleep duration and longer daytime napping are associated with non-alcoholic fatty liver disease in Chinese adults. J Diabetes (2017) 9 827-36. DOI: 10.1111/1753-0407.12489 9. Um YJ, Chang Y, Jung HS, Cho IY, Shin JH, Shin H. Sleep duration, sleep quality, and the development of nonalcoholic fatty liver disease: a cohort study. Clin Transl Gastroenterol (2021) 12 e00417. DOI: 10.14309/ctg.0000000000000417 10. Kim CW, Yun KE, Jung HS, Chang Y, Choi ES, Kwon MJ. Sleep duration and quality in relation to non-alcoholic fatty liver disease in middle-aged workers and their spouses. J Hepatol (2013) 59 351-7. DOI: 10.1016/j.jhep.2013.03.035 11. Kim D, Kim HJ, Kushida CA, Heo NY, Ahmed A, Kim WR. Short sleep duration is associated with abnormal serum aminotransferase activities and nonalcoholic fatty liver disease. Clin Gastroenterol Hepatol (2018) 16 588-90. DOI: 10.1016/j.cgh.2017.08.049 12. Kim JH, Jung DH, Kwon YJ, Lee JI, Shim JY. The impact of the sleep duration on NAFLD score in Korean middle-aged adults: a community-based cohort study. Sleep Med (2019) 57 144-50. DOI: 10.1016/j.sleep.2019.02.012 13. Wijarnpreecha K, Thongprayoon C, Panjawatanan P, Ungprasert P. Short sleep duration and risk of nonalcoholic fatty liver disease: a systematic review and meta-analysis. J Gastroenterol Hepatol (2016) 31 1802-7. DOI: 10.1111/jgh.13391 14. Buysse DJ, III Reynolds CF, Monk TH, Hoch CC, Yeager AL, Kupfer DJ. Quantification of subjective sleep quality in healthy elderly men and women using the Pittsburgh Sleep Quality Index (PSQI). Sleep (1991) 14 331-8. DOI: 10.1093/sleep/14.4.331 15. Buysse DJ, 3rd Reynolds CF, Monk TH, Berman SR, Kupfer DJ. The Pittsburgh Sleep Quality Index: a new instrument for psychiatric practice and research. Psychiatr Res (1989) 28 193-213. DOI: 10.1016/0165-1781(89)90047-4 16. Hirshkowitz M, Whiton K, Albert SM, Alessi C, Bruni O, DonCarlos L. National sleep foundation’s sleep time duration recommendations: methodology and results summary. Sleep Health (2015) 1 40-3. DOI: 10.1016/j.sleh.2014.12.010 17. Roenneberg T, Kuehnle T, Juda M, Kantermann T, Allebrandt K, Gordijn M. Epidemiology of the human circadian clock. Sleep Med Rev (2007) 11 429-38. DOI: 10.1016/j.smrv.2007.07.005 18. Roenneberg T, Wirz-Justice A, Merrow M. Life between clocks: daily temporal patterns of human chronotypes. J Biol Rhythm (2003) 18 80-90. DOI: 10.1177/0748730402239679 19. American Diabetes Association. Standards of medical care in diabetes – 2009. Diabetes Care (2009) 32 S13-61. DOI: 10.2337/dc09-S013 20. Seo Y, Sung GH, Lee S, Han KJ. Weekend catch-up sleep is associated with the alleviation of non-alcoholic fatty liver disease. Ann Hepatol (2022) 27 100690. DOI: 10.1016/j.aohep.2022.100690 21. Parthasarathy G, Revelo X, Malhi H. Pathogenesis of nonalcoholic steatohepatitis: an overview. Hepatol Commun (2020) 4 478-92. DOI: 10.1002/hep4.1479 22. Yu J, Marsh S, Hu J, Feng W, Wu C. The pathogenesis of nonalcoholic fatty liver disease: interplay between diet, gut microbiota, and genetic background. Gastroenterol Res Pract (2016) 2016 2862173. DOI: 10.1155/2016/2862173 23. Pafili K, Roden M. Nonalcoholic fatty liver disease (NAFLD) from pathogenesis to treatment concepts in humans. Mol Metab (2021) 50 101122. DOI: 10.1016/j.molmet.2020.101122 24. Mir HM, Stepanova M, Afendy H, Cable R, Younossi ZM. Association of sleep disorders with nonalcoholic fatty liver disease (NAFLD): a population-based study. J Clin Exp Hepatol (2013) 3 181-5. DOI: 10.1016/j.jceh.2013.06.004 25. Ye H, Huang S, Song Y, Liu H, Zhao X, Zhao D. Gene co-expression analysis identifies modules related to insufficient sleep in humans. Sleep Med (2021) 86 68-74. DOI: 10.1016/j.sleep.2021.08.010 26. Knutson KL, Spiegel K, Penev P, Van Cauter E. The metabolic consequences of sleep deprivation. Sleep Med Rev (2007) 11 163-78. DOI: 10.1016/j.smrv.2007.01.002
--- title: Changes in mortality of Polish residents in the early and late old age due to main causes of death from 2000 to 2019 authors: - Monika Burzyńska - Małgorzata Pikala journal: Frontiers in Public Health year: 2023 pmcid: PMC10025537 doi: 10.3389/fpubh.2023.1060028 license: CC BY 4.0 --- # Changes in mortality of Polish residents in the early and late old age due to main causes of death from 2000 to 2019 ## Abstract ### Purpose The aim of the study was to assess mortality trends in Poland between 2000 and 2019 in the early and late old age population (65–74 years and over 75 years). ### Methods The work used data on all deaths of Polish residents aged over 65 years ($$n = 5$$,496,970). The analysis included the five most common major groups of causes of death: diseases of the circulatory system, malignant neoplasms, diseases of the respiratory system, diseases of the digestive system and external causes of mortality. The analysis of time trends has been carried out with the use of joinpoint models. The Annual Percentage Change (APC) for each segments of broken lines, the Average Annual Percentage Change (AAPC) for the whole study period ($95\%$ CI), and standardized death rates (SDRs) were calculated. ### Results The percentage of deaths due to diseases of the circulatory system decreased in all the studied subgroups. Among malignant neoplasms, lung and bronchus cancers accounted for the largest percentage of deaths, for which the SDRs among men decreased, while those among women increased. In the early old age, the SDR value increased from 67.8 to 76.3 (AAPC = $0.6\%$, $p \leq 0.05$), while in the late old age group it increased from 112.1 to 155.2 (AAPC = $1.8\%$, $p \leq 0.05$). Among men, there was an upward trend for prostate cancer (AAPC = $0.4\%$ in the early old age group and AAPC = $0.6\%$ in the late old age group, $p \leq 0.05$) and a downward trend for stomach cancer (AAPC −3.2 and −$2.7\%$, respectively, $p \leq 0.05$). Stomach cancer also showed a decreasing trend among women (AAPC −3.2 and −$3.6\%$, $p \leq 0.05$). SDRs due to influenza and pneumonia were increasing. Increasing trends in mortality due to diseases of the digestive system in women and men in the early old age group have been observed in recent years, due to alcoholic liver disease. Among the external causes of mortality in the late old age group, the most common ones were falls. ### Conclusions It is necessary to conduct further research that will allow to diagnose risk and health problems of the elderly subpopulation in order to meet the health burden of the aging society. ## Introduction The process of population aging has demographic, economic, social and health dimensions. This is because the phenomenon is indirectly influenced by a number of factors, such as the level of affluence of the population, changes in the family model, professional activity of women, the quality of social and health care, education and government policies in the field of public health [1]. Demographic forecasts predict that in 2050 the percentage of elderly people in the world will reach $16\%$. In EU countries, there will be only two people of working age for every person aged 65 years or over, while in Poland the share of people aged 65 years and over will be nearly $40\%$. The oldest age group, i.e., individuals over 85 years will constitute the largest group of people. The size of this group is expected to increase by more than 2.5 times as compared to 2020 [2]. In Poland, at the end of 2020, the percentage of people aged 65 years and over was $23.8\%$, while the old-age dependency ratio, defined as the number of people aged 65 years and over per 100 people aged from 15 to 64 years, was 28.2. In view of the advancement of the population aging process, it is very important to analyse the health status of the elderly subpopulation [3, 4]. Data on deaths are one of the most important sources of information on a population's health status in all age groups. Due to the fact that deaths have to be registered, they provide a database of complete information on the causes of mortality in societies around the world. Based on the data from the Global Burden of Disease Study, between 1990 and 2017 as many as 12 million additional deaths worldwide were related to population aging. This accounted for $27.9\%$ of all deaths, with the largest share attributed to ischemic heart disease [5]. The mortality structure and trends among the elderly reflect the mortality of the general population. In Poland, the predominant cause of death is cardiovascular disease, accounting for $42.6\%$ of all deaths, with a ten-percentage-point decline since [6]. The percentage of deaths from this cause in people aged over 65 years has also declined and is $41.1\%$. What is characteristic for the elderly is that the rate of deaths from cardiovascular disease among men only slightly exceeds that among women, while in younger age groups, mortality among men significantly exceeds that of women [7]. The second most common cause of death in the general Polish population is cancer. The Health at a Glance 2021 report shows that the incidence of cancer in Poland has increased, which may indicate an improvement in early cancer diagnosis. However, the rate in Poland is still relatively low, reaching 267 per 100,000 population, with the average for OECD countries at 294, respectively. In contrast, the mortality rate from malignancies in Poland is one of the highest in OECD countries, at 228 deaths per 100,000 population, with an average of 191 per 100,000 [8]. Cancer is also the second cause of death in people aged over 65 years ($21.5\%$). Incidence trends in men in this age group showed an increase that continued until the mid-1990s, after which the phenomenon stabilized. In contrast, the elderly female population has seen an almost 1.6-fold increase in incidence over the past three decades. The majority of cancer deaths ($75\%$) occur after the age of 60. The risk of dying from cancer increases with age, reaching a peak in the eighth and ninth decades of life [9]. The third cause of death among the elderly, as in the general population, is respiratory diseases. They accounted for $6.5\%$ of all deaths in 2020 and have shown an upward trend over recent years. Among those aged 65 years and older, respiratory diseases are almost twice as common a cause of death as among those under the age of 65 years. The next most common causes of death in the elderly population are digestive diseases ($2.7\%$) and external causes of mortality ($2.0\%$) [7]. The described changes in the age structure of the population determine the health profile of the society and the nature of challenges facing the health care system. More than $30\%$ of patients using health care are affected by multi-morbidity, which is strongly related to age. The average annual cost of treating a patient over 65 years of age is almost three times higher than that of people in younger age groups. The demand for long-term care services is also growing. This gives rise to the need to look for solutions that will minimize the effects of the population aging i.a. by monitoring and forecasting the health needs of subpopulations in older age groups, separately from the population of younger people and those affected by premature mortality. The aim of this study was to assess mortality trends in Poland between 2000 and 2019 in the early and late old age population. ## Materials and methods The study used data on all deaths of Polish residents aged 65 years or more in the years 2000–2019 ($$n = 5$$,496,970). The database was based on death reports collected and made available for this study by the Department of Information of the Polish Central Statistical Office. Mortality was analyzed in two age groups: early old age (65–74 years) and late old age (over 75 years). The analysis included the five most common major groups of causes of death: diseases of the circulatory system (according to the International Statistical Classification of Diseases and Health-Related Problems—Tenth Revision—ICD-10, coded as I00–I99), malignant neoplasms (C00–C97), diseases of the respiratory system (J00–J99), diseases of the digestive system (K00–K93) and external causes of mortality (V01–Y98). In each group, the most important causes of death were identified: ischemic heart diseases (I20–I25), cerebrovascular diseases (I60–I69), diseases of arteries, arterioles and capillaries (I70–I79), cancers of the lungs and bronchi (C34), stomach (C16), colorectal (C18–C20), breast (C50), prostate (C61), and pancreas (C25), chronic obstructive pulmonary disease (J44), influenza and pneumonia (J09–J18), alcoholic liver disease (K70), transport accidents (V01–V99), falls (W00–W19) and intentional self-harm (X60–X84). The standardized death rates (SDRs) were calculated according to the following formula: where: ki is the number of deaths in this i-age group, pi is population size of this i-age group, wi is the weight assigned to this i-age group, resulting from the distribution of the standard population, N—number of the age groups The standardization procedure was performed using the direct method, in compliance with the European Standard Population, updated in 2012 [10]. The Revised European Standard Population is the unweighted average of the individual populations of EU-27 plus EFTA countries in each five-year age band (with the exception of individuals under the age of five and the highest band, i.e., those aged over 85 years). The analysis of time trends has been carried out with joinpoint models and Joinpoint Regression program, a statistical software package developed by the U.S. National Cancer Institute for the Surveillance, Epidemiology and End Results Program [11]. Joinpoint regression model is an advanced version of linear regression y = bx + a, where b is the slope coefficient, a is the y-intercept, y = ln(z), z is a measure evaluated in the study (SDR) and x is calendar year. Time trends were determined with the use of segments joining in joinpoints, where trend values significantly changed ($p \leq 0.05$). To confirm whether the changes were statistically significant, the Monte Carlo Permutation method was applied. In addition, the authors also calculated the Annual Percentage Change (APC) for each segment of broken lines and the Average Annual Percentage Change (AAPC) for the whole study period with corresponding $95\%$ confidence intervals (CI). The Annual Percent Change is one of the ways to characterize trends in death rates over time and it was calculated according to the following formula: where b is the slope coefficient. With this approach, the death rates are assumed to change at a constant percentage of the rate of the previous year. For example, if the APC is $1\%$, and the rate is 50 per 100,000 in 2,000, the rate is 50 × 1.01 = 50.5 in 2001 and 50.5 × 1.01 = 51.005 in 2002. Rates that change at a constant percentage every year change linearly on a log scale. The Average Annual Percent Change (AAPC) is a summary measure of the trend over a pre-specified fixed interval. It allows us to use a single number to describe the average APCs over a period of many years. It is valid even if the joinpoint model indicates that there were changes in trends during those years. It is computed as a weighted average of the APCs from the joinpoint model, with the weights equal to the length of the APC interval [12]. where bi is the slope coefficient for each segment in the desired range of years and wi corresponds to the length of each segment in the range of years. ## Results The most common major groups of causes of death among Polish residents aged over 65 years were the following: diseases of the circulatory system, malignant neoplasms, diseases of the respiratory system, diseases of the digestive system and external causes of mortality (Table 1). Table 1 \| Sex \| Men \| Men.1 \| Men.2 \| Women \| Women.1 \| Women.2 \| Unnamed: 7 \| Unnamed: 8 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Age group \| 65–74 \| 65–74 \| 75+ \| 75+ \| 65–74 \| 65–74 \| 75+ \| 75+ \| \| Year \| 2000 \| 2019 \| 2000 \| 2019 \| 2000 \| 2019 \| 2000 \| 2019 \| \| Diseases of the circulatory system (I00–I99) including: \| 45.63 \| 35.19 \| 56.22 \| 44.87 \| 48.40 \| 30.82 \| 63.34 \| 51.84 \| \| Ischemic heart diseases (I20–I25) \| 18.01 \| 11.91 \| 16.27 \| 13.05 \| 15.71 \| 9.13 \| 15.62 \| 12.62 \| \| Cerebrovascular diseases (I60–I69) \| 10.28 \| 6.46 \| 12.26 \| 7.31 \| 14.18 \| 6.57 \| 15.22 \| 9.11 \| \| Diseases of arteries, arterioles and capillaries (I70–I79) \| 5.62 \| 4.13 \| 13.28 \| 9.63 \| 5.58 \| 3.46 \| 16.53 \| 13.04 \| \| Malignant neoplasms (C00–C97) including: \| 31.96 \| 33.97 \| 18.90 \| 22.31 \| 29.83 \| 41.84 \| 12.35 \| 14.33 \| \| Malignant neoplasm of bronchus and lung (C34) \| 11.89 \| 10.99 \| 4.77 \| 4.82 \| 3.51 \| 10.24 \| 1.02 \| 1.77 \| \| Malignant neoplasm of stomach (C16) \| 2.58 \| 1.91 \| 1.75 \| 1.26 \| 1.82 \| 1.33 \| 0.94 \| 0.59 \| \| Malignant neoplasm of colon (C18–C20) \| 2.76 \| 4.11 \| 2.01 \| 3.37 \| 3.25 \| 4.07 \| 1.67 \| 2.08 \| \| Malignant neoplasm of breast (C50) \| 0.02 \| 0.04 \| 0.01 \| 0.05 \| 3.21 \| 5.10 \| 1.24 \| 2.14 \| \| Malignant neoplasm of prostate (C61) \| 2.04 \| 2.76 \| 2.59 \| 3.96 \| 0 \| 0 \| 0 \| 0 \| \| Malignant neoplasm of pancreas (C25) \| 1.15 \| 1.63 \| 0.66 \| 0.75 \| 1.66 \| 2.61 \| 0.73 \| 0.87 \| \| Diseases of the respiratory system (J00–J99) including: \| 6.09 \| 6.48 \| 7.85 \| 9.45 \| 4.10 \| 5.99 \| 5.14 \| 6.44 \| \| Chronic obstructive pulmonary disease (J44) \| 2.62 \| 2.02 \| 2.21 \| 2.24 \| 1.14 \| 2.07 \| 0.61 \| 0.97 \| \| Influenza and pneumonia (J09–J18) \| 1.71 \| 3.71 \| 3.68 \| 6.43 \| 1.70 \| 3.12 \| 3.61 \| 4.88 \| \| Diseases of the digestive system (K00–K93) including: \| 3.56 \| 4.58 \| 2.87 \| 2.34 \| 4.07 \| 4.25 \| 3.03 \| 2.63 \| \| Alcoholic liver disease (K70) (K70–K74) \| 0.15 \| 1.68 \| 0.03 \| 0.17 \| 0.04 \| 0.92 \| 0 \| 0.03 \| \| External causes of mortality (V01–Y98) including: \| 3.39 \| 3.61 \| 2.53 \| 2.18 \| 2.23 \| 1.87 \| 2.65 \| 1.92 \| \| Transport accidents (V01–V99) \| 0.90 \| 0.58 \| 0.49 \| 0.28 \| 0.60 \| 0.40 \| 0.25 \| 0.16 \| \| Falls (W00–W19) \| 0.57 \| 0.80 \| 1.05 \| 1.13 \| 0.66 \| 0.53 \| 1.76 \| 1.35 \| \| Intentional self-harm (X60–X84) \| 0.67 \| 0.84 \| 0.34 \| 0.34 \| 0.30 \| 0.29 \| 0.09 \| 0.05 \| The highest percentage of deaths in 2000 in both analyzed age groups (early old age and late old age cohorts), and in both gender groups were deaths caused by diseases of the circulatory system. Over the 20 years analyzed, this percentage decreased in all the subgroups studied. As a consequence, this led to a decrease in differences in death rates between diseases of the circulatory system and the second most common group of malignant neoplasms, while in the group of women in early old age in 2019, malignant neoplasms became the most frequent cause of death (Table 1). The fastest percentage decline in deaths from cardiovascular diseases occurred in the group of women in early old age. The standardized death rate in this group decreased from 1027.3 in 2000 to 471.5 in 2019 (AAPC = −$4.1\%$, $p \leq 0.05$) (Figure 1, Supplementary material 1, 2). In the other subgroups analyzed, AAPC was about −$3.0\%$ in 2019 (Supplementary material 2). In 2019, SDRs were 3,535.7 among women in late old age, 1,152.0 in the early old age male group, and 4,323.8 among men in late old age group (Supplementary material 1). Figure 1: SDR trends in the Polish population aged 65–74 years and 75 years and older due to major groups of causes of death in the years 2000–2019. Among cardiovascular diseases, ischemic heart diseases were the most common cause of death, except for women in late old age, where it was diseases of arteries, arterioles and capillaries (Figure 2). The third most common cause of death in the cardiovascular disease group involved cerebrovascular diseases. In each of the three aforementioned subgroups of causes of death among diseases of the circulatory system, in all the analyzed gender and age subgroups, decreasing trends were observed in the period between 2000 and 2019 (Supplementary material 2). A more detailed trend analysis, however, shows that SDRs due to ischemic heart disease have been increasing for a few years (Figure 2). In the early old age group of women and men, the upward trend began in 2015 and 2016, respectively, and was not statistically significant (APC = 2.1 and $2.0\%$). In the late old age group of women and men, the increase in SDR between 2014 and 2019 was statistically significant, i.e., APC was $3.5\%$ in the female group and $2.5\%$ in the male group (Supplementary material 2). Figure 2: SDR trends in the Polish population aged 65–74 years and 75 years and older due to diseases of the circulatory system in the years 2000–2019. The share of malignant neoplasms among causes of death differed by gender and age. Among women in early late age, SDR was 631.4 in 2000 and decreased until 2006 at a rate of $1.0\%$ ($p \leq 0.05$). After 2006, SDR began to increase (APC = $0.5\%$, $p \leq 0.05$). As a result, the SDR value in 2019 was 628.7 (Supplementary material 1, 2). There was a statistically insignificant decrease in SDR in the group of women in late old age between 2000 and 2011, and a statistically significant increase between 2011 and 2019 (APC = $0.7\%$). As a result, the SDR value decreased from 1,114.6 in 2000 to 1,004.1 in 2011, and then increased to 1,052.4 in 2019. Among men, declining trends in SDR due to malignant neoplasms were observed in both age groups analyzed (Figure 2). A slightly faster decline occurred in the early old age group—from 1,440.1 in 2000 to 1,106.6 in 2019 (AAPC = −$1.3\%$, $p \leq 0.05$). In the late old age group, the SDR values decreased from 2,226.5 in 2000 to 2,099.3 in 2019 (AAPC = −$0.4\%$, $p \leq 0.05$) (Supplementary material 1, 2). Among malignancies, lung and bronchus cancer accounted for the largest share among causes of death, and while SDRs declined gradually in the male group, a continuous increase was observed in the female group (Figure 3). Among women in early old age, SDR increased from 74.2 in 2000 to 153.2 in 2019, with a small and statistically insignificant increase between 2000 and 2005 (APC = $1.5\%$, $p \leq 0.05$). After 2005, SDRs began to increase at a rapid rate of $5.0\%$ ($p \leq 0.05$). Among women in late old age, there was an increase in SDR from 89.4 in 2000 to 105.0 in 2013 (APC = $1.1\%$, $p \leq 0.05$). Between 2013 and 2019, the increase accelerated to $3.5\%$ ($p \leq 0.05$), with SDR reaching 136.7 in 2019. Among men in early old age, the SDR values decreased between 2000 and 2019 from 535.9 to 357.4 (AAPC = −$2.1\%$, $p \leq 0.05$), and in late old age from 520.9 to 450.3 (AAPC = −$0.6\%$, $p \leq 0.05$). Figure 3: SDR trends in the Polish population aged 65–74 years and 75 years and older due to malignant neoplasms in the years 2000–2019. Among women, breast cancer was the second highest SDR cause of death among malignancies in the early old age group and the most common cancer causing death in the late old age group (Figure 3). In the early old age group, SDR increased in years between 2000 and 2014, and then began to decrease. As a result of these changes, the SDR value increased from 67.8 in 2000 to 76.3 in 2019 (AAPC = $0.6\%$, $p \leq 0.05$). In the group of women in late old age, a slight decrease in SDR between 2000 and 2012 (APC = −$0.1\%$, $p \leq 0.05$) was followed by a rapid increase between 2012 and 2019 (APC = $5.1\%$, $p \leq 0.05$). As a result, the SDR value increased from 112.1 in 2000 to 155.2 in 2019 (AAPC = $1.8\%$, $p \leq 0.05$) (Supplementary material 1, 2). Among men, prostate cancer and stomach cancer are the second and third causes of death among malignancies in the late old age group and the third and second causes in the early old age group. The opposite direction of trends was observed for these two cancers among men—increasing for prostate cancer (AAPC = $0.4\%$ in the early old age group and AAPC = $0.6\%$ in the late old age group, $p \leq 0.05$), and decreasing for stomach cancer (AAPC −3.2 and −$2.7\%$, respectively, $p \leq 0.05$). There was also a downward trend observed among women in both analyzed age groups (AAPC −3.2 and −$3.6\%$, respectively, $p \leq 0.05$) (Supplementary material 2). Colorectal cancer mortality trends were stable among women in the early old age group (APC = −$0.3\%$, $p \leq 0.05$) and in the late old age group (APC = −$0.1\%$, $p \leq 0.05$). In the group of elderly men, SDRs due to colorectal cancer increased in the years 2000–2010 at a rate of $1.8\%$ ($p \leq 0.05$), after 2010 they began to decrease at a rate of −$0.8\%$ ($p \leq 0.05$). In the group of elderly men, an increase in SDRs was observed in the years 2000–2016 (APC = $2.0\%$, $p \leq 0.05$) and a statistically insignificant decrease after 2016 (APC = −$1.5\%$, $p \leq 0.05$) (Supplementary material 2). Changes due to the fifth highest SDR cause of death among malignancies—pancreas cancer—were also analyzed. Increasing trends were observed in the early old age group (AAPC = $0.7\%$, $p \leq 0.05$ among women and AAPC = $0.2\%$, $p \leq 0.05$ among men) and decreasing trends in the late old age group (AAPC = −$0.3\%$, $p \leq 0.05$ among women and AAPC = −$0.1\%$, $p \leq 0.05$ among men) (Supplementary material 2). Diseases of the respiratory system are becoming an increasingly common cause of death among women (Figure 1). In the early old age group, SDR increased from 63.1 to 90.9 between 2002 and 2009 (APC = $2.0\%$, $p \leq 0.05$), while in the late old age group, after a decline between 2000 and 2011 from 526.6 to 349.9 (APC = −$2.3\%$, $p \leq 0.05$), an increase to a value of 445.5 in 2019 (APC = $3.1\%$, $p \leq 0.05$) began (Supplementary material 2). The increase in SDR due to diseases of the respiratory system was mainly influenced by influenza and pneumonia (Figure 4). Among early old age women, rates increased by $6.4\%$ annually since 2008 ($p \leq 0.05$), while in the late old age group they increased by $5.1\%$ per year since 2011 ($p \leq 0.05$). For the second most common chronic obstructive pulmonary disease, a stable SDR was observed throughout the analyzed period in the early old age group (APC = $0.1\%$, $p \leq 0.05$) and a decrease in the late old age group (APC = −$1.6\%$, $p \leq 0.05$) (Supplementary material 1). Figure 4: SDR trends in the Polish population aged 65–74 years and 75 years and older due to diseases of the respiratory and digestive system in the years 2000–2019. Among men, SDR values due to diseases of the respiratory system were decreasing. In the early old age group, SDRs decreased from 278.1 in 2000 to 213.4 in 2019 (APC = −$1.4\%$, $p \leq 0.05$). In the late old age group, SDRs decreased from 1092.1 to 910.5 (APC = −$0.5\%$, $p \leq 0.05$) (Supplementary material 1, 2). As in the female group, SDRs from influenza and pneumonia also increased in the male group. In the early old age group, the 2002–2019 APC was $4.0\%$ ($p \leq 0.05$), while in the late old age group, the 2010–2019 APC was $4.3\%$ ($p \leq 0.05$). In contrast, SDRs due to chronic obstructive pulmonary disease decreased, with an AAPC of −$4.6\%$ ($p \leq 0.05$) in the early old age group, and −$3.1\%$ ($p \leq 0.05$) in the late old age group (Supplementary material 2). As for mortality from gastrointestinal diseases, in the last few years of the period studied, increasing trends in mortality were observed in the early old age group of women (as of 2015 APC = $2.5\%$, $p \leq 0.05$) and men (as of 2016 APC = $6.6\%$, $p \leq 0.05$) (Figure 1, Supplementary material 2). It is alcoholic liver disease, the most common cause of death in this disease group, that is responsible for these unfavorable trends. In 2016, in the group of women in early old age, APC was $10.2\%$ ($p \leq 0.05$), while in 2015, in the group of men in early old age, it was $6.6\%$ ($p \leq 0.05$). Moreover, SDRs due to alcoholic liver disease have also begun to increase in recent years in the late old age group. Among women, APC has been $3.7\%$ ($p \leq 0.05$) since 2015, and among men it has been $3.3\%$ ($p \leq 0.05$) since 2016 (Supplementary material 2). The standardized death rates due to external causes decreased from 46.6 in 2000 to 28.1 in 2019 in the early old age group of women (AAPC = −$2.9\%$, $p \leq 0.05$), whereas in the late old age group from 264.5 to 131.8 (AAPC = −$3.7\%$, $p \leq 0.05$) (Figure 1, Supplementary material 1, 2). A decreasing trend in SDR due to transport accidents was observed in both age groups of women (in the early old age group AAPC = −$3.9\%$, $p \leq 0.05$, in the late old age group AAPC = −$3.2\%$, $p \leq 0.05$). Among men, a downward trend occurred between 2000 and 2016 in the early old age group (APC = −$5.0\%$, $p \leq 0.05$) and between 2000 and 2015 in the late old age group (APC = −$5.8\%$, $p \leq 0.05$). After this period, a statistically insignificant increase in SDR began (4.2 and $2.9\%$, respectively) (Figure 5, Supplementary material 2). Figure 5: SDR trends in the Polish population aged 65–74 years and 75 years and older due to external causes in the years 2000–2019. Among external causes of mortality in the late old age groups of both women and men, falls occurred most frequently. In the late old age group of women, a rapid decline in SDR between 2000 and 2009 (APC = −$7.4\%$, $p \leq 0.05$) was followed by a period of stabilization (APC = −$0.3\%$, $p \leq 0.05$). In the late old age group of men, there was a decreasing trend from 2000 to 2009 (APC = −$4.4\%$, $p \leq 0.05$), then an increasing trend from 2009 to 2013 (APC = 3.1, $p \leq 0.05$) and again a decreasing trend from 2013 to 2019 (APC = −$2.3\%$, $p \leq 0.05$) (Supplementary material 2). As for suicides, downward trends were observed in the groups of early and late old age women (both groups AAPC = −$2.8\%$, $p \leq 0.05$) as well as in the group of late old age of men (AAPC = −$0.9\%$, $p \leq 0.05$). In the group of early old-age group men, SDRs increased between 2000 and 2012 at a rate of $1.2\%$ ($p \leq 0.05$), after 2012 they began to decrease at a rate of −$5.5\%$ ($p \leq 0.05$) (Supplementary material 2). In both male age groups, SDRs due to suicide had higher values than those resulting from transport accidents in 2019 (Supplementary material 1). ## Discussion Mortality in the elderly population is influenced by health and non-health determinants, particularly those of psychosocial nature. According to a study by J. S. House, these primarily include anti-health behaviors, such as poor dietary patterns, lack of physical activity, use of stimulants, lack of social contacts and support, stress, and inability to make decisions about one's own life [13]. In order to explain the changes occurring in the mortality pattern of the elderly over the years, the causes of death have to be analyzed and their trends need to be assessed, which was accomplished in this study. The most common cause of death in the population of people aged 65 years and older were cardiovascular diseases. The risk of developing these conditions increases with age [14]. According to data from the American Heart Association on Heart Disease and Stroke Statistics, the incidence of cardiovascular disease among patients aged 40–60 years is on average 35–$40\%$, 60–80 years 75–$78\%$, while among those aged over 80 years it exceeds $85\%$. At the same time, more than $80\%$ of deaths in people aged over 65 years result from cardiovascular causes, and the same percentage of hospitalizations in this age group is due to this group of diseases [15]. The Framingham Heart Study showed a significant relation of an increase in the incidence of coronary heart disease with age, in both men and women [16]. In the Polish population, since 1990, favorable changes in overall mortality in all age groups have been observed, especially in relation to cardiovascular disease, which indicates the effectiveness of preventive measures taken, involving mainly those associated with lifestyle changes, including dietary improvements [17]. In Poland, at the turn of $\frac{1989}{1990}$, a socio-political transformation took place. Food was no longer subsidized after 1990; this caused big changes in relative prices. As a consequence, the structure of food consumed by Polish citizens changed substantially. For example, between 1989 and 2008 annual butter consumption decreased from 7 to 3.8 kg per head, and beef consumption fell by $75\%$. At the same time availability and consumption of fruits increased markedly [18]. According to Bandosz et al. in the period between 1991 and 2005 about $54\%$ of the deaths from coronary heart disease prevented or postponed were attributable to changes in risk factors and $37\%$ to the increased use of evidence based treatments. Most ($41\%$ of the fall in men and $33\%$ in women) were attributable to large decreases in mean cholesterol concentration (declining by 0.4 mmol/L). This fall in deaths concerns changes in mean cholesterol concentration related to diet only and was calculated by subtraction of drug related effects from total effect of mean cholesterol change. The effects of changes in smoking in men were observed also. The prevalence of smoking decreased by $15.7\%$, explaining about $15\%$ of their fall in mortality. Mean systolic blood pressure fell by 2.7 mm Hg in men and by 5.2 mm Hg in women. After subtraction of the effects of treatments for hypertension, these falls in blood pressure explained about $29\%$ of the decrease in mortality in women and $8\%$ of the increase in deaths in men. Increased leisure time physical activity explained about $10\%$ of the decrease in deaths. These gains were partially offset by about 1,810 additional deaths attributable to increases in BMI (−4 and −$5\%$ for men and women, respectively) and prevalence of diabetes (−1 and −$8\%$, respectively) [6]. However, these relatively favorable patterns weren't continued. As indicated by the results of the National Multicentre Health Survey WOBASZ II (2013–2014), the quality of Poles' eating habits, physical activity frequency, prevalence of obesity and overweight aren't satisfactory (19–21). This trend was confirmed in the present study. The favorable trend has been reversed for several years and the values of standardized mortality rates due to ischemic heart diseases (IHD) among the elderly, in all separate age groups in both sexes have increased. Studies show that the foundation of preventive and therapeutic measures among the elderly is regular physical activity [22, 23]. In seniors with ischemic heart disease, appropriate physical exercise effectively slows the progression of the disease and lowers the risk of acute cardiac incidents, commonly referred to as myocardial infarctions, thus reducing the risk of death [24]. Although the percentage of people aged 60–69 who are physically active, meeting the dose of PA required for health recommendations, increased in Poland between 2014 and 2018 from 31.7 to $46.3\%$, age is the determining factor in these trends [25]. In a study conducted in the Czech population, time spent on work-related and recreational physical activity decreased with age, while time spent in sedentary behaviors increased [26]. A study by Biernat and Piatkowska shows that the problem of inactivity begins at the age of 50 years. On average, as many as $48.2\%$ of Polish people aged 50–64 years do not follow the WHO recommendations [27]. In the PolSenior2 study, age was also the most important determinant of declining physical activity [28]. Also, a 2018 report issued by the Central Statistical Office (CSO) confirms that older Polish residents are less active as compared to younger individuals (25.1 vs. $46.4\%$ on average) [4]. This results in the fact the share of physically active people aged 65 years or more remains insufficient and much lower than in other EU countries [29]. Considering how important regular physical activity is for prevention of chronic non-communicable diseases, including ischemic heart disease, it may be assumed that it is this factor that plays a significant role in the unfavorable mortality trends due to IHD, offsetting the impact of favorable changes in other lifestyle components [30]. The worsening trend in mortality in the elderly population from this cause observed in recent years can also be attributed to the significant increase in the prevalence of obesity, diabetes and metabolic syndrome, the co-presence of which significantly increases the risk of death from IHD [31]. Another important health problem whose incidence is closely correlated with age is cancer [32]. In 2019, nearly $50\%$ of cancer deaths were reported in the subpopulation of people aged 65 years and older [33]. According to the authors' results, malignant neoplasms were, as in the general population, the second cause of death in the population aged over 65 years. However, mortality trends have been inconclusive both in general and with regard to individual malignancies. An overall increase in standardized mortality rates from malignant neoplasms in the female group has been observed in recent years, with a concomitant decline in men in both age groups. Studies of cancer mortality trends from 1970 to 2015 conducted in 11 countries around the world confirm these unfavorable Polish trends in comparison with other countries, for which mortality patterns over the past few decades have varied, however, have been more optimistic. They also confirm a significantly faster reduction in mortality levels for men than for women [34]. It is also worth referring to trends in mortality due to specific types of cancer. The analysis showed that bronchus and lung cancers accounted for the largest share among causes of death, and while SDRs declined steadily in the male group, a steady upward tendency was observed in the female group. Similar trends have been observed in most European countries, with decreases in incidence and mortality from lung cancer since 2000. A significant decrease has also been recorded in North America and the United Kingdom [35]. This is due to the decline in smoking prevalence among generations of men. In comparison, among women, smoking prevalence increased in the US and UK after World War II, and in the 1970's in most other countries as well, i.e., in the generation born between the 1930's and 1950's [36]. Moreover, middle-aged and older men were more likely to quit smoking than women. It should also be remembered that lung cancer risk factors translate into morbidity and resulting mortality with a lag of up to even more than 20 years. Therefore, the incidence and mortality of lung cancer in women aged over 65 years in various regions of the world continues to increase [37]. The positive change in the mortality trend among Polish men is also a consequence of the declining prevalence of smoking in all age groups. In contrast to women, where active smoking varies greatly by cohort effect (period of birth in calendar time). The highest smoking rate was observed in the generation of women born between 1940 and 1960. In the population of women born after 1960, smoking prevalence has halved and is now 20.0–$25.0\%$. Exposure to the carcinogens of tobacco smoke, after taking into account the 20-year latency period, accurately explains the trends of lung cancers in older women in Poland, while the observed cohort effect means that the incidence of the disease, and the resulting mortality, still shows an upward trend that will continue for some time in the future [38]. Prostate cancer is also listed among malignancies strongly associated with smoking. It is estimated that by 2040, mortality related to prostate cancer in the general population will double as compared to 2018, reaching 379,005 deaths worldwide. The highest mortality rate will occur in Africa (+$124.4\%$) and Asia ($116.7\%$), while the lowest in Europe (+$58.3\%$) [39]. Currently, prostate cancer mortality trends are not clear-cut and show global territorial variations. In the population of older men, after an increase in mortality occurring until the 1990's, significant declines in SDR from this cause are observed in North American countries, Argentina, Australia and most European countries, except Poland and Russia. The most favorable changes are recorded in Japan. The rates declined between 2002 and 2012 ($9.8\%$), reaching $\frac{61.6}{100}$,000 men in 2012 [40]. Since 2015, the number of deaths from prostate cancer in EU countries has dropped by an average of $7\%$, which is attributed to improved treatment and better diagnosis [41]. Unfortunately, Poland is the only country to which this indicator does not apply, as for the past 5 years there has been a steady increase in mortality due to late diagnosis, among others. A significant number of patients still remain undiagnosed. According to the National Cancer Registry in Poland, the annual rate of increase in incidence is estimated at $2.5\%$, however, the risk of incidence increases markedly after the age of 50, and after the age of 80 the cancer is found in almost $80\%$ of men [42], which explains the increasing trend for prostate cancer observed in our study in men aged over 65 years, in early and late old age. Negative trends have also been observed in the early old age group for pancreas cancer in both men and women. Similar trends have also been observed in younger age groups in the rest of the world. However, the reason for these unfavorable tendencies remains largely unexplained [43]. Beginning in the 1990s, as a result of the introduction of screening tests, early diagnosis and improved treatment, favorable global trends in breast cancer mortality among older women have been observed [44]. At the same time, however, upward trends have been observed in Asian countries. In Japan, the rate of increase in the mortality rate between 1970 and 2015 was 2.2. An upward trend in the mortality rate was also observed in Russia (by $10.3\%$), as well as in Poland [45], which was confirmed in our study in the group of women in late old age. At the time they entered the age of increased risk of developing the disease, preventive measures leading to early detection and high survival rates were not yet as widespread as they are today. The reduction in mortality from breast cancer is influenced by population-based screening programs, participation in which increases the chance of rapid diagnosis and effective treatment. In Poland, the breast cancer screening program began in 2006. By comparison, in the United States it was introduced 20 years earlier. Thus, the current epidemiological picture does not yet show clear unidirectional changes resulting from the participation of Polish women in this program [46]. Our study also analyzed trends in mortality from stomach cancer, showing a decrease in all four age and gender groups. The absolute incidence of stomach cancer has been growing slightly worldwide as a result of an increase in the size and average age of some populations. However, in most countries, the incidence of stomach cancer has declined by about $75\%$ over the past 50 years. Mortality from this cause in all age groups has also declined. In the United States, the mortality rate has dropped from 37 to 6 per 100,000 people. Japan, too, has seen a decline of almost $40\%$. Studies suggest that this is due to early detection of stomach cancer, changes in dietary habits, increased levels of hygiene, reduced tobacco smoking among men and, most importantly, a decrease in the incidence of Helicobacter pylori infection [47]. A study by Ostrowski et al. found an ~$30\%$ lower prevalence of Helicobacter pylori infection in Poland as compared to studies conducted 15 years ago [48]. Our study showed that diseases of the respiratory system were an increasingly common cause of death in the group of women aged over 65 years during the period analyzed, mainly due to an increase in mortality from influenza and pneumonia. Although, in general, decreasing trends in mortality from diseases of the respiratory system were observed in the group of men, the values of standardized mortality rates for influenza and pneumonia were increasing. This unfavorable trend observed in Poland is attributed to the unsatisfactory level of vaccination against influenza and the change in its etiological factor. Year by year, the disease is increasingly caused by the A strain ($78\%$ of cases in 2019), which is responsible for the severe course of the disease and increases the risk of complications such as pneumonia, exacerbation of chronic disease or myocarditis, which become the ultimate cause of death in the elderly. The likelihood of death, as well as severe flu complications requiring hospitalization, increase nearly threefold in people aged over 65 years. Of critical importance in protecting the safety of the elderly is immunization [49]. According to the WHO recommendations, influenza vaccination among the elderly in the WHO European Region should be implemented at $75\%$ of the vaccination status in this age group. Data on the influenza vaccination status of the elderly in EU countries show that it is about $44\%$ on average, but varies from country to country (above $75\%$ in the Netherlands, $43\%$ in Denmark, $68\%$ in the UK, 57.6 in Ireland, $10\%$ in Poland, $6.9\%$ in Latvia and $4.8\%$ in Estonia) [50]. According to data from the National Institute of Public Health - National Institute of Hygiene, the level of influenza vaccination in the population aged over 65 years fluctuated between 2009 and 2018, and unfortunately shows a downward trend from $11.35\%$ in 2009 to the lowest value in 2016–$6.87\%$. In 2018, the percentage of seniors vaccinated against influenza was $8.31\%$ [51]. In Poland, in response to these unfavorable trends, a $50\%$ reimbursement of influenza vaccination for people aged over 65 years was introduced in 2018, while in 2020 it was extended to include free vaccination for people aged over 75 years. Interest in this form of prevention, especially in the senior population, increased during the COVID-19 pandemic, which gives hope that social awareness of the role of immunization in the fight against infectious diseases will gradually improve. Another group of diseases whose incidence increases with age are those of the digestive system. The most common cause of deaths analyzed, accounting for adverse mortality trends in this group, was alcoholic liver disease (ALD). Mortality from this cause continues to be an important public health problem. Globally, alcohol accounts for $7.6\%$ of deaths in men and $4.0\%$ of deaths in women. It is Europe that consumes the most-−10.9 liters per person per year. For the past 25 years or so, average alcohol consumption in Central and Eastern Europe has remained stable, in Western and Southern Europe it has decreased, while in the UK and Finland it has increased [52]. According to data from the National Agency for Solving Alcohol Problems, consumption of $100\%$ alcohol per capita in Poland, despite isolated declines (related to higher rates excise tax introduced in 2009 and 2014, among others) has shown an upward trend. Currently, recorded levels are significantly higher than in the early 1990s [53]. Deaths related to excessive alcohol consumption are most often due to cardiovascular diseases, transport accidents and alcoholic liver disease. In the European Union, $41\%$ of liver disease deaths have alcohol-consumption background, and in $46\%$ the cause is unknown, however, it is likely to be very often related to alcohol as well. The social and economic costs of excessive alcohol consumption are enormous, hence ALD remains a very important civilization challenge [54]. Unfavorable trends regarding total deaths related to alcohol consumption were demonstrated in a study by Zatoński et al. Although the highest mortality rates were recorded in the group of Polish residents aged 45–64 years, the rate of increase in the years 2002–2017 was the fastest in the population of people aged over 65 years, both among men and women (AAPCs were 8.5 and 12.2, respectively). These unfavorable trends can be fully linked to the weakening of alcohol control measures in Poland. At the same time, alcohol-related mortality has decreased in countries such as Russia and Lithuania, where new, stricter methods for controlling alcohol consumption in the population have been introduced [55]. In the United States, between 1999 and 2019, there was also a statistically significant increase in mortality from alcoholic cirrhosis in each of the 10-year age groups analyzed (25–85 years and older). The largest increase also occurred in early old age—in individuals aged 65 to 74 years—and the differences between men and women in this group gradually disappeared to the disadvantage of women [56]. A study on drinking culture among people aged 60–64 years in Poland was the Standardized European Alcohol Survey (RARHA SEAS). The subpopulation covered by it included retirees from the so-called “baby boomers” generation, those born between 1945 and 1964. This is the generation of the post-war demographic peak coinciding with widespread shortages of consumer goods. As they entered adulthood, echoes of the cultural revolution of the 1960's reached Poland and influenced the generation of Polish baby boomers, including in terms of alcohol consumption. Statistics from the 1970's and early 1980's show very high levels of alcohol consumption, which may indicate a risky drinking pattern for many people of this generation [57]. Considering the fact that ALD is diagnosed with a long delay [58], after many years of alcohol dependence, this may explain the unfavorable trends related to alcoholic liver disease among those included in this study. Undoubtedly, these alarming trends in mortality from this cause represent a health challenge aimed at reducing alcohol consumption in Polish society [59], especially in the era of the COVID-19 pandemic, which had a negative impact on its patterns [60]. One of the most important public health problems globally are injuries, resulting from external causes, mainly transport accidents, self-harm and suicide attempts [61]. Data on the incidence of hospital treatment in Europe indicate that the incidence of injuries is bimodal—clearly increasing among both young people and those aged over 60 years, however, with a change in the hierarchy of their causes [62]. In the old population, the share of falls increases, accounting for nearly $69\%$ of outpatient and inpatient treatment for all external causes in EU countries vs. $41\%$ in the under-65 group. In contrast, the share of transport accidents decreases with age (6 vs. $10\%$, respectively). Despite the relatively favorable trends in mortality from falls and transport accidents shown in this study, the risk of death for Polish seniors from each of these causes is higher than the EU average by 23 and $53\%$, respectively [63]. A significant problem in the group of external causes of death among the elderly is suicide [64]. The study including a group of people aged 65 years and more shows that the average suicide rate among older men in Europe is significantly higher than the average among older women [65], as also shown in a study by Law et al. conducted in Australia [66], as well as our own study. At the same time, the mental health of the Polish population is deteriorating. Between 1997 and 2010, the number of people suffering from mental disorders increased [67]. This is particularly worrisome in old age, when deteriorating health with age, multi-morbidity and polypharmacy increase the risk of mental disorders predisposing to a suicidal act [68]. A factor that increases this risk is the moment when people decide to retire. The inability to fulfill oneself at work, as well as deterioration of the financial condition often associated with retirement, increase the risk of depression, from which the risk of death increases with the severity of symptoms [69, 70]. Studies also show that seniors do not report their suicidal intentions and are more likely to make attempts in conditions where intervention is not possible, which is especially true for men [71]. Our own study showed favorable trends in mortality from suicide in all groups except for men in early old age. This calls for special observation in order to take appropriate preventive measures in the male population, for whom SDRs due to suicide had higher values in 2019 than those due to transport accidents in both early and late old age. The results of the Survey of Health, Aging and Retirement in Europe, carried out in 2017 in 27 European countries, indicate that the health status of Polish people aged 50 years and older is significantly worse than that of populations of other countries, such as Sweden, Greece, or Spain. The survey also proves a higher risk of chronic diseases and a faster rate of their increase with age [72]. Our study had some limitations. Quality of the analyses performed on the mortality statistics depend on the completeness and accuracy of the information contained in the death certificate and the proper and precise description of the cause of death. Poland is a country with $100\%$ completeness of death registration. In order to standardize death causes, which are subject to further statistical analyses, it was determined that the doctor who pronounces death is responsible for filling in the death card, into which he or she puts the primary, secondary and direct death cause, whereas qualified teams of doctors are responsible for coding death causes according to the ICD-10 classification. The data relating to 2000 shows that the cause of $24.8\%$ of deaths were inaccurately described. In 2015 this percentage was the highest and amounted to $31.2\%$, after which it steadily decreased and in 2019 it amounted to $27.4\%$.In the majority of cases garbage codes concerned deaths due to cardiovascular diseases. Significantly fewer incorrect codes number concerns other causes of death [73]. However, from the perspective of public health, it is so important to assess the health burden of the elderly population [74]. It will allow for taking appropriate measures aimed at improving the quality of life and gradually increasing the years lived in health. ## Conclusions The percentage of deaths due to diseases of the circulatory system decreased in the studied subgroups but this problem still remains the greatest health risk in the elderly population, primarily due to ischemic heart disease for which growing trends were observed in recent years of analyzed period. Among malignant neoplasms, lung and bronchus cancer accounted for the largest percentage of deaths, for which the analyzed trends were growing among women and decreasing in male group. Unfavorable trends in mortality due to prostate cancer in the group of men in the early old age and due to breast cancer in the group of women in the late old age were observed. Mortality due to stomach cancer was steadily decreasing in all analyzed subgroups. Diseases of the respiratory system are becoming an increasingly common cause of death among women, mainly due to influenza and pneumonia. Increasing trends in mortality due to diseases of the digestive system in women and men in the early old age group have been observed in recent years, due to alcoholic liver disease—the most common cause of death in this disease group. Downward trends of mortality due to external causes, mainly according to suicides, were observed in both gender groups. It is necessary to conduct further research that will allow to diagnose risk and health problems of the elderly subpopulation in order to meet the health burden of the aging society. ## Institutional review board statement The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Bioethics Committee of the Medical University of Lodz on 22 May 2012 No. RNN/$\frac{422}{12}$/KB. ## 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 Bioethics Committee of the Medical University of Lodz, No. RNN/$\frac{422}{12}$/KB. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements. ## Author contributions MB contributed to study design and writing the article. MP conducted the statistical analysis and interpreted the data. All authors participated in the critical revision of the article and approved the final article. ## 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/fpubh.2023.1060028/full#supplementary-material ## References 1. Ward M, May P. Normand Ch, Kenny RA, Nolan A. Mortality risk associated with combinations of loneliness and social isolation findings from the Irish longitudinal study on ageing (TILDA). Age Ageing. (2021.0) 50 1329-35. DOI: 10.1093/ageing/afab004 2. 2.Eurostat, Statisctics Explained,. Population Projections in the EU. Available online at: https://ec.europa.eu/eurostat/statisticsexplained/index.php?title=Population_projections_in_the_EU (accessed September 15, 2022). 3. 3.Eurostat, Statistics Explained,. Population Structure and Aging of the Society. Available online at: https://ec.europa.eu/eurostat/statistics-explained/index.php?oldid=510211 (accessed September 2, 2022). 4. 4.Główny Urzad Statystyczny Urzad Statystyczny w Białymstoku. Sytuacja osób starszych w Polsce w 2020 roku. Warszawa, Białystok (2021).. Sytuacja osób starszych w Polsce w 2020 roku (2021.0) 5. Cheng X, Yang Y, Schwebel DC, Liu Z, Li L, Cheng P. et al. Population ageing and mortality during 1990–2017: a global decomposition analysis. PLoS Med (2020.0) 17 e1003138. DOI: 10.1371/journal.pmed.1003138 6. Bandosz P, O'Flaherty M, Drygas W, Rutkowski M, Koziarek J, Wyrzykowski B. Decline in mortality from coronary heart disease in Poland after socioeconomic transformation: modelling study. BMJ. (2012.0) 344 d8136. DOI: 10.1136/bmj.d8136 7. Statistics Poland. Demographic situation in Poland up to 2020 (2021.0) 8. 8.Organisation for Economic Cooperation and Development. Health at a Glance. (2021) (accessed September 2, 2022).. Health at a Glance (2021.0) 9. 9.Krajowy, Rejestr Nowotworów,. Nowotwory złośliwe ogółem. Available online at: http://onkologia.org.pl/nowotwory-zlosliwe-ogolem-2/#r (accessed August 30, 2022). 10. 10.Eurostat. Revision of the European Standard Population. Available online at: https://ec.europa.eu/eurostat/documents/3859598/5926869/KS-RA-13-028-EN.PDF/e713fa79-1add-44e8-b23d-5e8fa09b3f8f (accessed August 5, 2022). 11. Kim H, Fay MP, Feuer EJ, Midthune DN. Permutation tests for joinpoint regression with applications to cancer rates. Stat Med. (2000.0) 19 335-51. DOI: 10.1002/(sici)1097-0258(20000215)19:3<335::aid-sim336>3.0.co;2-z 12. Surveillance Research Programme. Average Annual Percent Change (AAPC) 13. Herd P, Schoeni RF, House JS. Upstream solutions: does the supplemental security income program reduce disability in the elderly?. Milbank Q. (2008.0) 86 5-45. DOI: 10.1111/j.1468-0009.2007.00512.x 14. Rodgers JL, Jones J, Bolleddu SI, Vanthenapalli S, Rodgers LE, Shah K. et al. Cardiovascular risks associated with gender and aging. J Cardiovasc Dev Dis (2019.0) 6 19. DOI: 10.3390/jcdd6020019 15. Tsao CW, Aday AW, Almarzooq ZI, Alonso A, Beaton AZ, Bittencourt MS. et al. Heart disease and stroke statistics-−2022 update: a report from the American heart association. Circulation (2022.0) 145 e153-e639. DOI: 10.1161/CIR.0000000000001052 16. Mahmood SS, Levy D, Vasan RS, Wang TJ. The framingham heart study and the epidemiology of cardiovascular disease: a historical perspective. Lancet (2014.0) 15 999-1008. DOI: 10.1016/S0140-6736(13)61752-3 17. Pajak A, Jankowski P, Zdrojewski T. The burden of cardiovascular disease risk factors: A current problem. Kardiol Pol. (2022.0) 80 5-15. DOI: 10.33963/KP.a2022.0018 18. Zatonski WA, McMichael AJ, Powles JW. Ecological study of reasons for sharp decline in mortality from ischaemic heart disease in Poland since 1991. BMJ. (1998.0) 316 1047-51. DOI: 10.1136/bmj.316.7137.1047 19. Kwaśniewska M, Pikala M, Bielecki W, Dziankowska-Zaborszczyk E, Rebowska E. Ten-year changes in the prevalence and socio-demographic determinants of physical activity among polish adults aged 20 to 74 years. Results of the national multicenter health surveys WOBASZ (2003–2005) and WOBASZ II (2013–2014). PLoS ONE (2016.0) 11 e0156766. DOI: 10.1371/journal.pone.0156766 20. Stepaniak U, Micek A, Waśkiewicz A, Bielecki W, Drygas W, Janion M. et al. Prevalence of general and abdominal obesity and overweight among adults in Poland. Results of the WOBASZ II study (2013–2014) and comparison with the WOBASZ study (2003–2005). Pol Arch Med Wewn (2016.0) 126 662-671. DOI: 10.20452/pamw.3499 21. Waśkiewicz A, Szcześniewska D, Szostak-Wegierek D, Kwaśniewska M, Pajak A, Stepaniak U. Are dietary habits of the polish population consistent with the recommendations for prevention of cardiovascular disease?. Kardiol Pol. (2016.0) 74 969-77. DOI: 10.5603/KP.a2016.0003 22. Nikitas C, Kikidis D, Bibas A, Pavlou M, Zachou Z, Bamiou DE. Recommendations for physical activity in the elderly population: A scoping review of guidelines. J Frailty Sarcopenia Falls. (2022.0) 7 18-28. DOI: 10.22540/JFSF-07-018 23. Langhammer B, Bergland A, Rydwik E. The importance of physical activity exercise among older people. Biomed Res Int. (2018.0) 5 7856823. DOI: 10.1155/2018/7856823 24. Dobrosielski DA. How can exercise reduce cardiovascular disease risk? A primer for the clinician. Pol Arch Intern Med. (2021.0) 131 16122. DOI: 10.20452/pamw.16122 25. Biernat E, Piatkowska M. Leisure-time physical activity participation trends 2014–2018: a cross-sectional study in Poland. Int J Environ Res Public Health (2019.0) 17 208. DOI: 10.3390/ijerph17010208 26. Hamrik Z, Sigmundová D, Kalman M, Pavelka J, Sigmund E. Physical activity and sedentary behaviour in Czech adults: results from the GPAQ study. Eur J Sport Sci. (2014.0) 14 193-8. DOI: 10.1080/17461391.2013.822565 27. Biernat E, Piatkowska M. Stay active for life: physical activity across life stages. Clin Interv Aging. (2018.0) 13 1341-52. DOI: 10.2147/CIA.S167131 28. Błedowski P, Grodzicki T, Mossakowska M, Zdrojewski T. POLSENIOR 2. Badanie poszczególnych obszarów stanu zdrowia osób starszych, w tym jakości zycia zwiazanej ze zdrowiem (2021.0) 29. European Commission. Eurobarometer. Sport and Physical Activity 30. Virtanen M, Vahtera J, Singh-Manoux A, Elovainio M, Ferrie JE, Kivimäki M. Unfavorable and favorable changes in modifiable risk factors and incidence of coronary heart disease: the Whitehall II cohort study. Int J Cardiol. (2018.0) 269 7-12. DOI: 10.1016/j.ijcard.2018.07.005 31. Nichols M, Townsend N, Scarborough P, Rayner M. Trends in age-specific coronary heart disease mortality in the European Union over three decades: 1980–2009. Eur Heart J. (2013.0) 34 3017-27. DOI: 10.1093/eurheartj/eht159 32. Pikala M, Burzyńska M, Maniecka-Bryła I. Years of life lost due to cervical cancer in Poland in 2000 to 2015. Int J Environ Res Public Health. (2019.0) 16 1545. DOI: 10.3390/ijerph16091545 33. Pilleron S, Sarfati D, Janssen-Heijnen M. Global cancer incidence in older adults, 2012 and 2035: a population-based study. Int J Cancer. (2019.0) 144 49-58. DOI: 10.1002/ijc.31664 34. Carioli G, Malvezzi M, Bertuccio P, Hashim D, Waxman S, Negri E. et al. Cancer mortality in the elderly in 11 countries worldwide, 1970–2015. Ann Oncol (2019.0) 30 1344-55. DOI: 10.1093/annonc/mdz178 35. Levi F, Lucchini F, Negri E, Boyle P, LaVecchia C. Changed trends of cancer mortality in the elderly. Ann Oncol. (2001.0) 12 1467-77. DOI: 10.1023/A:1012539213643 36. Malvezzi M, Carioli G, Bertuccio P, Negri E, La Vecchia C. Relation between mortality trends of cardiovascular diseases and selected cancers in the European Union, in 1970–2017. Focus on cohort and period effects. Eur J Cancer. (2018.0) 103 341-55. DOI: 10.1016/j.ejca.2018.06.018 37. Bosetti C, Malvezzi M, Rosso T, Bertuccio P, Gallus S, Chatenoud L. Lung cancer mortality in European women: trends and predictions. Lung Cancer. (2012.0) 78 171-8. DOI: 10.1016/j.lungcan.2012.08.008 38. Zatoński W, Przewozniak K. Tobacco smoking in Poland in the years 1923–1987. Pol Popul Rev. (1993.0) 3 103-24. PMID: 28224973 39. Global Cancer Observatory: Cancer Tomorrow 40. Rawla P. Epidemiology of prostate cancer. World J Oncol. (2019.0) 10 63-89. DOI: 10.14740/wjon1191 41. Mottet N, Bellmunt J, Bolla M, Briers E, Cumberbatch MG, De Santis M. EAU-ESTRO-SIOG guidelines on prostate cancer. Part 1: screening, diagnosis, and local treatment with curative. Int Eur Urol. (2017.0) 71 618-29. DOI: 10.1016/j.eururo.2016.08.003 42. Nowotwory złośliwe gruczołu krokowego 43. Lucas AL, Malvezzi M, Carioli G, Negri E, La Vecchia C, Boffetta P. Global trends in pancreatic cancer mortality from 1980 through 2013 and predictions for 2017. Clin Gastroenterol Hepatol. (2016.0) 14 1452-62. DOI: 10.1016/j.cgh.2016.05.034 44. Carioli G, Malvezzi M, Rodriguez T, Bertuccio P, Negri E, La Vecchia C. Trends and predictions to 2020 in breast cancer mortality: Americas and Australasia. Breast. (2018.0) 37 163-9. DOI: 10.1016/j.breast.2017.12.004 45. Carioli G, Malvezzi M, Rodriguez T et al. Trends and predictions to 2020 in breast cancer mortality in Europe. Breast. (2017.0) 36 89-95. DOI: 10.1016/j.breast.2017.06.003 46. Malinowska M, Nasierowska-Guttmejer A. Epidemiologia i patogeneza raka zoładka. Pol J Pathol. (2013.0) 64 17-26 47. Ostrowski B, Budzyńska A, Janczewska A, Gabryelczak M, Kadłubicki B. Zapadalność i czynniki ryzyka infekcji Helicobacter pylori wśród krewnych pacjentów zakazonych. Pediatr Med Rodz. (2017.0) 13 377-83. DOI: 10.15557/PiMR.2017.0040 48. Lee JK, Lam GK, Shin T, Kim J, Krishnan A, Greenberg DP. Efficacy and effectiveness of high-dose vs. standard-dose influenza vaccination for older adults: a systematic review and meta-analysis. Expert Rev Vaccines. (2018.0) 17 435-43. DOI: 10.1080/14760584.2018.1471989 49. 44% of elderly people vaccinated against influenza 50. Jaki jest poziom zaszczepienia przeciw grypie w Polsce? 51. 51.World Health Organization. Global status report on alcohol and health 2014. Geneva (2014).. Global status report on alcohol and health 2014 (2014.0) 52. Zgliczyński WS. Alkohol w Polsce. Infos (2016.0) 11 1-4 53. Hartleb M. Choroby watroby zwiazane z alkoholem. Omówienie wytycznych European Association for the Study of the Liver 2018 oraz wytycznych American College of Gastroenterology. Med Prakt (2019.0) 54. Zatonski WA, Zatonski M, Janik-Koncewicz K, Wojtyla A. Alcohol-related deaths in Poland during a period of weakening alcohol control measures. JAMA. (2021.0) 325 1108-9. DOI: 10.1001/jama.2020.25879 55. Yoon YH, Chen CM, Slater ME, Jung MK, White AM. Trends in premature deaths from alcoholic liver disease in the U.S.. Am J Prev Med. (2020.0) 59 469-80. DOI: 10.1016/j.amepre.2020.04.024 56. Zin-Sesek M. Rozpowszechnienie i style picia alkoholu wśród osób starszych na podstawie wyników Standaryzowanego Europejskiego Sondazu na temat Alkoholu (RARHA SEAS). Alcohol Drug Addict. (2019.0) 32 1-24. DOI: 10.5114/ain.2019.85765 57. Torruellas C, French SW, Medici V. Diagnosis of alcoholic liver disease. World J Gastroenterol (2014.0) 20 11684-99. DOI: 10.3748/wjg.v20.i33.11684 58. Julien J, Ayer T, Bethea ED, Tapper EB, Chhatwal J. Projected prevalence and mortality associated with alcohol-related liver disease in the USA, 2019-40: a modelling study. Lancet Public Health (2020.0) 5 e316-e323. DOI: 10.1016/S2468-2667(20)30062-1 59. Szajnoga D, Klimek-Tulwin M, Piekut A. COVID-19 lockdown leads to changes in alcohol consumption patterns. Results from the Polish national survey. J Addict Dis (2021.0) 39 215-25. DOI: 10.1080/10550887.2020.1848247 60. Imm P, Grogan B, Diallo O. Self-harm injury hospitalisations: an analysis of case selection criteria. Inj Prev. (2021.0) 27 i49-55. DOI: 10.1136/injuryprev-2019-043514 61. Sirven N, Rapp T. The dynamics of hospital use among older people evidence for Europe using share data. Health Serv Res. (2017.0) 52 1168-84. DOI: 10.1111/1475-6773.12518 62. 62.European Association for Injury Prevention Safety Promotion. Falls Among Older Adults in the EU-28: Key Facts from the Available Statistics. Available online at: https://www.eurosafe.eu.com/uploads/inline-files/POLICY%20BRIEFING%2020_Facts%20on%20falls%20in%20older%20adults%20in%20EU.pdf (accessed September 1, 2022). 63. De Leo D. Late-life suicide in an aging world. Nat Aging (2022.0) 2 7-12. DOI: 10.1038/s43587-021-00160-1 64. Wu J, Li Y, Waern M. Suicide among older people in different European welfare regimes: does economic (in)security have implications for suicide prevention?. Int J Environ Res Public Health. (2022.0) 19 7003. DOI: 10.3390/ijerph19127003 65. Law CK, Kõlves K, De Leo D. Influences of population-level factors on suicides in older adults: a national ecological study from Australia. Int J Geriatr Psychiatry. (2015.0). DOI: 10.1002/gps.4343 66. Lubecka B, Lubecki M, Kasperczyk J, Jośko-Ochojska J, Pudlo R. Risk modifying factors of anxiety and depressive disorders, using the example of a population study in the Zywiec District. Int J Environ Res Public Health. (2021.0) 18 10248. DOI: 10.3390/ijerph181910248 67. Huang L, Shujuan X, Lei S, Xiao Z, Yaqing X, Qilong Y. Impact of multimorbidity on symptoms of depression, anxiety, and stress in older adults: is there a sex difference?. Front Psychol. (2021.0) 12 762310. DOI: 10.3389/fpsyg.2021.762310 68. Odone A, Gianfredi V, Vigezzi GP, Amerio A, Ardito C. d'Errico A, Stuckler D, Costa G; Italian Working Group on Retirement and Health. Does retirement trigger depressive symptoms? A systematic review and meta-analysis. Epidemiol Psychiatr Sci. (2021.0) 30 e77. DOI: 10.1017/S2045796021000627 69. Zuelke AE, Roehr S, Schroeter M, Witte AV, Hinz A, Glaesmer H. Depressive symptomatology in early retirees associated with reason for retirement—results from the population-based life-adult-study. Front Psychiatry. (2020.0) 11 565442. DOI: 10.3389/fpsyt.2020.565442 70. Gramaglia C, Martelli M, Scotti L, Bestagini L, Gambaro E, Romero M. Attempted suicide in the older adults: a case series from the psychiatry ward of the university hospital maggiore Della Carità, Novara, Italy. Front Public Health. (2022.0) 9 732284. DOI: 10.3389/fpubh.2021.732284 71. Acciai F. The age pattern of social inequalities in health at older ages: are common measures of socio-economic status interchangeable?. Public Health. (2018.0) 157 135-41. DOI: 10.1016/j.puhe.2018.01.002 72. Cierniak M. Na co Umarł Pacjent–Czyli, co Jest Wpisywane na Kartach Zgonów?. (2019.0) 73. Burzynska M, Bryla M, Bryla P, Maniecka-Bryla I. Factors determining the use of social support services among elderly people living in a city environment in Poland. Health Soc Care Commun. (2016.0) 24 758-68. DOI: 10.1111/hsc.12259 74. Tang B, Li Z, Hu S, Xiong J. Economic implications of health care burden for elderly population. Inquiry (2022.0) 59 1-13. DOI: 10.1177/00469580221121511
--- title: 'Associations of metal mixtures with metabolic-associated fatty liver disease and non-alcoholic fatty liver disease: NHANES 2003–2018' authors: - Zhilan Xie - Ruxianguli Aimuzi - Mingyu Si - Yimin Qu - Yu Jiang journal: Frontiers in Public Health year: 2023 pmcid: PMC10025549 doi: 10.3389/fpubh.2023.1133194 license: CC BY 4.0 --- # Associations of metal mixtures with metabolic-associated fatty liver disease and non-alcoholic fatty liver disease: NHANES 2003–2018 ## Abstract ### Objective The hepatotoxicity of exposure to a single heavy metal has been examined in previous studies. However, there is limited evidence on the association between heavy metals mixture and non-alcoholic fatty liver disease (NAFLD) and metabolic-associated fatty liver disease (MAFLD). This study aims to investigate the associations of 13 urinary metals, individually and jointly, with NAFLD, MAFLD, and MAFLD components. ### Methods This study included 5,548 adults from the National Health and Nutrition Examination Survey (NHANES) 2003–2018. Binary logistic regression was used to explore the associations between individual metal exposures and MAFLD, NAFLD, and MAFLD components. Bayesian kernel machine regression (BKMR) and Quantile-based g-computation (QGC) were used to investigate the association of metal mixture exposure with these outcomes. ### Results In single metal analysis, increased levels of arsenic [OR 1.09 ($95\%$CI 1.03–1.16)], dimethylarsinic acid [1.17 ($95\%$CI 1.07–1.27)], barium [1.22 ($95\%$CI 1.14–1.30)], cobalt [1.22 ($95\%$CI 1.11–1.34)], cesium [1.35 ($95\%$CI 1.18–1.54)], molybdenum [1.45 ($95\%$CI 1.30–1.62)], antimony [1.18 ($95\%$CI 1.08–1.29)], thallium [1.49 ($95\%$CI 1.33–1.67)], and tungsten [1.23 ($95\%$CI 1.15–1.32)] were significantly associated with MAFLD risk after adjusting for potential covariates. The results for NAFLD were similar to those for MAFLD, except for arsenic, which was insignificantly associated with NAFLD. In mixture analysis, the overall metal mixture was positively associated with MAFLD, NAFLD, and MAFLD components, including obesity/overweight, diabetes, and metabolic dysfunction. In both BKMR and QGC models, thallium, molybdenum, tungsten, and barium mainly contributed to the positive association with MAFLD. ### Conclusion Our study indicated that exposure to heavy metals, individually or cumulatively, was positively associated with NAFLD, MAFLD, and MAFLD components, including obesity/overweight, diabetes, and metabolic dysfunction. Additional research is needed to validate these findings in longitudinal settings. ## 1. Introduction Non-alcoholic fatty liver disease (NAFLD), the leading cause of cirrhosis and hepatocellular carcinoma, prevails worldwide with an estimated prevalence of $32.4\%$ [1]. The prevalence of NAFLD has increased parallel to the increasing prevalence of obesity, type 2 diabetes, and other metabolic syndromes [2]. Recently, experts from European Liver Patients Association (ELPA) proposed a new nomenclature, metabolic associated fatty liver disease (MAFLD), to replace the earlier NAFLD term [3]. The new nomenclature changes the emphasis from excluding other liver diseases or excessive alcohol consumption to identifying cases with concomitant metabolic dysfunction [4]. This change in definition affects the prevalence, risk factors, and outcomes of these two diseases [5, 6]. The pooled prevalence of MAFLD was reported to be $39.22\%$, which was higher than that of NAFLD at $33.86\%$ [5]. Compared with NAFLD, MAFLD tended to be more closely associated with obesity, diabetes, and high fibrosis scores [5]. Emerging data have also suggested that patients with MAFLD tended to have higher risks of cardiovascular disease and all-cause mortality than those with NAFLD [6]. Furthermore, similar to the increasing trend of NAFLD, the prevalence of MAFLD in the United States also increased from $34.4\%$ in 2011 to $38.1\%$ in 2018 [7]. Given their increasing prevalence and adverse outcomes, identifying determinants of MAFLD and NAFLD is of substantial public health interest. NAFLD and MAFLD are heterogeneous disorders with genetic and environmental factors involved in their pathogenesis and progression. Beyond dietary factors and physical activity, previous animal and human studies suggested that heavy metals may play essential roles in the etiology of NAFLD and MAFLD [8, 9]. Heavy metals are metallic elements with high density and atomic weight and have adverse health impacts on humans [10]. Heavy metal exposure is widespread in humans due to its various sources, including the atmosphere, domestic effluents, industrial waste, and agriculture [11, 12]. Heavy metals threaten human health because they are non-biodegradable and can be deposited in body tissues or organs to produce harm after initial exposure [13]. Toxicology studies have shown that heavy metals (lead, cadmium, and arsenic) could disturb the hypothalamic dopaminergic system and endoplasmic reticulum proteostasis [14], impair adipogenesis and adipocytokines secretion, and induce hepatic inflammation and steatosis [15, 16]. In addition, exposure to heavy metals was a risk factor for many metabolic abnormalities, such as diabetes [17], metabolic syndrome [18], obesity, and hypertension [17]. Several previous studies reported positive associations between mercury [19], arsenic [20], lead [9], cadmium [21], and metal mixture [22, 23] with NAFLD. However, the evidence of the association between heavy metals and the risk of MAFLD is limited. The exact physiological roles of other metals in MAFLD and NAFLD patients are still unknown. Additionally, previous studies on metals generally evaluated the influence of single metals, but this approach could not reflect the reality that individuals are exposed to multiple metals simultaneously. To fill these knowledge gaps, we conducted this study among US adults who participated in the National Health and Nutrition Examination Survey (NHANES) 2003–2018 survey cycles to examine the associations of urinary metal mixtures and individual metals with MAFLD, NAFLD, and MAFLD components using Bayesian Kernel Machine Regression (BKMR) and Quantile based g-computation (QGC). Our study might be informative and instructive for MAFLD and NAFLD etiology and prevention. ## 2.1. Study population The NHANES examines a representative sample of the resident population across the United States, combining interviews and physical examinations [24]. Written informed consent was obtained from each participant, and the NHANES protocol was approved by the National Center for Health Statistics (NCHS) Institutional Review Board. The current study is based on an analysis of data from the combined eight continuous NHANES survey cycles (2003–2018). Two earlier cycles of NHANES (1999–2000 and 2001–2002) were not included because arsenic species were not measured in those cycles. Urinary measurements of heavy metals were taken from 14,058 adults 20 years of age or older. We excluded 8,510 participants due to the missing data in [1] the calculation of the U.S. fatty liver index (USFLI) (remaining, $$n = 6$$,068), [2] covariates (i.e., ratio of family income to poverty, education, smoking status, and physical activity) (remaining, $$n = 5$$,548), leaving 5,548 participants for the analyses of MAFLD. The NAFLD analysis sample additionally excluded individuals with excessive alcohol consumption ($$n = 725$$), positive HBV surface antigen ($$n = 29$$), and positive HCV RNA ($$n = 66$$). The final sample size for NAFLD analyses was 4,750. The flow chart of the inclusion and exclusion criteria for the sample population was presented in Supplementary Figure 1. ## 2.2. Measurements of heavy metals A total of 14 heavy metals were measured in urine, including total arsenic (As), arsenobetaine (Asb), dimethylarsinic acid (DMA), monomethylarsonic acid (MMA), barium (Ba), cadmium (Cd), cobalt (Co), cesium (Cs), mercury (Hg), lead (Pb), molybdenum (Mo), antimony (Sb), thallium (Tl), and tungsten (W). Arsenic is a metalloid rather than a heavy metal. DMA, Asb, and MMA are metabolites of As. However, As might induce toxic effects by combining and inactivating sulfhydryl enzymes similar to heavy metals [10, 25]. Thus, we also listed As, Asb, DMA, and MMA as heavy metals according to previous studies [17, 26]. Detailed information on the sample preparations and detection methods was summarized in Supplementary Table 1 and previously published elsewhere [27]. The detection rate (%) and limit of detection (LOD) of heavy metals are presented in Supplementary Table 2. Metal concentration below the LOD was recorded as the LOD divided by the square root of two. Urinary creatinine was measured by Jaffé rate reaction before 2010 and Enzymatic Roche Cobas 6000 Analyzer in later research cycles. ## 2.3. Assessment of outcome We used USFLI, a well-validated steatosis score, to define hepatic steatosis as a substitute for the liver biopsy [28, 29]. The details of the calculation formula are presented in Supplementary material. Hepatic steatosis was defined to be present in the USFLI score ≥30 [29]. This cut-off point has been previously validated with a sensitivity and specificity of 62 and $88\%$, respectively [30]. MAFLD was defined by the presence of hepatic steatosis, demonstrated by serologic score (USFLI ≥ 30), with at least one of the MAFLD components [4]: overweight/obesity [body mass index (BMI) ≥25 kg/m2], diabetes mellitus [fasting glucose levels ≥7 mmol/L, or hemoglobin A1c (HbA1c) ≥$6.5\%$, or 2-h post-load plasma glucose levels (2h-OGTT) ≥11 mmol/L], and metabolic dysfunction (at least two metabolic risk abnormalities). The diagnostic criteria of metabolic abnormalities are displayed in Supplementary material [31]. NAFLD was defined as the USFLI ≥ 30 in the absence of viral hepatitis (HBV or HCV) and excessive consumption history of alcohol (alcohol consumption ≥$\frac{30}{20}$ g/d for men and women) [28]. The difference between was summarized in Supplementary Table 3. ## 2.4. Statistical analysis The proportions of categorical variables and median (inter-quantile range, IQR) of continuous variables are presented among comparison groups. We compared baseline characteristics using Chi-square tests for categorical variables and the Mann-Whitney U-test for continuous variables. Before association analysis, we adopted a covariate-adjusted standardization method to adjust for the urine dilution of urinary metal, which was generally applied by previous studies due to lower statistical bias than former methods [32, 33]. In this approach, log-transformed creatinine was first regressed on the variables [race, gender, age (in years), and BMI were included in the present study] known to affect urine dilution [34]. Then a ratio is produced by dividing observed creatinine values by the predicted creatinine values obtained from the previous model. Finally, we standardized metal concentration by dividing the biomarker concentration by this ratio. Binary logistic regression, Bayesian Kernel Machine Regression (BKMR), and quantile-based g-computation (QGC) were then applied to evaluate the associations below. ## 2.4.1. Binary logistic regression Binary logistic regression models were applied to evaluate the associations of individual metals with NAFLD, MAFLD, and its components. In regression models, dilution-adjusted heavy metals were modeled as continuous (Ln-transformed) and categorical (i.e., quartiles). A linear trend test was performed by modeling the categorized metals as ordinal variables. Given the sex difference in the prevalence of NAFLD and MAFLD [35], these association analyses were further stratified by sex. ## 2.4.2. Bayesian kernel machine regression BKMR was implemented to estimate the joint and potential non-linear association of metal exposure with MAFLD, NAFLD, and MAFLD components. BKMR is a statistical approach combining Bayesian and statistical learning methods to investigate mixed exposure-response functions using a Gaussian Kernel function [36]. This approach is a non-parameter statistical method without hypothesis testing but visualizes the exposure-response associations of each chemical and the joint influence of all chemicals. The probit BKMR model was applied to binary outcomes. The core function formula in this study is presented as follows: Where Φ−1 is a probit link function and (Yi = 1) represents the probability of the relative outcome. Other covariates and their coefficients are denoted by xi and β, respectively. The function h() represents the exposure-response function considering the non-linear and interactive relationship between exposure and a latent continuous outcome (>0 equal to MAFLD or NAFLD, < 0 equal to non-MAFLD or non-NAFLD). A possible interpretation of h(z) in the probit BKMR model could be the correlation between metal exposures and a latent outcome. We applied the option of variable selection and 20000 iterations by the Markov Chain Monte Carlo algorithm. A posterior inclusion probability (PIP) was calculated to evaluate the relative importance of metal exposure to health outcomes [37]. The BKMR results contain univariate exposure-response and cumulative mixture exposure relationships. ## 2.4.3. Quantile-based g-computation To validate the association of exposure to multiple metals with the outcomes, we employed the QGC for mixture analyses. QGC is a parameter-based statistical method that combines weighted quantile sum regression (WQS) and g-computation. Compared to WQS, QGC has particular advantages in allowing for directional heterogeneity and non-linear or non-additive effects of components of the mixture [38]. This novel strategy was used to estimate the change in MAFLD, NAFLD, and MAFLD components risk for a synchronous one-quartile increase for all 13 heavy metals. The plot depicts heavy metal and health outcomes prediction at the joint exposure levels via g-computation and bootstrap variance with bootstrap up to 200. ## 2.4.4. Covariates We selected covariates based on a priori as potential confounders [20, 39, 40]. Age at interview (“≥20 and <40,” “≥40 and <60,” “≥60”), gender (male, female), education (High school or less, College, Graduate or higher), race/ethnicity (Hispanic, Non-Hispanic White, Non-Hispanic Black, Other), smoking status, and physical activity were collected using self-administered questionnaires. The ratio of family income to poverty (PIR) was calculated by dividing annual family income by the poverty threshold and dichotomized (PIR < 1 and PIR ≥ 1) for analysis. Smoking status was grouped into three categories: current smoker (smoking at least 100 cigarettes in lifetime, and smoking every day or some days at the time of interview), former smoker (smoking at least 100 cigarettes in lifetime, but not smoking at the time of interview), and never smoker (having not smoked 100 cigarettes during life). The participants engaged in vigorous or moderate recreational activities were identified as having regular physical activity. Diabetes, hypertension, BMI, HDL cholesterol, and high TG were not adjusted due to application in MAFLD diagnosis. The samples were weighted to reduce the selection bias among subgroups for age, gender, and race/ethnicity in the NHANES survey. However, these variables for calculating sample weights are already included in the models, especially for BKMR and QGC models. Therefore, as recommended, we used unweighted estimation for the main results and logistic models incorporating sampling weights in the sensitivity analysis [22, 41, 42]. ## 2.4.5. Sensitivity analyses To test the robustness of our results, we conducted several sensitivity analyses. First, we used the natural log-transformed heavy metals in regression models to validate the role of creatinine. Second, we reanalyzed the regression model accounting for sample design, sampling weights, and strata. NHANES selected representative participants using a complex, multistage, and probability sampling design. Specifying the sampling design parameters (including sample weights) should be considered to reduce biased estimates [43]. For combing multiple survey cycles, the sample weight was calculated by “WTMEC2YR (variable name of weight)/n (the number of survey circles)” [43]. SAS PROC SURVEYLOGISTIC was used for logistic regression analyses while incorporating survey design. BKMR and QGC methods do not support the survey design, and these analyses were limited to conventional binary logistic regression. Statistics analyses were performed using SAS 9.4 (SAS Institute Inc., Cary, NC) and R 4.1.1 [44]. BKMR and QGC were conducted using “bkmr” and “qgcomp” packages, respectively. The P-value was 0.05 for the significance level. ## 3.1. Population characteristics A total of 1,811 ($32.6\%$) and 1,624 ($34.2\%$) individuals were diagnosed with MAFLD and NAFLD, respectively. Participants with MAFLD were older, less educated, Hispanic, and more likely to be current/past smokers and less physical activity than those without MAFLD (Table 1). The characteristics of participants with NAFLD were similar to those with MAFLD. Cases with MAFLD or NAFLD had a higher prevalence of MAFLD components (i.e., diabetes, overweight/obesity, and metabolic dysfunction). Table 1 \| Variables \| Non-MAFLD \| MAFLD \| Pa -value \| Non-NAFLD \| NAFLD \| Pa -value.1 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Demographic variable \| Demographic variable \| Demographic variable \| Demographic variable \| Demographic variable \| Demographic variable \| Demographic variable \| \| Age, n (%) \| \| \| 0.001 \| \| \| <0.001 \| \| ≥20 and <40 \| 1,470 (39.3%) \| 442 (24.4%) \| \| 1,225 (39.2%) \| 388 (23.9%) \| \| \| ≥40 and <60 \| 1,171 (31.3%) \| 617 (34.1%) \| \| 929 (29.7%) \| 523 (32.2%) \| \| \| ≥60 \| 1,096 (29.3%) \| 752 (41.5%) \| \| 972 (31.1%) \| 713 (43.9%) \| \| \| Sex, male, n (%) \| 1,727 (46.2%) \| 1,030 (56.9%) \| < 0.001 \| 1,379 (44.1%) \| 900 (55.4%) \| <0.001 \| \| Ratio of family income to poverty, ≥1, n (%) \| 3,019 (80.8%) \| 1,429 (78.9%) \| 0.107 \| 2,511 (80.3%) \| 1,272 (78.3%) \| 0.113 \| \| Education, n (%) \| \| \| <0.001 \| \| \| <0.001 \| \| High school or less \| 1,663 (44.5%) \| 1,007 (55.6%) \| \| 1,414 (45.2%) \| 904 (55.7%) \| \| \| College \| 1,092 (29.2%) \| 527 (29.1%) \| \| 904 (28.9%) \| 478 (29.4%) \| \| \| Graduate or higher \| 982 (26.3%) \| 277 (15.3%) \| \| 808 (25.8%) \| 242 (14.9%) \| \| \| Race, n (%) \| \| \| <0.001 \| \| \| <0.001 \| \| Hispanic \| 760 (20.3%) \| 643 (35.5%) \| \| 659 (21.1%) \| 591 (36.4%) \| \| \| Non-Hispanic White \| 1,664 (44.5%) \| 806 (44.5%) \| \| 1,348 (43.1%) \| 714 (44.0%) \| \| \| Non-Hispanic Black \| 885 (23.7%) \| 231 (12.8%) \| \| 750 (24.0%) \| 194 (11.9%) \| \| \| Other race \| 428 (11.5%) \| 131 (7.23%) \| \| 369 (11.8%) \| 125 (7.70%) \| \| \| Smoking, n (%) \| \| \| <0.001 \| \| \| <0.001 \| \| Never \| 2,132 (57.1%) \| 877 (48.4%) \| \| 1,886 (60.3%) \| 827 (50.9%) \| \| \| Former \| 819 (21.9%) \| 595 (32.9%) \| \| 669 (21.4%) \| 513 (31.6%) \| \| \| Current \| 786 (21.0%) \| 339 (18.7%) \| \| 571 (18.3%) \| 284 (17.5%) \| \| \| Have regular physical activity, yes, n (%) \| 2,098 (56.1%) \| 765 (42.2%) \| <0.001 \| 1,738 (55.6%) \| 679 (41.8%) \| <0.001 \| \| MAFLD components, yes, n (%) \| MAFLD components, yes, n (%) \| MAFLD components, yes, n (%) \| MAFLD components, yes, n (%) \| MAFLD components, yes, n (%) \| MAFLD components, yes, n (%) \| MAFLD components, yes, n (%) \| \| Diabetes \| 342 (9.15%) \| 625 (34.5%) \| <0.001 \| 317 (10.1%) \| 561 (34.5%) \| <0.001 \| \| Overweight/obesity \| 2,097 (56.3%) \| 1,778 (98.2%) \| <0.001 \| 1,809 (58.0%) \| 1,549 (95.6%) \| <0.001 \| \| Metabolic dysfunction \| 2,331 (62.4%) \| 1,791 (98.9%) \| <0.001 \| 1,976 (63.2%) \| 1,607 (99.0%) \| <0.001 \| \| High C-reaction protein \| 1,120 (40.3%) \| 920 (66.7%) \| <0.001 \| 950 (41.0%) \| 837 (67.7%) \| <0.001 \| \| Central obesity \| 1,545 (41.3%) \| 1,584 (87.5%) \| <0.001 \| 1,355 (43.3%) \| 1,400 (86.2%) \| <0.001 \| \| Insulin resistance \| 971 (26.0%) \| 1,731 (95.6%) \| <0.001 \| 821 (26.3%) \| 1,557 (95.9%) \| <0.001 \| \| Low HDL cholesterol \| 698 (18.7%) \| 759 (41.9%) \| <0.001 \| 610 (19.5%) \| 716 (44.1%) \| <0.001 \| \| Hypertension \| 1,400 (38.3%) \| 1,128 (63.1%) \| <0.001 \| 1,177 (38.5%) \| 1,000 (62.5%) \| <0.001 \| \| Prediabetes \| 1,588 (44.0%) \| 1,126 (76.4%) \| <0.001 \| 1,337 (44.5%) \| 1,018 (77.1%) \| <0.001 \| \| High triglyceride \| 647 (17.5%) \| 798 (44.5%) \| <0.001 \| 529 (17.1%) \| 723 (45.0%) \| <0.001 \| Among 14 metals in the present study, we excluded MMA from the association analyses because the detection rate was <$50\%$ (Supplementary Table 2). The distribution of urinary metals stratified by MAFLD and NAFLD was presented in Supplementary Table 4. The participants with MAFLD or NAFLD had significantly higher levels of heavy metals. The Spearman correlations among these heavy metals varied from weak (0.05) to strong (0.82), as presented in Supplementary Figure 2. The strongest correlations were detected between DMA and As ($r = 0.82$). ## 3.2. Associations of single metal exposure with MAFLD, NAFLD, and MAFLD components The results from the binary logistic regression models adjusted for the covariates are shown in Table 2. We found a significant positive association between As [OR 1.09 ($95\%$CI 1.03–1.16)], DMA [OR 1.17 ($95\%$CI 1.07–1.27)], Ba [OR 1.22 ($95\%$CI 1.14–1.30)], Co [OR 1.22 ($95\%$CI 1.11–1.34)], Cs [OR 1.35 ($95\%$CI 1.18–1.54)], Mo [OR 1.45 ($95\%$CI 1.30–1.62)], Sb [OR 1.18 ($95\%$CI 1.08–1.29)], Tl [OR 1.49 ($95\%$CI 1.33–1.67)], and W [OR 1.23 ($95\%$CI 1.15–1.32)] with MAFLD. For NAFLD, patterns of associations with heavy metals were similar to those of MAFLD, except for As, where no significant association between As and NAFLD was observed (Table 2). The significant linear trend of the associations with these two outcomes was also observed when heavy metals were modeled as quartiles (P trend < 0.05, Supplementary Figures 3, 4). Additionally, Cd was significantly associated with MAFLD compared in Q2 [OR 1.30 ($95\%$CI 1.08–1.56)], Q3 [OR 1.43 ($95\%$CI 1.18–1.74)], and Q4 [OR 1.25 ($95\%$CI 1.01–1.54)] to Q1, respectively, suggesting a non-linear association. Sex-stratified analyses revealed that the associations of these heavy metals with MAFLD and NAFLD were generally similar (P-int > 0.05, Supplementary Table 5) except for Hg, with stronger associations observed among females (P-int < 0.001). Positive associations generally remained significant when no adjustment for urinary dilution was made (Supplementary Table 6, model 1) or the NHANES complex survey design, the sample weight, was incorporated (Supplementary Table 6, model 2). Table 2 \| Metal \| MAFLD, OR (95%CI) \| NAFLD, OR (95%CI) \| \| --- \| --- \| --- \| \| Asb \| 1.03 (0.99, 1.07) \| 1.02 (0.98, 1.06) \| \| As \| 1.09 (1.03, 1.16) \| 1.07 (1.00, 1.14) \| \| Ba \| 1.22 (1.14, 1.30) \| 1.22 (1.14, 1.31) \| \| Cd \| 1.08 (0.98, 1.19) \| 1.09 (0.98, 1.21) \| \| Co \| 1.22 (1.11, 1.34) \| 1.22 (1.10, 1.35) \| \| Cs \| 1.35 (1.18, 1.54) \| 1.32 (1.14, 1.52) \| \| DMA \| 1.17 (1.07, 1.27) \| 1.12 (1.02, 1.23) \| \| Hg \| 0.95 (0.89, 1.01) \| 0.92 (0.86, 0.99) \| \| Mo \| 1.45 (1.30, 1.62) \| 1.51 (1.34, 1.69) \| \| Pb \| 1.00 (0.91, 1.10) \| 0.98 (0.89, 1.09) \| \| Sb \| 1.18 (1.08, 1.29) \| 1.19 (1.08, 1.31) \| \| Tl \| 1.49 (1.33, 1.67) \| 1.47 (1.30, 1.67) \| \| W \| 1.23 (1.15, 1.32) \| 1.23 (1.14, 1.33) \| The associations of heavy metals with MAFLD components are presented in Supplementary Figure 5. The positive associations of Co, Cs, Mo, Tl, and W with MAFLD were also observed for diabetes mellitus, overweight/obesity, and metabolic dysfunction, specifically central obesity, insulin resistance, and prediabetes. The positive associations of MAFLD with As, DMA, and Sb were also observed for diabetes mellitus, whereas positive associations with Ba were observed for overweight/obesity. ## 3.3. Associations of metal mixture exposure and MAFLD, NAFLD, and MALFD components The results of the QGC models showed that a quartile increase in the metal mixture was significantly associated with increased odds of being MAFLD [OR 1.58 ($95\%$CI 1.40–1.78)]. As shown in Figure 1, Tl (0.21) contributed most to the positive association between heavy metals and MAFLD, followed by Mo (0.16), W (0.16), and Ba (0.16). Pb had the largest negative contribution to the overall effect, followed by Hg. We also observed a quartile increase in the QGC index was significantly associated with NAFLD [OR 1.52 ($95\%$CI 1.33–1.74)], diabetes [OR 1.38 ($95\%$CI 1.19–1.59)], overweight/obesity [OR 1.43 ($95\%$CI 1.26–1.62)], and metabolic dysfunction [OR 1.35 ($95\%$CI 1.20–1.51)]. Urinary Tl exposure was assigned the largest positive weights with NAFLD and obesity/overweight. Urinary W and Ba exposure were assigned to the strongest positive weights in the relationship with diabetes and metabolic dysfunction, respectively (Figure 1; Supplementary Figure 6). Figure 1: Combined association (95%CI) and qgcomp weights of metal mixture with MAFLD, NAFLD, and components of MAFLD by QGC models. Models were adjusted for age, gender, race, research cycle, education level, smoking status, poverty income ratio, and physical activity. Qgcomp, quantile g-computation; MAFLD, metabolic associated fatty liver disease; NAFLD, non-alcoholic fatty liver disease; MD, metabolic dysfunction. In our study, the BKMR model was developed to estimate the combined effects of 13 urinary metal mixtures on MAFLD, NAFLD, and the component of MAFLD. Figure 2 presents the cumulative effect of the metal mixtures by comparing when all metals were at their 50th percentile and $95\%$ confidence interval. The overall positive effects of metal mixtures on MAFLD and NAFLD were observed. Figure 3 displays the dose-response relationship with other metals set at median concentrations after adjusting for the covariates. Positive exposure-response relationships were observed between Ba, Mo, Tl, and W with MAFLD, while negative associations were observed for Hg and Pb. Similar patterns were observed for NAFLD. Additionally, the PIP for each metal was estimated (Supplementary Table 7). Among metal mixtures, the chemicals with the highest PIPs (1.00) were Ba, Cd, Hg, Mo, Pb, Tl, and W in the MAFLD model and Ba, Cd, Hg, Mo, Tl, and W in the NAFLD model. Figure 2: Combined association (95%CI) of metal mixture with MAFLD, NAFLD, and components of MAFLD by BKMR models, comparing all chemicals set at different levels with their 50th percentiles. Models were adjusted for age, gender, race, research cycle, education level, smoking status, poverty income ratio, and physical activity. BKMR, Bayesian kernel machine regression; MAFLD, metabolic associated fatty liver disease; NAFLD, non-alcoholic fatty liver disease; MD, metabolic dysfunction. Figure 3: Univariate exposure-response function (95%CI) showed associations of heavy metals with MAFLD, NAFLD, and MAFLD components. All the remaining metal exposures are fixed at their median values. Results were adjusted for age, gender, race, research cycle, education level, smoking status, PIR, and physical activity. BKMR, Bayesian kernel machine regression; MAFLD, metabolic associated fatty liver disease; NAFLD, non-alcoholic fatty liver disease; MD, metabolic dysfunction; PIR, poverty income ratio. Regarding the components of MAFLD, we observed a positive correlation of the overall metals mixture with diabetes, overweight/obesity, metabolic dysfunction, central obesity, prediabetes, insulin resistance, high C-reactive protein (CRP), and high triglyceride (TG) (Figure 2; Supplementary Figure 7). For single metal response, the exposure-response of Ba and Tl in MAFLD was similar to that of obesity/overweight and metabolic dysfunction (Figure 3; Supplementary Figure 8). ## 4. Discussion Using multivariate logistic regression, BKMR, and QGC analysis of the metal mixture, we found that mixture of 13 analyzed metals was significantly associated with MAFLD, NAFLD, and the component of MAFLD. Tl, Mo, W, and Ba were positively associated with MAFLD based on logistic regression and BKMR. Tl, Mo, W, and Ba were also positively weighted in QGC. The positive weight of Tl and Ba on MAFLD may attribute to obesity/overweight and metabolic dysfunction, while that of Mo and W may mostly attribute to diabetes mellitus. Interestingly, urinary Hg and Pb were inversely associated with MAFLD or NAFLD risk in BKMR and QGC models. Studies on the association between human exposure to individual and joint metals and MAFLD are sporadic. In both QGC and BKMR models, Tl, Mo, W, and Ba mainly contributed to the positive associations of heavy metal exposure with MAFLD risk. Findings in the single metal analysis of these metals also presented similarly positive associations. In line with our study, Asprouli et al. [ 45] reported a positive association of Tl with NAFLD risk or liver function indices among Greece's population. High Tl toxicity is mainly due to increased reactive oxygen species (ROS) and the interference of K-dependent reactions to secret insulin [46]. Concerning Mo, decreased serum Mo was associated with a higher risk of NAFLD in Chinese males [47]. The discordance in results may attribute to disparities in the study population or varied metrics used for the exposure assessment of metals. Excessive amounts of essential Mo may also cause toxicity by inducing the generation of reactive oxygen species (ROS) [48]. For Ba and W, despite rare evidence from the population study, the hepatotoxicity of Ba and W was partially revealed by previous in vivo research. In the hepatocyte of rats, the high dose of Ba might increase the biomarkers with the implication of oxidative stress and disturb the activities of membrane-bound ATPases [49]. The generation of ROS was similarly found in human liver cells exposed to an increased dose of W [48]. Thus, more epidemiological studies are warranted to confirm the link between these metals and MAFLD. In addition, to elucidate the MAFLD risk and heavy metal exposure, we also investigated associations with MAFLD components and heavy metals. The increased risk of MAFLD might be related to the development of diabetes mellitus, obesity/overweight, and metabolic dysfunction. Consistent results were shown by previous research. In U.S. adults or adolescents, combined impacts of metals, including Ba and Tl, were associated with obesity and type 2 diabetes [17, 50]. Results from South Korea suggested the positive association between joint effects of Hg, Pb, and Cd with metabolic dysfunction, including hypertension, high TG, and central obesity [18]. These findings indicated that exposure to heavy metals affects metabolic function, which is a significant risk factor for MAFLD and NAFLD. Although the underlying mechanisms are not fully understood, both in vivo and in vitro studies provided valuable hints. Most metals, such as Cd, Ba, Tl, W, and As could induce ROS production, which in turn induces the release of apoptosis cytokine, activation of hepatic stellate cells, and finally, formation of fibrosis [51, 52]. Meanwhile, the increased oxidative stress may also be generated by impaired homeostasis of essential trace elements, which act as important cofactors in many enzymes mediating such progress [53, 54]. In rat liver mitochondria, Co could induce oxidative stress, in the presence of calcium, by highly damaging hydroxyl radical, finally resulting in apoptosis [55]. Notably, such adverse effects are additive. For example, concurrent As and Cd exposure in rats is more damaging than separate exposure in triggering oxidant stress [56]. Oxidant stress could also attribute to lipotoxic species, which are produced due to the overwhelmed disposal of fatty acids through beta-oxidation [57]. In addition to ROS production, an essential factor advancing the development of fatty liver, heavy metals could directly disturb fatty acids' metabolism and increase fat accumulation in the liver [58]. For example, Cd in vivo inhibits the fatty acid oxidation in the mitochondrial of hepatocytes, potentially through the sirtuin 1 signaling pathway [59]. As could inhibit beta-oxidation of fatty acid by reacting with protein sulfhydryl groups and inactivating enzymes [10]. The inactivation enzymes, less energy production, and more lipids production may also accelerate the development of MAFLD components. For example, Tl and Ba were associated with increased obesity, with similar mechanisms [17]. By disrupting lipid metabolism, Cd could impair pancreatic β-cell function and exaggerate diabetes [60]. Previous evidence showed that lipid accumulation could lead to hepatic insulin resistance and hepatic inflammation [61]. Thus, the lipid metabolism disturbance could be a link between the comprehensive influence of metal exposure to fatty liver disease, obesity, diabetes, and other metabolic syndromes. Our result indicated that urinary Pb and Hg were negatively associated with MAFLD risk, which was contrary to previous findings using blood biomarkers [62]. We propose several potential explanations for this contradictory result. First, no significant association was observed between Hg and Pb with MAFLD in individual metal analysis. Thus, this inverse association might be attributed to complex antagonism between metals. The antagonistic interactions are common among metal mixtures due to the competition for carriers, metabolic interference, and morphological factors [63, 64]. The adverse effects of these two metals on fatty liver disease might be alleviated in the metal mixture. Second, the various metrics (i.e., blood and urine) assessing the exposure level of Hg and Pb may be another possible reason. In U.S. populations, it is reported that the urinary Hg declined over the period 1999–2016, whereas there was a steady increase in blood organic Hg [65]. Another research also suggested the difference between urinary and blood lead in analysis [66]. However, the above hypothetical explanations need to be further validated by future studies. Additionally, the inverse association between Hg with MAFLD and NAFLD was merely shown in men. Previous studies also reported the sex differences in health hazards of heavy metals. Heavy metals tend to correlate positively with NAFLD or liver fibrosis, more so in women than men [21, 22]. Some hypotheses may explain this difference. For example, a lack of iron may contribute to the compensatory increase of heavy metal absorption by women [67]. Rat models showed the expression of organic anion transporter (Oat) might also relate to the sex-specific organ toxicity of Hg exposure [68]. In the male rat, the declined expression of Oat3 in the hepatocytes membranes after exposure would lower the intake of Hg, leading to a higher accumulation of Hg in the female liver. However, no sex disparity was found in other metals in our study, indicating future validation studies in a sex-specific manner. We found significantly positive associations of the overall metal mixtures, including all 13 metals (As, Asb, DMA, Ba, Cd, Co, Cs, Hg, Pb, Mo, Sb, Tl, and W), with the risk of MAFLD or NAFLD, using BKMR and QGC models. Previous research on the combined associations of metal exposure on fatty liver disease is somewhat limited. Moon et al. utilized QGC to estimate the overall effects of Hg, Pb, and Cd on the hepatic steatosis index (HSI) and NAFLD risk [62]. In this result among the Korean population, Elevated levels of Pb and Hg in total blood were associated with high HSI and increased risk of NAFLD. Similar results for the mixture of Hg, Pb, and Cd were reported by another study using the weighted quantile sum (WQS) regression model, QGC, and BKMR [69]. In a cross-sectional study of Chinese males, the least absolute shrinkage and selection operator (LASSO) regression was used to explore the associations of 22 serum metals with NAFLD, which reported a negative association for Mo, and a positive association with Zn [47]. Principle components analysis combined with Pearson correlation coefficients was also used to investigate multiple metal exposures, which suggested that the co-exposure to As-Hg, Pb-Cd, and Se-Zn pair patterns were linked to metabolic syndrome [70]. Another two studies also used NHANES datasets for analysis. Li et al. used WQS and participants from NHANES 1999–2014 for analysis [23]. Their results also showed a positive association between the metal mixture and NAFLD. Contrary to BKMR and QGC, WQS makes a unidirectionality assumption that all chemicals are positively or negatively associated with the given exposure [38] and this study merely incorporated the positive circumstance. Simultaneously, their study failed to take account of creatinine, an important marker affecting urinary concentrations of environmental contaminants [32]. In another study using NHANES 2017–2020 [22], controlled attenuation parameter (CAP) and liver stiffness measurement (LSM) were utilized for the indicators of NAFLD and liver fibrosis. We used USFLI to diagnose NAFLD and MAFLD to include more participants because these two indices were merely available in two research cycles. No above studies applied the MAFLD or MAFLD components as the health outcomes. BKMR and QGC models were used for multiple-metal exposure in this study. The BKMR model can resolve non-linear, multiple, and complex interactions between mixed exposures to metals or other chemicals [36]. However, the BKMR model is not based on parametric inference. Similar to the BKMR model, QGC can address non-linear and non-additive effects but provide parametric inference results. Our research involved more diverse metals in the analysis and the final response function, compared with prior studies [47, 62, 69, 70]. There are other established methods for analyzing mixture exposure in previous studies, including WQS, latent class analysis (LCA), and Lasso. Nevertheless, we were mainly interested in the overall association and interactions of all metals and chose BKMR and QGC for analysis. There are several strengths of this study. First, to the best of our knowledge, this is the first study to illustrate the association between exposure to various heavy metals and MAFLD, the novel nomenclature form of fatty liver disease. Second, we explored the associations between heavy metals with MAFLD and NAFLD directly and with MAFLD components, serving as indirect evidence. Third, we used both BKMR and QGC models, which have complementary advantages. However, this study still has some notable limitations. First, this is a cross-sectional study, which does not allow the determination of temporality. Future prospective studies are required to investigate the causal relationships between joint metal exposure and MAFLD and NAFLD. Second, a definitive diagnosis of MAFLD and NAFLD by liver biopsy was unavailable from the NHANES database. Thus, we used USFLI to identify liver steatosis cases. However, compared with another generally used index, the hepatic steatosis index (HSI), USFLI was suggested to be more precise in the assessment of steatosis for the NHANES database [30]. ## 5. Conclusion In conclusion, this study suggested that exposure to metal mixtures is associated with the risk of MAFLD, NAFLD, and MAFLD components in US adults. Tl, Mo, W, and Ba contributed most to the MALFD risk, which could be instructive in MALFD prevention. Our findings apply the new definition of fatty liver disease and test its associations with risk factors in the real world. However, given the cross-sectional design of the present study, these findings are warranted to be confirmed by future longitudinal studies. ## Data availability statement Publicly available datasets were analyzed in this study. This data can be found here: https://wwwn.cdc.gov/nchs/nhanes/Default.aspx. ## Ethics statement The studies involving human participants were reviewed and approved by National Center for Health Statistics (NCHS) Institutional Review Board. The patients/participants provided their written informed consent to participate in this study. ## Author contributions ZX and RA: methodology, data curation, formal analysis, visualization, writing—original draft, and writing—review and editing. MS: data curation. YQ: writing—review and editing and supervision. YJ: conceptualization, supervision, funding acquisition, and writing—review and editing. All authors 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. ## Supplementary material The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fpubh.2023.1133194/full#supplementary-material ## References 1. Riazi K, Azhari H, Charette JH, Underwood FE, King JA, Afshar EE. The prevalence and incidence of NAFLD worldwide: a systematic review and meta-analysis. Lancet Gastroenterol Hepatol. (2022.0) 7 851-61. DOI: 10.1016/S2468-1253(22)00165-0 2. Rinella ME. Nonalcoholic fatty liver disease: a systematic review. JAMA. (2015.0) 313 2263-73. DOI: 10.1001/jama.2015.5370 3. Eslam M, Sanyal AJ, George J, International Consensus Panel. MAFLD: a consensus-driven proposed nomenclature for metabolic associated fatty liver disease. Gastroenterology (2020.0) 158 1999-2014.e1. DOI: 10.1053/j.gastro.2019.11.312 4. Eslam M, Newsome PN, Sarin SK, Anstee QM, Targher G, Romero-Gomez M. A new definition for metabolic dysfunction-associated fatty liver disease: an international expert consensus statement. J Hepatol. (2020.0). DOI: 10.1016/j.jhep.2020.07.045 5. Lim GEH, Tang A, Ng CH, Chin YH, Lim WH, Tan DJH. An observational data meta-analysis on the differences in prevalence and risk factors between MAFLD vs. NAFLD. Clin Gastroenterol Hepatol (2021.0). DOI: 10.1016/j.cgh.2021.11.038 6. Kim D, Konyn P, Sandhu KK, Dennis BB, Cheung AC, Ahmed A. Metabolic dysfunction-associated fatty liver disease is associated with increased all-cause mortality in the United States. J Hepatol. (2021.0) 75 1284-91. DOI: 10.1016/j.jhep.2021.07.035 7. Wong RJ, Cheung R. Trends in the prevalence of metabolic dysfunction-associated fatty liver disease in the United States, 2011-2018. Clin Gastroenterol Hepatol. (2022.0) 20 e610-3. DOI: 10.1016/j.cgh.2021.01.030 8. Treviño LS, Katz TA. Endocrine disruptors and developmental origins of nonalcoholic fatty liver disease. Endocrinology. (2018.0) 159 20-31. DOI: 10.1210/en.2017-00887 9. Wan H, Wang Y, Zhang H, Zhang K, Chen Y, Chen C. Chronic lead exposure induces fatty liver disease associated with the variations of gut microbiota. Ecotoxicol Environ Saf. (2022.0) 232 113257. DOI: 10.1016/j.ecoenv.2022.113257 10. Tchounwou PB, Yedjou CG, Patlolla AK, Sutton DJ. Heavy metal toxicity and the environment. Exp Suppl. (2012.0) 101 133-64. DOI: 10.1007/978-3-7643-8340-4_6 11. He ZL, Yang XE, Stoffella PJ. Trace elements in agroecosystems and impacts on the environment. J Trace Elem Med Biol. (2005.0) 19 125-40. DOI: 10.1016/j.jtemb.2005.02.010 12. Kim J, Lee K. Changes in urinary cotinine concentrations in non-smoking adults from the Korean National Environmental Health Survey (KoNEHS) II (2012-2014) after implementation of partial smoke-free regulations. Int J Hyg Environ Health. (2020.0) 224 113419. DOI: 10.1016/j.ijheh.2019.113419 13. 13.World Health Organization (WHO). Health Risks of Heavy Metals from Long-Range Transboundary Air Pollution. WHO Regional Office for Europe (2007). Available online at: https://www.euro.who.int/__data/assets/pdf_file/0007/78649/E91044.pdf?language=?. Health Risks of Heavy Metals from Long-Range Transboundary Air Pollution (2007.0) 14. Lebeaupin C, Vallée D, Hazari Y, Hetz C, Chevet E, Bailly-Maitre B. Endoplasmic reticulum stress signalling and the pathogenesis of non-alcoholic fatty liver disease. J Hepatol. (2018.0) 69 927-47. DOI: 10.1016/j.jhep.2018.06.008 15. Toomey CB, Cauvi DM, Hamel JC, Ramirez AE, Pollard KM. Cathepsin B regulates the appearance and severity of mercury-induced inflammation and autoimmunity. Toxicol Sci. (2014.0) 142 339-49. DOI: 10.1093/toxsci/kfu189 16. Tinkov AA, Ajsuvakova OP, Skalnaya MG, Popova EV, Sinitskii AI, Nemereshina ON. Mercury and metabolic syndrome: a review of experimental and clinical observations. Biometals. (2015.0) 28 231-54. DOI: 10.1007/s10534-015-9823-2 17. Wang X, Mukherjee B, Park SK. Associations of cumulative exposure to heavy metal mixtures with obesity and its comorbidities among U.S. adults in NHANES 2003-2014. Environ Int (2018.0) 121 683-94. DOI: 10.1016/j.envint.2018.09.035 18. Lee BK, Kim Y. Blood cadmium, mercury, and lead and metabolic syndrome in South Korea: 2005-2010 Korean National Health and Nutrition Examination Survey. Am J Ind Med. (2013.0) 56 682-92. DOI: 10.1002/ajim.22107 19. Stratakis N, Golden-Mason L, Margetaki K, Zhao Y, Valvi D, Garcia E. Hepatology. (2021.0) 74 1546-59. DOI: 10.1002/hep.31809 20. Frediani JK, Naioti EA, Vos MB, Figueroa J, Marsit CJ, Welsh JA. Arsenic exposure and risk of nonalcoholic fatty liver disease (NAFLD) among U.S. adolescents and adults: an association modified by race/ethnicity, NHANES 2005-2014. Environ Health (2018.0) 17 6. DOI: 10.1186/s12940-017-0350-1 21. Chung SM, Moon JS, Yoon JS, Won KC, Lee HW. The sex-specific effects of blood lead, mercury, and cadmium levels on hepatic steatosis and fibrosis: Korean nationwide cross-sectional study. J Trace Elem Med Biol (2020.0) 62 126601. DOI: 10.1016/j.jtemb.2020.126601 22. Wan H, Jiang Y, Yang J, Ma Q, Liu L, Peng L. Sex-specific associations of the urinary fourteen-metal mixture with NAFLD and liver fibrosis among US adults: A nationally representative study. Ecotoxicol Environ Saf. (2022.0) 248 114306. DOI: 10.1016/j.ecoenv.2022.114306 23. Li W, Xiao H, Wu H, Pan C, Deng K, Xu X. Analysis of environmental chemical mixtures and nonalcoholic fatty liver disease: NHANES 1999-2014. Environ Pollut (2022.0) 311 119915. DOI: 10.1016/j.envpol.2022.119915 24. 24.Centers for Disease Control Prevention (CDC). About the National Health and Nutrition Examination Survey. (2022). Available online at: https://www.cdc.gov/nchs/nhanes/about_nhanes.htm (accessed September 22, 2022).. About the National Health and Nutrition Examination Survey. (2022.0) 25. Gupta PK. Chapter 18 - Toxic effects of metals in Fundamentals of Toxicology (2016.0) 203-13 26. Hepp NM, Mindak WR, Gasper JW, Thompson CB, Barrows JN. Survey of cosmetics for arsenic, cadmium, chromium, cobalt, lead, mercury, and nickel content. J Cosmet Sci. (2014.0) 65 125-45. PMID: 25043485 27. 27.Centers for Disease Control Prevention (CDC). NHANES Laboratory Procedure Manual (LPM). (2017). Available online at: https://wwwn.cdc.gov/nchs/data/nhanes/2017-2018/manuals/2017_MEC_Laboratory_Procedures_Manual.pdf (accessed September 20, 2022).. NHANES Laboratory Procedure Manual (LPM). (2017.0) 28. EASL-EASD-EASO Clinical Practice Guidelines for the management of non-alcoholic fatty liver disease. J Hepatol (2016.0) 64 1388-402. DOI: 10.1016/j.jhep.2015.11.004 29. Ruhl CE, Everhart JE. Fatty liver indices in the multiethnic United States National Health and Nutrition Examination Survey. Aliment Pharmacol Ther. (2015.0) 41 65-76. DOI: 10.1111/apt.13012 30. Barb D, Repetto EM, Stokes ME, Shankar SS, Cusi K. Type 2 diabetes mellitus increases the risk of hepatic fibrosis in individuals with obesity and nonalcoholic fatty liver disease. Obesity. (2021.0) 29 1950-60. DOI: 10.1002/oby.23263 31. Chalasani N, Younossi Z, Lavine JE, Charlton M, Cusi K, Rinella M. The diagnosis and management of nonalcoholic fatty liver disease: practice guidance from the American Association for the Study of Liver Diseases. Hepatology. (2018.0) 67 328-57. DOI: 10.1002/hep.29367 32. O'Brien KM, Upson K, Cook NR, Weinberg CR. Environmental chemicals in urine and blood: improving methods for creatinine and lipid adjustment. Environ Health Perspect. (2016.0) 124 220-7. DOI: 10.1289/ehp.1509693 33. Kuiper JR, Vuong AM, Lanphear BP, Calafat AM, Ospina M, Cecil KM. Early life organophosphate ester exposures and bone health at age 12 years: the health outcomes and measures of the environment (HOME) Study. Sci Total Environ (2022.0) 851 158246. DOI: 10.1016/j.scitotenv.2022.158246 34. Barr DB, Wilder LC, Caudill SP, Gonzalez AJ, Needham LL, Pirkle JL. Urinary creatinine concentrations in the US population: implications for urinary biologic monitoring measurements. Environ Health Perspect. (2005.0) 113 192-200. DOI: 10.1289/ehp.7337 35. Pan JJ, Fallon MB. Gender and racial differences in nonalcoholic fatty liver disease. World J Hepatol. (2014.0) 6 274-83. DOI: 10.4254/wjh.v6.i5.274 36. Bobb JF, Valeri L, Claus Henn B, Christiani DC, Wright RO, Mazumdar M. Bayesian kernel machine regression for estimating the health effects of multi-pollutant mixtures. Biostatistics. (2015.0) 16 493-508. DOI: 10.1093/biostatistics/kxu058 37. Coker E, Chevrier J, Rauch S, Bradman A, Obida M, Crause M. Association between prenatal exposure to multiple insecticides and child body weight and body composition in the VHEMBE South African birth cohort. Environ Int. (2018.0) 113 122-32. DOI: 10.1016/j.envint.2018.01.016 38. Keil AP, Buckley JP, O'Brien KM, Ferguson KK, Zhao S, White AJ. Quantile-based g-computation approach to addressing the effects of exposure mixtures. Environ Health Perspect. (2020.0) 128 47004. DOI: 10.1289/EHP5838 39. Wahlang B, Beier JI, Clair HB, Bellis-Jones HJ, Falkner KC, McClain CJ. Toxicant-associated steatohepatitis. Toxicol Pathol. (2013.0) 41 343-60. DOI: 10.1177/0192623312468517 40. Yang HH, Chen GC, Li DM, Lan L, Chen LH, Xu JY. Serum iron and risk of nonalcoholic fatty liver disease and advanced hepatic fibrosis in US adults. Sci Rep. (2021.0) 11 10387. DOI: 10.1038/s41598-021-89991-x 41. Kim S, Kim S, Won S, Choi K. Considering common sources of exposure in association studies - Urinary benzophenone-3 and DEHP metabolites are associated with altered thyroid hormone balance in the NHANES 2007-2008. Environ Int (2017.0) 107 25-32. DOI: 10.1016/j.envint.2017.06.013 42. Zhang Y, Dong T, Hu W, Wang X, Xu B, Lin Z. Association between exposure to a mixture of phenols, pesticides, and phthalates and obesity: comparison of three statistical models. Environ Int. (2019.0) 123 325-36. DOI: 10.1016/j.envint.2018.11.076 43. 43.Centers for Disease Control Prevention (CDC). National Health and Nutrition Examination Survey. Sample Design. Available online at: https://wwwn.cdc.gov/nchs/nhanes/tutorials/sampledesign.aspx (accessed January 29, 2023).. National Health and Nutrition Examination Survey 44. 44.R Core Team. R: A Language and Environment for Statistical Computing. Vienna: R Foundation for Statistical Computing (2019).. R: A Language and Environment for Statistical Computing (2019.0) 45. Asprouli E, Kalafati IP, Sakellari A, Karavoltsos S, Vlachogiannakos J, Revenas K. Evaluation of plasma trace elements in different stages of nonalcoholic fatty liver disease. Biol Trace Elem Res. (2019.0) 188 326-33. DOI: 10.1007/s12011-018-1432-9 46. Kiliç GA, Kutlu M. Effects of exogenous metallothionein against thallium-induced oxidative stress in rat liver. Food Chem Toxicol. (2010.0) 48 980-7. DOI: 10.1016/j.fct.2010.01.013 47. Li L, Huang L, Huang S, Luo X, Zhang H, Mo Z. Non-linear association of serum molybdenum and linear association of serum zinc with nonalcoholic fatty liver disease: Multiple-exposure and Mendelian randomization approach. Sci Total Environ. (2020.0) 720 137655. DOI: 10.1016/j.scitotenv.2020.137655 48. Sachdeva S, Maret W. Comparative outcomes of exposing human liver and kidney cell lines to tungstate and molybdate. Toxicol Mech Methods. (2021.0) 31 690-8. DOI: 10.1080/15376516.2021.1956031 49. Elwej A, Chaabane M, Ghorbel I, Chelly S, Boudawara T, Zeghal N. Effects of barium graded doses on redox status, membrane bound ATPases and histomorphological aspect of the liver in adult rats. Toxicol Mech Methods. (2017.0) 27 677-86. DOI: 10.1080/15376516.2017.1351016 50. Wu B, Jiang Y, Jin X, He L. Using three statistical methods to analyze the association between exposure to 9 compounds and obesity in children and adolescents: NHANES 2005-2010. Environ Health. (2020.0) 19 94. DOI: 10.1186/s12940-020-00642-6 51. Yuan G, Dai S, Yin Z, Lu H, Jia R, Xu J. Toxicological assessment of combined lead and cadmium: acute and sub-chronic toxicity study in rats. Food Chem Toxicol. (2014.0) 65 260-8. DOI: 10.1016/j.fct.2013.12.041 52. Zhong G, Wan F, Wu S, Jiang X, Tang Z, Zhang X. Arsenic or/and antimony induced mitophagy and apoptosis associated with metabolic abnormalities and oxidative stress in the liver of mice. Sci Total Environ. (2021.0) 777 146082. DOI: 10.1016/j.scitotenv.2021.146082 53. Zoroddu MA, Aaseth J, Crisponi G, Medici S, Peana M, Nurchi VM. The essential metals for humans: a brief overview. J Inorg Biochem. (2019.0) 195 120-9. DOI: 10.1016/j.jinorgbio.2019.03.013 54. Sule K, Umbsaar J. Prenner EJ. Mechanisms of Co, Ni, and Mn toxicity: From exposure and homeostasis to their interactions with and impact on lipids and biomembranes. Biochim Biophys Acta Biomembr. (2020.0) 1862 183250. DOI: 10.1016/j.bbamem.2020.183250 55. Battaglia V, Compagnone A, Bandino A, Bragadin M, Rossi CA, Zanetti F. Cobalt induces oxidative stress in isolated liver mitochondria responsible for permeability transition and intrinsic apoptosis in hepatocyte primary cultures. Int J Biochem Cell Biol. (2009.0) 41 586-94. DOI: 10.1016/j.biocel.2008.07.012 56. Wu X, Cobbina SJ, Mao G, Xu H, Zhang Z, Yang L. review of toxicity and mechanisms of individual and mixtures of heavy metals in the environment. Environ Sci Pollut Res Int. (2016.0) 23 8244-59. DOI: 10.1007/s11356-016-6333-x 57. Friedman SL, Neuschwander-Tetri BA, Rinella M, Sanyal AJ. Mechanisms of NAFLD development and therapeutic strategies. Nat Med. (2018.0) 24 908-22. DOI: 10.1038/s41591-018-0104-9 58. Abdel Azim A, Rittmann SKR, Fino D, Bochmann G. The physiological effect of heavy metals and volatile fatty acids on Methanococcus maripaludis S2. Biotechnol Biofuels. (2018.0) 11 301. DOI: 10.1186/s13068-018-1302-x 59. He X, Gao J, Hou H, Qi Z, Chen H, Zhang XX. Inhibition of mitochondrial fatty acid oxidation contributes to development of nonalcoholic fatty liver disease induced by environmental cadmium exposure. Environ Sci Technol. (2019.0) 53 13992-4000. DOI: 10.1021/acs.est.9b05131 60. Hong H, Xu Y, Xu J, Zhang J, Xi Y, Pi H. Cadmium exposure impairs pancreatic β-cell function and exaggerates diabetes by disrupting lipid metabolism. Environ Int. (2021.0) 149 106406. DOI: 10.1016/j.envint.2021.106406 61. Sung KC, Jeong WS, Wild SH, Byrne CD. Combined influence of insulin resistance, overweight/obesity, and fatty liver as risk factors for type 2 diabetes. Diabetes Care. (2012.0) 35 717-22. DOI: 10.2337/dc11-1853 62. Moon MK, Lee I, Lee A, Park H, Kim MJ, Kim S. Lead, mercury, and cadmium exposures are associated with obesity but not with diabetes mellitus: Korean National Environmental Health Survey (KoNEHS) 2015-2017. Environ Res (2022.0) 204 111888. DOI: 10.1016/j.envres.2021.111888 63. Magos L, Webb M. Theoretical and practical considerations on the problem of metal–metal interaction. Environ Health Perspect (1978.0) 25 151-4. DOI: 10.1289/ehp.7825151 64. Feng J, Gao Y, Ji Y, Zhu L. Quantifying the interactions among metal mixtures in toxicodynamic process with generalized linear model. J Hazard Mater. (2018.0) 345 97-106. DOI: 10.1016/j.jhazmat.2017.11.013 65. So SCA, Tsoi MF, Cheung AJ, Cheung TT, Cheung BMY. Blood and urine inorganic and organic mercury levels in the United States from 1999 to 2016. Am J Med. (2021.0) 134 e20-30. DOI: 10.1016/j.amjmed.2020.06.023 66. Shimbo S, Zhang ZW, Moon CS, Watanabe T, Nakatsuka H, Matsuda-Inoguchi N. Correlation between urine and blood concentrations, and dietary intake of cadmium and lead among women in the general population of Japan. Int Arch Occup Environ Health. (2000.0) 73 163-70. DOI: 10.1007/s004200050023 67. Lee BK. Kim Y. Sex-specific profiles of blood metal levels associated with metal-iron interactions. Saf Health Work. (2014.0) 5 113-7. DOI: 10.1016/j.shaw.2014.06.005 68. Hazelhoff MH, Torres AM. Gender differences in mercury-induced hepatotoxicity: potential mechanisms. Chemosphere. (2018.0) 202 330-8. DOI: 10.1016/j.chemosphere.2018.03.106 69. Nguyen HD, Kim MS. Cadmium, lead, and mercury mixtures interact with non-alcoholic fatty liver diseases. Environ Pollut. (2022.0) 309 119780. DOI: 10.1016/j.envpol.2022.119780 70. Bulka CM, Persky VW, Daviglus ML, Durazo-Arvizu RA, Argos M. Multiple metal exposures and metabolic syndrome: a cross-sectional analysis of the National Health and Nutrition Examination Survey 2011-2014. Environ Res. (2019.0) 168 397-405. DOI: 10.1016/j.envres.2018.10.022
--- title: 'Serum uric acid: A risk factor for right ventricular dysfunction and prognosis in heart failure with preserved ejection fraction' authors: - Xiang-liang Deng - Han-wen Yi - Jin Xiao - Xiao-fang Zhang - Jin Zhao - Min Sun - Xue-song Wen - Zhi-qiang Liu - Lei Gao - Zi-yang Li - Ping Ge - Qi Yu - Dong-ying Zhang journal: Frontiers in Endocrinology year: 2023 pmcid: PMC10025558 doi: 10.3389/fendo.2023.1143458 license: CC BY 4.0 --- # Serum uric acid: A risk factor for right ventricular dysfunction and prognosis in heart failure with preserved ejection fraction ## Abstract ### Background Hyperuricemia and right ventricular dysfunction (RVD) are both widespread in heart failure with preserved ejection fraction (HFpEF) patients. RVD is associated with a poor prognosis in HFpEF. The correlation between serum uric acid (UA) levels and right ventricular function is unclear. The prognostic performance of UA in patients with HFpEF needs further validation. ### Methods and results A total of 210 patients with HFpEF were included in the study and divided into two groups according to UA level: the normal UA group (≤7 mg/dl) and the high UA group (>7 mg/dl). The variables examined included clinical characteristics, echocardiography, and serum biochemical parameters. Right ventricular function was assessed by tricuspid annular plane systolic excursion (TAPSE) and tricuspid annular peak systolic velocity (TAPSV). Baseline characteristics were compared between the two groups, and the correlation between baseline UA and RVD was assessed using multifactorial binary logistic regression. Kaplan–Meier curves were used to describe all-cause mortality and heart failure readmission. Results showed that right ventricular function parameters were worse in the high UA group. After adjusting for UA, left ventricular posterior wall thickness (LVPWT), N-terminal B-type natriuretic peptide (NT-proBNP), atrial fibrillation (AF), and low-density lipoprotein cholesterol (LDL-C), UA (odds ratio = 2.028; $p \leq 0.001$) was independently associated with RVD, and UA >7 mg/dl (HR = 2.98; $p \leq 0.001$) was associated with heart failure readmission in patients with HFpEF. ### Conclusion Elevated serum UA is closely associated with RVD and significantly associated with the heart failure readmission rate in patients with HFpEF. ## Introduction Heart failure with preserved ejection fraction (HFpEF) is generally considered a syndrome with pathophysiological heterogeneity, whose prevalence has increased rapidly over the past two decades [1, 2]. Factors affecting the prognosis (mortality and hospitalization) of HFpEF include metabolic syndrome, renal insufficiency, and so forth [3]. Hyperuricemia is an important comorbidity in heart failure patients and is usually associated with advanced severity of heart failure [4]. As the end product of purine metabolism in the human body, uric acid (UA) is commonly associated with the development and progression of cardiovascular diseases such as peripheral artery disease, coronary artery disease (CAD), hypertension, and atrial fibrillation (AF) (5–7). The prevalence of hyperuricemia ranges from $13.4\%$ to $20.1\%$ in different populations (8–10), and the total number of people with hyperuricemia in China has gradually increased to 170 million [11]. Elevated UA was particularly common in people with heart failure in China [12], and previous studies have shown that hyperuricemia may contribute to worse clinical outcomes in patients with cardiovascular diseases [13, 14]. Right ventricular dysfunction (RVD) is one of the common manifestations in the HFpEF population [15]. Former studies have demonstrated that RVD leads to a worse clinical prognosis compared to HFpEF patients without RVD. However, the current treatment of RVD has not been as effective as anticipated (16–18). This study aimed to investigate the relationship between UA and RVD in the context of HFpEF and to illustrate the relationship between UA and the prognosis of HFpEF. ## Study design and study population This is a prospective observational study to assess the association between baseline UA and RVD in patients with HFpEF and to investigate the relationship between elevated UA and patient prognosis. Study patients were enrolled between October 2020 and April 2022. All enrolled patients met the inclusion criteria for a definitive diagnosis of HFpEF according to the HFA-PEFF diagnostic algorithm. The exclusion criteria were [1] acute coronary syndrome or right myocardial infarction history, [2] severe renal impairment (eGFR < 30 ml/min/1.73 m2, based on CKD-EPI formula), [3] urate-lowering therapy, [4] malignant tumor, [5] severe hepatic impairment (elevated liver enzymes: three times over upper reference limit or liver cirrhosis), and [6] infections. According to Chinese guidelines for the diagnosis and management of hyperuricemia and gout in 2019, hyperuricemia is defined as above 7.0 mg/dl [19]. Patients enrolled in the study were divided into two groups: normal UA group (UA ≤7.0 mg/dl) and high UA group (UA >7.0 mg/dl). All the study population signed informed consents that were prospectively registered and agreed to be followed up for the collection of outcome data. Patients were followed up by phone every 4 months, and three patients were lost during the follow-up period. The last follow-up visit ended in August 2022. Ultimately, 210 patients were enrolled in the study (Figure 1). The study was in accordance with the Declaration of Helsinki, approved by the Clinical Research Review Board of the First Affiliated Hospital of Chongqing Medical University (No. 2021-473), and registered on clinicaltrials.gov with an identifier of NCT05053256. Figure 1: The enrollment flowchart. HFpEF, heart failure with preserved ejection fraction; HF, heart failure; LVEF, left ventricular ejection fraction; TAPSE, tricuspid annular plane systolic excursion; TAPSV, tricuspid annular peak systolic velocity; UA, uric acid. ## Data collection The baseline clinical data collection was conducted by trained researchers following the same protocol at the time of enrollment. Patients’ demographics, comorbidities, personal histories, medications, laboratory tests, and echocardiography were collected. Biochemical indexes were detected and analyzed, including albumin (Alb), blood urea nitrogen (BUR), creatinine (Cr), direct bilirubin (DB), hemoglobin (Hb), glycosylated hemoglobin (HbAlc), high-sensitivity C-reactive protein (hs-CRP), LDL-C, N-terminal B-type natriuretic peptide (NT-proBNP), and UA. ## Serum uric acid measurement Blood samples were taken on the second morning after admission. UA was performed through the central laboratory using ABBOTT. Conversion of each UA measurement from micromoles per liter to milligrams per deciliter was conducted by dividing it by 60. ## Echocardiography and assessment of right ventricular function All the echocardiographic examinations were conducted by trained echocardiographers according to the guidelines of the American Society of Echocardiography (ASE) [20]. The standard four-chamber method was used to measure the right atrial transverse diameter, right ventricular anteroposterior diameter, tricuspid annular plane systolic excursion (TAPSE), and tricuspid annular peak systolic velocity (TAPSV). We defined RVD as TAPSE <17 mm and TAPSV <9.5 cm/s. Pulmonary systolic pressure (PASP) was calculated as 4 * (peak tricuspid regurgitation velocity (TR))2 + right atrial pressure, estimated based on the diameter and collapse of the inferior vena cava. ## Outcomes and clinical follow-up Endpoints examined include readmission for heart failure and all-cause mortality. Heart failure readmission was determined by two senior doctors in the heart failure ward. Deaths were confirmed by population management consultations and hospital death certificates. Enrolled patients were followed up by telephone or WeChat every 4 months until the end of August 2022 or death. Postcharge clinical events were obtained through telephone follow-up and medical records from other hospitals. ## Statistical analysis We used percentages for qualitative data. Normally distributed quantitative data were presented as mean ± standard deviation (SD), and abnormally distributed quantitative data were presented as median (interquartile range (IQR)). The receiver operating characteristic (ROC) curve was used to determine the predictive value of UA for RVD. When comparing baseline data for HFpEF patients with UA > 7 mg/dl and UA ≤ 7 mg/dl, independent sample t-test, rank sum test, or Chi-square test were selected based on data characteristics. Pearson’s or Spearman’s tests were used to assess the association of variables with UA, TAPSE, and TAPSV. Based on published data and clinical relevance, we performed a univariate analysis of UA, gender, CAD, diabetes, AF, body mass index (BMI), systolic blood pressure (SBP), heart rate, left ventricular posterior wall thickness (LVPWT), and interventricular septal thickness (IVST). Based on the results of univariate binary logistic analysis, different models were developed to determine the odds ratio (OR) between UA and RVD. The long-term cumulative incidence of all-cause mortality and heart failure readmission was estimated using Kaplan–Meier curves. The predictive value of variables for heart failure readmission was tested by Cox’s univariate proportional hazards regression analysis. Variables in univariate Cox regression were included in multivariate Cox regression, and subgroup analysis was performed to demonstrate the potential effects of UA >7 mg/dl. Statistical significance was $p \leq 0.05.$ Results were expressed as a hazard ratio (HR) with $95\%$ confidence intervals ($95\%$ CI). All statistical analyses were performed using SPSS version 24.0 (SPSS, Chicago, IL, USA). ## Clinical characteristics In total, 210 eligible HFpEF patients were recruited to the study ($59\%$ women), with 80 patients assigned to the high UA group. Compared to the normal UA group, patients in the high UA group were more prevalent with CAD, AF, and higher New York Heart Association (NYHA) class heart failure, higher SBP, diastolic blood pressure (DBP), heart rate, BUR, Cr, NT-proBNP, DB, and hs-CRP, but lower Alb and LDL-C at baseline ($p \leq 0.05$; Table 1). Except for the higher frequency of diuretics use in the high UA group, there was no difference in medication use or other characteristics, including age, BMI, HbAlc, history of smoking, or alcohol consumption between the two groups (Table 1). Table 1 \| Unnamed: 0 \| Total (n = 210) \| Normal UA (≤7 mg/dl, n = 130) \| High UA (>7 mg/dl, n = 80) \| p-value \| \| --- \| --- \| --- \| --- \| --- \| \| Demographics \| Demographics \| Demographics \| Demographics \| Demographics \| \| Age (years) \| 74 (67, 81) \| 72 (67, 79) \| 77 (66, 82) \| 0.132 \| \| Gender/men (n, %) \| 86 (41.0%) \| 45 (34.6%) \| 41 (51.2%) \| 0.017 \| \| BMI (kg/m2) \| 23.6 (21.1, 26.4) \| 23.6 (21.5, 25.9) \| 23.6 (20.6, 26.7) \| 0.668 \| \| SBP (mmHg) \| 125 ± 15.6 \| 121 ± 12.5 \| 132 ± 17.5 \| <0.001 \| \| DBP (mmHg) \| 73 ± 9.7 \| 70 ± 7.7 \| 77 ± 11.1 \| <0.001 \| \| Heart rate \| 76 (66, 89) \| 74 (65, 85) \| 84 (71, 98) \| 0.001 \| \| Comorbidities \| Comorbidities \| Comorbidities \| Comorbidities \| Comorbidities \| \| CAD (n, %) \| 94 (44.8%) \| 45 (34.6%) \| 49 (61.3%) \| <0.001 \| \| Hypertension (n, %) \| 136 (64.8%) \| 83 (63.8%) \| 53 (66.3%) \| 0.723 \| \| Diabetes (n, %) \| 74 (35.2%) \| 49 (37.7%) \| 25 (31.3%) \| 0.343 \| \| AF (n, %) \| 84 (40.0%) \| 39 (30.0%) \| 45 (56.3%) \| <0.001 \| \| NYHA class \| NYHA class \| NYHA class \| NYHA class \| NYHA class \| \| II \| 106 (50.5%) \| 81 (62.3%) \| 25 (31.3%) \| <0.001 \| \| III \| 92 (43.8%) \| 46 (35.4%) \| 46 (57.5%) \| 0.002 \| \| IV \| 12 (5.7%) \| 3 (2.3%) \| 9 (11.3%) \| 0.007 \| \| Personal history \| Personal history \| Personal history \| Personal history \| Personal history \| \| Smoking (n, %) \| 54 (25.7%) \| 31 (23.8%) \| 23 (28.7%) \| 0.430 \| \| Drinking (n, %) \| 41 (19.5%) \| 20 (15.4%) \| 21 (26.3%) \| 0.054 \| \| Medications \| Medications \| Medications \| Medications \| Medications \| \| Diuretics (n, %) \| 141 (67.1%) \| 78 (60%) \| 63 (78.8%) \| 0.005 \| \| ACEI/ARB/ARNI (n, %) \| 132 (62.9%) \| 83 (63.8%) \| 49 (61.3%) \| 0.705 \| \| Statins (n, %) \| 158 (75.2%) \| 103(79.2%) \| 55(68.8%) \| 0.087 \| \| Laboratory values \| Laboratory values \| Laboratory values \| Laboratory values \| Laboratory values \| \| UA (mg/dl) \| 6.14 (5.05, 7.60) \| 5.24 (4.67, 5.92) \| 8.10 (7.48, 9.12) \| <0.001 \| \| BUR (mmol/L) \| 6.8 (5.6, 9.1) \| 6.4 (5.3, 7.9) \| 8 (6.1, 9.9) \| <0.001 \| \| Cr (µmol/L) \| 79 (63, 96) \| 69 (60, 85) \| 94 (79, 115) \| <0.001 \| \| NT-proBNP (pmol/L) \| 114.8 (52.8, 262.4) \| 83.3 (50.5, 164.7) \| 220.6 (109.3, 344.1) \| <0.001 \| \| hs-CRP (mg/L) \| 2.03(0.74, 6.94) \| 1.40 (0.67, 4.69) \| 2.91 (1.39, 9.79) \| 0.001 \| \| Hb (g/L) \| 130 (119, 143) \| 128 (120, 141) \| 131 (120, 146) \| 0.177 \| \| LDL-C (mmol/L) \| 2.13 (1.63, 2.59) \| 2.20 (1.71, 2.64) \| 1.97 (1.46, 2.47) \| 0.020 \| \| TB (µmol/L) \| 11.9 (8.4, 17.6) \| 11.0 (8.0, 14.4) \| 14.4 (9.5, 20.7) \| 0.001 \| \| DB (µmol/L) \| 5.1 (3.4, 7.2) \| 4.4 (3.2, 6.0) \| 6.2 (4.4, 9.9) \| <0.001 \| \| Alb (g/L) \| 40 (38, 43) \| 42 (39, 44) \| 39 (37, 41) \| <0.001 \| \| HbAlc (%) \| 5.6 (6, 6.4) \| 6.0 (5.6, 6.4) \| 5.9 (5.7, 6.4) \| 0.687 \| \| Echocardiography \| Echocardiography \| Echocardiography \| Echocardiography \| Echocardiography \| \| TAPSE (mm) \| 17.9 ± 3.7 \| 18.9 ± 3.3 \| 16.2 ± 3.7 \| <0.001 \| \| TAPSV (cm/s) \| 10.5 (9.0, 12.8) \| 11.6 (10.2, 13.2) \| 9.1 (7.9, 9.8) \| <0.001 \| \| PASP (mmHg) \| 42 (34, 49) \| 39 (32, 45) \| 45 (36, 56) \| 0.002 \| \| RA diameter (mm) \| 39 (35, 45) \| 37 (33, 42) \| 43 (37, 50) \| <0.001 \| \| RV diameter (mm) \| 21 (20, 24) \| 20 (19.22) \| 22 (20, 26) \| <0.001 \| \| LA diameter (mm) \| 36 (32, 41) \| 36 (32, 40) \| 38 (32, 42) \| 0.133 \| \| LVEDD (mm) \| 47 ± 5.9 \| 46 ± 5.8 \| 47 ± 6.2 \| 0.264 \| \| LAVI (ml/m2) \| 40 (33, 54) \| 40 (33, 53) \| 44 (33, 56) \| 0.147 \| \| LVMI (kg/m) \| 112 (96, 140) \| 113 (96, 140) \| 111(96, 140) \| 0.894 \| \| LVPWT (mm) \| 10.0 (10.0, 11.0) \| 10.0 (10.0, 11.0) \| 10.0 (10.0, 12.0) \| 0.723 \| \| IVST (mm) \| 11.0 (10.0, 12.0) \| 11.0 (10.0, 12.0) \| 10.5 (10.0, 12.0) \| 0.979 \| \| LVEF (%) \| 61 (58, 65) \| 62 (58, 65) \| 60 (57, 64) \| 0.108 \| \| H2FPEF score \| 4 (3, 5) \| 3 (3, 5) \| 5 (3, 6) \| 0.004 \| \| HFA-PEFF score \| 6 (5, 6) \| 6 (5, 6) \| 6 (5, 6) \| 0.088 \| \| RVD (n, %) \| 72 (34.3%) \| 19 (14.6%) \| 53 (66.3%) \| <0.001 \| The echocardiographic characteristics of patients are presented in Table 1. Compared with the normal UA group, patients in the high UA group displayed worse right heart structure and function, including larger right ventricular (RV) diameter and right atrium (RA) diameter, higher PASP, and lower TAPSE and TAPSV, but similar left ventricular end-diastolic dimension (LVEDD), left atrial volume index (LAVI), left ventricular mass index (LVMI), and left ventricular ejection fraction (LVEF). In the present study, the prevalence of RVD was $34.3\%$ among all participants, and the prevalence in the high UA group was four times higher than that in the normal UA group ($66.3\%$ vs. $14.6\%$; $p \leq 0.001$; Table 1). ## The association among UA, TAPSE, TAPSV, and selected variables Spearman’s correlations among UA, TAPSE, TAPSV, and selected variables are summarized in Table 2. UA was positively correlated with PASP, heart rate, SBP, Cr, NT-proBNP, hs-CRP, and DB ($p \leq 0.05$) but negatively correlated with TAPSE, TAPSV, LDL-C, and Alb ($p \leq 0.05$). In addition, TAPSE and TAPSV had positive associations with IVST, LVPWT, LDL-C, and Alb, but were negatively associated with PASP, heart rate, Cr, NT-proBNP, hs-CRP, and DB ($p \leq 0.05$; Table 2). The correlation between UA and echocardiographic characteristics representing right heart dysfunction is indicated in Supplementary Figure S1. Table 2 \| Unnamed: 0 \| TAPSE \| TAPSE.1 \| TAPSV \| TAPSV.1 \| UA \| UA.1 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| \| r \| p-value \| r \| p-value \| r \| p-value \| \| TAPSE (mm) \| – \| – \| 0.711 \| <0.001 \| −0.441 \| <0.001 \| \| TAPSV (cm/s) \| 0.711 \| <0.001 \| – \| – \| −0.495 \| <0.001 \| \| UA (mg/dl) \| −0.441 \| <0.001 \| −0.495 \| <0.001 \| – \| – \| \| PASP (mmHg) \| −0.290 \| <0.001 \| −0.249 \| <0.001 \| 0.250 \| <0.001 \| \| LVPWT (mm) \| 0.216 \| 0.002 \| 0.187 \| 0.007 \| −0.010 \| 0.867 \| \| IVST (mm) \| 0.209 \| 0.001 \| 0.160 \| 0.016 \| −0.050 \| 0.581 \| \| Age (year) \| −0.113 \| 0.104 \| −0.164 \| 0.018 \| 0.088 \| 0.203 \| \| Heart rate \| −0.230 \| 0.001 \| −0.193 \| 0.005 \| 0.214 \| 0.002 \| \| SBP (mmHg) \| 0.081 \| 0.242 \| 0.005 \| 0.940 \| 0.277 \| 0.001 \| \| LDL-C (mmol/L) \| 0.202 \| 0.003 \| 0.225 \| 0.001 \| −0.155 \| 0.025 \| \| Creatinine (µmol/L) \| −0.165 \| 0.017 \| −0.245 \| <0.001 \| 0.514 \| <0.001 \| \| hs-CRP (mg/L) \| −0.205 \| 0.005 \| −0.151 \| 0.040 \| 0.255 \| <0.001 \| \| NT-proBNP (pmol/L) \| −0.473 \| <0.001 \| −0.473 \| <0.001 \| 0.381 \| <0.001 \| \| DB (µmol/L) \| −0.343 \| <0.001 \| −0.327 \| <0.001 \| 0.339 \| <0.001 \| \| Alb (g/L) \| 0.217 \| 0.002 \| 0.195 \| 0.005 \| −0.297 \| <0.001 \| ## ROC curve for the prediction of RVD ROC curves are shown in Figure 2 to demonstrate the diagnostic UA value for the prediction of RVD, which was defined as TAPSE < 17 mm and TAPSV < 9.5 cm/s. The area under the curve (AUC) for RVD was 0.825 ($95\%$ CI, 0.764–0.886; $p \leq 0.001$). The best cutoff value of UA for predicting RVD was 7.15 mg/dl, yielding sensitivity and specificity of $69.4\%$ and $85.5\%$, respectively (Figure 2). Figure 2: The receiver operator characteristic (ROC) curve of UA for predicting RVD. Notes: The area under the curve (AUC) for RVD was 0.825(95% CI, 0.764–0.886). The best cutoff of serum UA to predict RVD was 7.15 mg/dl with a sensitivity of 69.4% and a specificity of 85.5%. ## Univariate and multiple logistic regression analysis with RVD In univariate binary logistic regression analysis, RVD was significantly associated with UA (OR = 2.061; $95\%$ CI, 1.654–2.568; $p \leq 0.001$), AF (OR = 3.508; $95\%$ CI, 1.933–6.366; $p \leq 0.001$), IVST (OR = 0.807; $95\%$ CI, 0.658–0.990; $$p \leq 0.040$$), LVPWT (OR = 0.742; $95\%$ CI, 0.593–0.928; $$p \leq 0.009$$), and other biochemical indexes including NT-proBNP, Cr, DB, and Alb (Table 3). Table 3 \| Unnamed: 0 \| OR \| 95% CI \| p-value \| \| --- \| --- \| --- \| --- \| \| UA (mg/dl) \| 2.061 \| 1.654–2.568 \| <0.001 \| \| Gender (n) \| 1.244 \| 0.699–2.216 \| 0.458 \| \| CAD (n) \| 1.068 \| 0.603–1.893 \| 0.822 \| \| AF (n) \| 3.508 \| 1.933–6.366 \| <0.001 \| \| Diuretics (n) \| 3.099 \| 1.555–6.179 \| 0.001 \| \| SBP (mmHg) \| 0.992 \| 0.974–1.011 \| 0.397 \| \| Heart rate \| 1.023 \| 1,009–1.037 \| 0.001 \| \| LVPWT (mm) \| 0.742 \| 0.593–0.928 \| 0.009 \| \| IVST (mm) \| 0.807 \| 0.658–0.990 \| 0.040 \| \| Cr (µmol/L) \| 1.021 \| 1.010–1.032 \| <0.001 \| \| NT-proBNP (pmol/L) \| 1.006 \| 1.004–1.008 \| <0.001 \| \| hs-CRP (mg/L) \| 0.999 \| 0.992–1.006 \| 0.711 \| \| LDL-C (mmol/L) \| 0.538 \| 0.362–0.800 \| 0.002 \| \| DB (µmol/L) \| 1.174 \| 1.084–1.271 \| <0.001 \| \| Alb (g/L) \| 0.842 \| 0.774–0.916 \| <0.001 \| Multivariable binary logistic regression analysis was performed using variables that were significant in univariate binary logistic regression, and four separate models were developed by the stepwise regression analysis method. Variables of the same type or related were included in one model. Variables that were significant in the first three models were taken into model 4. The results showed that UA was independently associated with RVD in all models (Figure 3). Details of the OR and p-value are listed in Supplementary Table S1. Figure 3: Multivariate binary logistic regression analysis of the effect of UA on the absolute value of right ventricular dysfunction. ## The correlation between UA and the prognosis of HFpEF During a median follow-up period of 278 (190–443) days, 52 ($24.8\%$) patients were readmitted for heart failure, and 11 ($5.2\%$) patients died. Kaplan–Meier curves for heart failure readmission and all-cause mortality are displayed in Figure 4. The rate of heart failure readmission was higher in the high UA group ($$p \leq 0.001$$) compared to the normal UA group, and all-cause mortality also trended to be higher without statistical significance ($$p \leq 0.062$$). The rate of heart failure readmission was higher in both male ($$p \leq 0.002$$) and female ($$p \leq 0.018$$) patients in the high UA group (Supplementary Figure S2A). To better understand the effect of UA on heart failure readmission, univariate and multivariate Cox regression analyses were performed. In univariate Cox regression, high UA (>7 mg/dl), RVD, and NT-proBNP were related to heart failure readmission ($p \leq 0.05$). Indicators in univariate Cox regression were taken into multivariate Cox regression, and high UA (HR = 3.027; $$p \leq 0.002$$) and NT-proBNP (HR = 1.002 for 1 pmol/L increase; $$p \leq 0.01$$) were independently related to heart failure readmission rate after adjusting of high UA, gender, NYHA class, CAD, AF, RVD, NT-proBNP, and Cr (Supplementary Table S2). To gain further insight into the role of UA in patients with HFpEF, a subgroup analysis was also conducted. The results indicated that UA > 7 mg/dl might be a risk factor for heart failure readmission regardless of gender, NT-proBNP, TAPSE, PASP, LVPWT, Alb, and DB values (Supplementary Figure S3). The association between high UA and heart failure readmission was stronger in male than in female patients (male (HR 3.88) and female (HR 2.43) patients). Figure 4: Kaplan–Meier analysis for heart failure readmission (A) and all-cause mortality (B) categorized by serum UA level. ## Discussion The results of this prospective cohort analysis showed that UA levels were associated with the adverse change in right ventricular function in HFpEF patients, with RVD measured by TAPSE and TAPSV. Patients in the high UA group had a higher rate of heart failure readmission, but no statistical difference in all-cause mortality between the two groups was detected. The European Society of Cardiology (ESC) guidelines highlight UA measurement as an additional marker for the stratification of cardiovascular risk [21], although hyperuricemia is common in patients with chronic heart failure with a prevalence of $50\%$ [22], in HFpEF ($26\%$) [23], and in our research ($38\%$). Limited data are available regarding the relationships of UA levels in HFpEF, especially between UA and right ventricular function. In the present study, we found that hyperuricemia was independently associated with right ventricular function in HFpEF. There are few reports about whether UA correlates with right ventricular function in patients with HFpEF, but some information from previous studies suggests that correlations may exist. A previous study in asymptomatic patients with type 2 diabetes demonstrated an independent relationship between UA and biventricular systolic function, regardless of renal function or diabetic control [24]. In patients with idiopathic pulmonary artery hypertension, higher UA levels were suggested to be associated with a lower cardiac index and higher pulmonary vascular resistance [25, 26]. Another study in patients with ischemic heart disease or dilated cardiomyopathy showed hyperuricemia was associated with elevated right atrial pressures [27]. According to those reports and our results, UA levels might correlate with pulmonary artery, right atrial pressure, and right ventricular function. Variables frequently used to assess RV function in patients with heart failure include RV ejection fraction, the longitudinal strain of the RV, TAPSE, and TAPSV. In the current study, TAPSE and TAPSV were used to assess right ventricular function, which is recommended by ASE to improve the accuracy of RVD [20]. They are negatively associated with PASP and TR, the latter two being used in HFpEF diagnosis by the diagnostic algorithm of the HFA-PEFF [28, 29]. The relationship between TAPSE and UA has rarely been reported. In our study, we declared high UA levels were significantly related to lower TAPSE, which is consistent with previous reports in type 2 diabetes patients [24], suggesting that UA might be a biomarker of RVD in patients with HFpEF. The mechanisms to account for elevated UA and RVD in HFpEF patients appear to be unclear and remain to be elucidated. The possible explanations for the findings are as follows: Firstly, increased UA production due to the upregulation of xanthine oxidase and decreased UA excretion owing to lactic acid accumulation and reduced renal perfusion resulted in a high prevalence of hyperuricemia in heart failure patients [30, 31]. Secondly, animal and cell experiments have demonstrated that elevated UA may lead to an increase in cytokine activation, insulin resistance, and oxidative stress, impairing endothelial function and activating the renin–angiotensin system (32–35), which may promote pulmonary vascular remodeling. Meanwhile, the right ventricle is very sensitive to increased pulmonary vascular resistance [36]. Increased PASP promotes right ventricular remodeling. Further studies are still needed to demonstrate whether UA evolves in the development of RVD or is only a risk factor for RVD in HFpEF patients. Previous studies in China did not report gender differences in UA levels in HFpEF patients, so we grouped patients according to Chinese hyperuricemia guidelines: the normal UA group (UA ≤ 7.0 mg/dl) and the high UA group (UA > 7.0 mg/dl). However, the enrolled population showed that gender differences in UA levels did exist. In addition, several studies have shown that the effect of gender on the prognosis of HFpEF is still controversial (37–39). In our study, we performed a gender-specific adjusted analysis and found that hyperuricemia was associated with heart failure readmissions in all patients, more prominently in men. This is in line with the higher comorbidity burden in male heart failure patients [40, 41]. For our results, it is important to carefully consider the gender difference in UA levels and the impact of hyperuricemia on the clinical outcomes of patients with HFpEF. It is widely accepted that UA is an independent predictor of heart failure morbidity and worse outcomes in heart failure population (42–44). However, previous studies demonstrated that the relationship between UA and all-cause mortality or cardiovascular death was controversial both in HFpEF and heart failure with reduced ejection fraction (HFrEF). In the DAPA-HF trial, UA per 1 mg/dl unit increase was not associated with cardiovascular death (HR, 1.06 (0.99–1.14); $$p \leq 0.07$$) or all-cause mortality (HR, 1.03 (0.97–1.1); $$p \leq 0.25$$) in patients with HFrEF [45]. In the EMPEROR-reduced trial, serum UA was an independent predictor of increased mortality (all-cause and cardiovascular mortality) and hospitalization for heart failure when the highest serum UA tertile was compared to the lowest serum UA tertile [4]. In addition, the PARAGON-HF trial and the RELAX trial displayed inconsistent results regarding the relationship between elevated UA and all-cause mortality in HFpEF [23, 46]. Multiple studies in HFrEF have shown that UA-lowering treatment with benzbromarone or allopurinol failed to improve clinical outcomes, exercise capacity, quality of life, and left ventricular systolic function [47, 48]. Research in patients with hyperuricemia and HFpEF showed that UA was a predictor for the composite of all-cause mortality and HF rehospitalization and that lowering UA may improve prognosis [49]. This suggests that UA-lowering therapy might help improve the prognosis of the patient. To date, there has been no study on the relationship between UA-lowering treatment and RV function in HFpEF. Therefore, whether reducing UA would improve RV function in patients with HFpEF need to be demonstrated by designing specific studies. ## Limitations Our study has the following drawbacks: First, the sample size is limited, and the follow-up time is not long enough. Second, we did not collect the dynamic changes of UA and evaluate the prognostic effects of UA reduction. Third, improvement or deterioration of RVD was not assessed during follow-up, so the effect of UA on changes in RVD outcomes could not be obtained. Fourth, as a prospective observational study, we cannot evaluate the potential role of UA in the development and progression of HFpEF. ## Conclusion Overall, elevated UA levels are associated with RVD in HFpEF patients and may be related to heart failure readmission in patients with HFpEF. ## 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 Clinical Research Review Board of the First Affiliated Hospital of Chongqing 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 X-LD contributed to the study design, data analysis, and manuscript preparation. H-WY, JX, X-FZ, JZ, MS, X-SW, Z-QL, LG, Z-YL, and PG were involved in the acquisition of data. D-YZ and QY worked on the study concept, design, and final proof. 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/fendo.2023.1143458/full#supplementary-material ## References 1. Dunlay SM, Roger VL, Redfield MM. Epidemiology of heart failure with preserved ejection fraction, nature reviews. Cardiology (2017) 14 591-602. DOI: 10.1038/nrcardio.2017.65 2. Zhang Y, Zhang J, Butler J, Yang X, Xie P, Guo D. Contemporary epidemiology, management, and outcomes of patients hospitalized for heart failure in China: Results from the China heart failure (China-HF) registry. J cardiac failure (2017) 23. DOI: 10.1016/j.cardfail.2017.09.014 3. Shah SJ, Kitzman DW, Borlaug BA, van Heerebeek L, Zile MR, Kass DA. Phenotype-specific treatment of heart failure with preserved ejection fraction: A multiorgan roadmap. Circulation (2016) 134 73-90. DOI: 10.1161/CIRCULATIONAHA.116.021884 4. Doehner W, Anker SD, Butler J, Zannad F, Filippatos G, Ferreira JP. Uric acid and sodium-glucose cotransporter-2 inhibition with empagliflozin in heart failure with reduced ejection fraction: the EMPEROR-reduced trial. Eur Heart J (2022) 43. DOI: 10.1093/eurheartj/ehac320 5. Gill D, Cameron AC, Burgess S, Li X, Doherty DJ, Karhunen V. Urate, blood pressure, and cardiovascular disease: Evidence from mendelian randomization and meta-analysis of clinical trials. Hypertension (Dallas Tex. 1979) (2021) 77. DOI: 10.1161/HYPERTENSIONAHA.120.16547 6. Kuwabara M, Hisatome I, Niwa K, Hara S, Roncal-Jimenez CA, Bjornstad P. Uric acid is a strong risk marker for developing hypertension from prehypertension: A 5-year Japanese cohort study. Hypertension (Dallas Tex. 1979) (2018) 71 78-86. DOI: 10.1161/HYPERTENSIONAHA.117.10370 7. Li S, Cheng J, Cui L, Gurol ME, Bhatt DL, Fonarow GC. Cohort study of repeated measurements of serum urate and risk of incident atrial fibrillation. J Am Heart Assoc (2019) 8. DOI: 10.1161/JAHA.119.012020 8. Chen-Xu M, Yokose C, Rai SK, Pillinger MH, Choi HK. Contemporary prevalence of gout and hyperuricemia in the united states and decadal trends: The national health and nutrition examination survey, 2007-2016. Arthritis Rheumatol (Hoboken N.J.) (2019) 71. DOI: 10.1002/art.4080710.1016/j.jchf.2020.02.008 9. Koto R, Nakajima A, Horiuchi H, Yamanaka H. Real-world treatment of gout and asymptomatic hyperuricemia: A cross-sectional study of Japanese health insurance claims data. Modern Rheumatol (2021) 31. DOI: 10.1080/14397595.2020.1784556 10. Ting K, Gill TK, Keen H, Tucker GR, Hill CL. Prevalence and associations of gout and hyperuricaemia: results from an Australian population-based study. Internal Med J (2016) 46. DOI: 10.1111/imj.13006 11. Hao Y, Li H, Cao Y, Chen Y, Lei M, Zhang T. Uricase and horseradish peroxidase hybrid CaHPO. J Biomed nanotechnol (2019) 15. DOI: 10.1166/jbn.2019.2752 12. Huang WM, Hsu PF, Cheng HM, Lu DY, Cheng YL, Guo CY. Determinants and prognostic impact of hyperuricemia in hospitalization for acute heart failure. Circ J (2016) 80. DOI: 10.1253/circj.CJ-15-0964 13. Borghi C, Rodriguez-Artalejo F, De Backer G, Dallongeville J, Medina J, Nuevo J. Serum uric acid levels are associated with cardiovascular risk score: A. Int J Cardiol (2018) 253. DOI: 10.1016/j.ijcard.2017.10.045 14. Yang Y, Zhang X, Jin Z, Zhao Q. Association of serum uric acid with mortality and cardiovascular outcomes in patients with hypertension: a meta-analysis. J Thromb thrombolysis (2021) 52. DOI: 10.1007/s11239-021-02453-z 15. Obokata M, Reddy YNV, Melenovsky V, Pislaru S, Borlaug BA. Deterioration in right ventricular structure and function over time in patients with heart failure and preserved ejection fraction. Eur Heart J (2019) 40. DOI: 10.1093/eurheartj/ehy809 16. Dietz MF, Prihadi EA, van der Bijl P, Ajmone Marsan N, Delgado V, Bax JJ. Prognostic implications of staging right heart failure in patients with significant secondary tricuspid regurgitation, JACC. Heart failure (2020) 8. DOI: 10.1016/j.jchf.2020.02.008 17. Gorter TM, van Veldhuisen DJ, Bauersachs J, Borlaug BA, Celutkiene J, Coats AJS. Right heart dysfunction and failure in heart failure with preserved ejection fraction: mechanisms and management. position statement on behalf of the heart failure association of the European society of cardiology. Eur J Heart failure (2018) 20 16-37. DOI: 10.1002/ejhf.1029 18. Rieth AJ, Richter MJ, Tello K, Gall H, Ghofrani HA, Guth S. Exercise hemodynamics in heart failure patients with preserved and mid-range ejection fraction: key role of the right heart. Clin Res Cardiol Off J German Cardiac Soc (2022) 111 393-405. DOI: 10.1007/s00392-021-01884-1 19. Guideline for the diagnosis and management of hyperuricemia and gout in China. Chin J Endocrinol Metab (2020) 36 1-13. DOI: 10.3760/cmaj.issn.1000-6699.2020.01.001 20. Lang RM, Badano LP, Mor-Avi V, Afilalo J, Armstrong A, Ernande L. 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. DOI: 10.1016/j.echo.2014.10.003 21. McDonagh TA, Metra M, Adamo M, Gardner RS, Baumbach A, Böhm M. 2021 ESC guidelines for the diagnosis and treatment of acute and chronic heart failure. Eur Heart J (2021) 42. DOI: 10.1093/eurheartj/ehab368 22. Borghi C, Palazzuoli A, Landolfo M, Cosentino E. Hyperuricemia: a novel old disorder-relationship and potential mechanisms in heart failure. Heart failure Rev (2020) 25 43-51. DOI: 10.1007/s10741-019-09869-z 23. Selvaraj S, Claggett BL, Pfeffer MA, Desai AS, Mc Causland FR, McGrath MM. Serum uric acid, influence of sacubitril-valsartan, and cardiovascular outcomes in heart failure with preserved ejection fraction: PARAGON-HF. Eur J Heart failure (2020) 22. DOI: 10.1002/ejhf.1984 24. Liu JH, Wu MZ, Li SM, Chen Y, Ren QW, Lin QS. Association of serum uric acid with biventricular myocardial dysfunction in patients with type 2 diabetes mellitus. Nutrition metabolism Cardiovasc Dis NMCD (2021) 31. DOI: 10.1016/j.numecd.2021.06.012 25. Cerik IB, Dindas F, Koyun E, Dereli S, Sahin A, Turgut OO. New prognostic markers in pulmonary arterial hypertension: CRP to albumin ratio and uric acid. Clin Biochem (2022) 100. DOI: 10.1016/j.clinbiochem.2021.11.004 26. Yan L, Huang Z, Zhao Z, Zhao Q, Tang Y, Zhang Y. The prognostic impact of serum uric acid on disease severity and 5-year mortality in patients with idiopathic pulmonary artery hypertension. Front Med (2022) 9. DOI: 10.3389/fmed.2022.805415 27. Hoeper MM, Hohlfeld JM, Fabel H. Hyperuricaemia in patients with right or left heart failure. Eur Respir J (1999) 13. DOI: 10.1183/09031936.99.13368299 28. Hsiao SH, Lin SK, Wang WC, Yang SH, Gin PL, Liu CP. Severe tricuspid regurgitation shows significant impact in the relationship among peak systolic tricuspid annular velocity, tricuspid annular plane systolic excursion, and right ventricular ejection fraction. J Am Soc (2006) 19. DOI: 10.1016/j.echo.2006.01.014 29. Mohammed SF, Hussain I, AbouEzzeddine OF, Takahama H, Kwon SH, Forfia P. Right ventricular function in heart failure with preserved ejection fraction: a community-based study. Circulation (2014) 130. DOI: 10.1161/CIRCULATIONAHA.113.008461 30. Bardin T, Richette P. Impact of comorbidities on gout and hyperuricaemia: an update on prevalence and treatment options. BMC Med (2017) 15 123. DOI: 10.1186/s12916-017-0890-9 31. Bergamini C, Cicoira M, Rossi A, Vassanelli C. Oxidative stress and hyperuricaemia: pathophysiology, clinical relevance, and therapeutic implications in chronic heart failure. Eur J Heart failure (2009) 11. DOI: 10.1093/eurjhf/hfp042 32. Gao K, Li Y, Su Y, Lin Z, Yang X, Xu M. High uric acid promotes mitophagy through the ROS/CaMKIIδ/Parkin pathway in cardiomyocytes. Am J Trans Res (2021) 13 33. Kimura Y, Yanagida T, Onda A, Tsukui D, Hosoyamada M, Kono H. Soluble uric acid promotes atherosclerosis. Arteriosclerosis thrombosis Vasc Biol (2020) 40. DOI: 10.1161/ATVBAHA.119.313224 34. Sun X, Jiao H, Zhao J, Wang X, Lin H. Unexpected effect of urate on hydrogen peroxide-induced oxidative damage in embryonic chicken cardiac cells. Free Radical Res (2017) 51 693-707. DOI: 10.1080/10715762.2017.1362106 35. Zhi L, Yuzhang Z, Tianliang H, Hisatome I, Yamamoto T, Jidong C. High uric acid induces insulin resistance in cardiomyocytes. PloS One (2016) 11. DOI: 10.1371/journal.pone.0147737 36. Vonk Noordegraaf A, Westerhof BE, Westerhof N. The relationship between the right ventricle and its load in pulmonary hypertension. J Am Coll Cardiol (2017) 69. DOI: 10.1016/j.jacc.2016.10.047 37. Duca F, Zotter-Tufaro C, Kammerlander AA, Aschauer S, Binder C, Mascherbauer J. Gender-related differences in heart failure with preserved ejection fraction. Sci Rep (2018) 8 1080. DOI: 10.1038/s41598-018-19507-7 38. O'Neal WT, Sandesara P, Hammadah M, Venkatesh S, Samman-Tahhan A, Kelli HM. Gender differences in the risk of adverse outcomes in patients with atrial fibrillation and heart failure with preserved ejection fraction. Am J Cardiol (2017) 119. DOI: 10.1016/j.amjcard.2017.02.045 39. Sotomi Y, Hikoso S, Nakatani D, Mizuno H, Okada K, Dohi T. Sex differences in heart failure with preserved ejection fraction. J Am Heart Assoc (2021) 10. DOI: 10.1161/JAHA.120.018574 40. Lau ES, Cunningham T, Hardin KM, Liu E, Malhotra R, Nayor M. Sex differences in cardiometabolic traits and determinants of exercise capacity in heart failure with preserved ejection fraction. JAMA Cardiol (2020) 5. DOI: 10.1001/jamacardio.2019.4150 41. Merrill M, Sweitzer NK, Lindenfeld J, Kao DP. Sex differences in outcomes and responses to spironolactone in heart failure with preserved ejection fraction: A secondary analysis of TOPCAT trial, JACC. Heart failure (2019) 7. DOI: 10.1016/j.jchf.2019.01.003 42. Gu J, Fan YQ, Zhang HL, Zhang JF, Wang CQ. Serum uric acid is associated with incidence of heart failure with preserved ejection fraction and cardiovascular events in patients with arterial hypertension. J Clin hypertension (Greenwich Conn.) (2018) 20. DOI: 10.1111/jch.13210 43. Krishnan E. Hyperuricemia and incident heart failure, circulation. Heart failure (2009) 2. DOI: 10.1161/CIRCHEARTFAILURE.108.797662 44. Wannamethee SG, Papacosta O, Lennon L, Whincup PH. Serum uric acid as a potential marker for heart failure risk in men on antihypertensive treatment: The British regional heart study. Int J Cardiol (2018) 252. DOI: 10.1016/j.ijcard.2017.11.083 45. McDowell K, Welsh P, Docherty KF, Morrow DA, Jhund PS, de Boer RA. Dapagliflozin reduces uric acid concentration, an independent predictor of adverse outcomes in DAPA-HF. Eur J Heart failure (2022) 24. DOI: 10.1002/ejhf.2433 46. Carnicelli AP, Sun JL, Alhanti B, Bjursell M, Perl S, Lytle B. Elevated uric acid prevalence and clinical outcomes in patients with heart failure with preserved ejection fraction: Insights from RELAX. Am J Med (2020) 133. DOI: 10.1016/j.amjmed.2020.03.054 47. Givertz MM, Anstrom KJ, Redfield MM, Deswal A, Haddad H, Butler J. Effects of xanthine oxidase inhibition in hyperuricemic heart failure patients: The xanthine oxidase inhibition for hyperuricemic heart failure patients (EXACT-HF) study. Circulation (2015) 131. DOI: 10.1161/CIRCULATIONAHA.114.014536 48. Ogino K, Kato M, Furuse Y, Kinugasa Y, Ishida K, Osaki S. Uric acid-lowering treatment with benzbromarone in patients with heart failure: a double-blind placebo-controlled crossover preliminary study, circulation. Heart failure (2010) 3 73-81. DOI: 10.1161/CIRCHEARTFAILURE.109.868604 49. Nishino M, Egami Y, Kawanami S, Sugae H, Ukita K, Kawamura A. Lowering uric acid may improve prognosis in patients with hyperuricemia and heart failure with preserved ejection fraction. J Am Heart Assoc (2022) 11. DOI: 10.1161/JAHA.122.026301
--- title: 'Hidden Blood Loss in Transforaminal Lumbar Interbody Fusion: An Analysis of Underlying Factors' journal: Cureus year: 2023 pmcid: PMC10025578 doi: 10.7759/cureus.35126 license: CC BY 3.0 --- # Hidden Blood Loss in Transforaminal Lumbar Interbody Fusion: An Analysis of Underlying Factors ## Abstract Background In the management of lumbar spine diseases, various techniques have been described for minimizing intraoperative blood loss. Soft tissue extravasation and hemolysis have been referred to as hidden blood loss (HBL). By acknowledging HBL and accounting for it in our postoperative care, strategies of fluid infusion and blood transfusion may be altered. Our study aims to estimate HBL in transforaminal lumbar interbody fusion (TLIF) surgeries and to analyze associated factors. Methods This is a retrospective cohort study. Records of patients who underwent TLIF between January 2016 and December 2020 were reviewed. Patients with both minimally invasive (MIS) and open TLIF were included. Patients with infection, tumors, or fractures being the indication for surgery were excluded. Moreover, patients with known blood-related diseases, aged younger than 18 years, patients requiring blood transfusion, or patients with estimated intra-operative blood loss greater than 1.5 L were excluded. HBL was calculated according to the formulae depending on patients’ weight, height, and hematocrit. Statistical analyses were performed to determine associations between HBL and other factors. Results A total of 95 patients were included. The mean estimated blood loss (EBL) was 231 mL, whereas the mean HBL was 265 mL, and the mean total blood loss is 629.7 ml with HBL accounting for $42\%$ of it. Significant associated factors with HBL were the type of surgery, patient’s total blood volume, preoperative hemoglobin and hematocrit, and decrease in hemoglobin and hematocrit. Conclusion Significant HBL may occur after TLIF, which was shown to be more than EBL. Although MIS had less EBL, it was associated with more HBL. Patients’ preoperative hemoglobin and hematocrit, and a decrease in them, have been shown to be associated with HBL. All these factors should be considered for postoperative management of blood loss. ## Introduction In the management of degenerative lumbar spine diseases, various techniques have been described in pursuit of minimizing intra-operative blood loss and blood transfusion [1-4]. However, there remains a large number of patients who deal with anemia and its associated conditions, which did not correlate with the total blood loss measured. Total blood loss during routine practice is measured by intra-operative bleeding and postoperative drainage, which neglects to take into account the amount of extravasation in the soft tissue and the loss from hemolysis. The concept of hidden blood loss (HBL) was first described in 2000 by Sehat et al., who reported that $26\%$ and $49\%$ of the total blood loss was attributed to HBL in total knee replacement and total hip replacement, respectively [5]. Smorgick et al. reported $39\%$ of HBL from total blood loss in posterior spine fusion surgery [6]. Many studies report HBL in lumbar fusion surgeries that ranged from 227-600 mL according to Zhou et al. [ 7]. By retrospectively reviewing patients' medical records the aim of this study was to measure the amount of HBL in transforaminal lumbar interbody fusion (TLIF) and to analyze the associated factors. ## Materials and methods Study design and participants This retrospective cohort study was carried out in a tertiary hospital, King Abdulaziz Medical City, Jeddah, Saudi Arabia, and included patients who underwent TLIF that was performed between January 2016 and December 2020, and the surgery did not exceed three vertebral levels. We excluded patients with spondylodiscitis, lumber fracture, tumor, known blood-related diseases, age younger than 18 years, patients requiring blood transfusion, or patients with estimated intra-operative blood loss greater than 1.5 L. Consecutive sampling technique was used. Surgical techniques Open TLIF Open TLIF was done as described by Harms and Jeszensky using a longitudinal midline incision [8]. Minimally Invasive Surgery (MIS)-TLIF Above the symptomatic side, a paramidline skin incision measuring approximately 3 cm is made, then the quadrant retractor system is placed until the desired working diameter is achieved. After that, the posterolateral element of the spine is exposed. Then laminectomy and facetectomy are performed and followed by subtotal discectomy and preparation of the endplate for fusion. A cage is placed and a bone graft is packed anterior to it. Screws and rods were then placed percutaneously on both sides, and compression was applied across the cage. Thorough washing and hemostasis followed by the closure of the skin in layers are done. Data collection Electronic hospital records were reviewed to collect patients’ demographic data, such as age, gender, weight, height, and comorbidities. Moreover, patients’ preoperative hemoglobin (Hgb), hematocrit (Hct), prothrombin time (PT), activated partial thromboplastin time (aPTT), and international normalized ratio (INR) were collected. Paraspinal muscle thickness and subcutaneous fat thickness were determined using pre-operative magnetic resonance imaging (MRI). Furthermore, the American Society of Anesthesiologist (ASA)'s classification and indication for surgery for each patient were also gathered. Perioperative variables such as vertebral level involved, surgical duration, estimated blood loss (EBL), and type of surgery, whether open or minimally invasive, were obtained. The immediate postoperative parameters that were collected were the length of hospital stay (LOH) and the minimum time before ambulation (TBA). Calculating HBL The patient’s blood volume (PBV) was calculated according to the formula of Nadler et al., which is PBV = k1 × height (m)3 + k2 × weight (kg) + k3 (for male: k1 = 0.3669, k2 = 0.03219, and k3 = 0.6041; for female: k1 = 0.3561, k2 = 0.03308, and k3 =0.1833) [9]. Total blood loss was then calculated by multiplying PBV by the change of Hct, according to the gross formula in which total blood loss = PBV (Hctpre−Hctpost)/Hctave, where Hctpre is the preoperative Hct, Hctpost is the postoperative Hct, nd Hctave is the average of Hctpre and Hctpost [10]. Finally, HBL was calculated according to the formula of Sehat et al., which is HBL= total blood loss − measured blood loss [5]. Statistical analyses Statistical analyses were performed using IBM SPSS Statistics for Windows, Version 25.0 (Released 2017; IBM Corp., Armonk, New York, United States). Data were summarized using descriptive statistics, such as mean ± standard deviation, frequency, and percentage. After examining all variables for normality by calculating the skewness and kurtosis z-values, and after using the Shapiro-Wilk test, a Pearson correlation, Student’s t-test, and ANOVA for normally distributed variables, and Spearman correlation for non-normally distributed variables, were used to determine the association between the dependent and independent variables. A p-value of less than 0.05 was considered significant. Ethical considerations The Institutional Review Board of King Abdullah International Medical Research Center, Jeddah, Saudi Arabia, approved the study (approval number: RJ$\frac{19}{090}$/J). Patients’ names, medical record numbers, and identifying data were not collected. Data were accessible exclusively by the authors. ## Results A total of 95 patients (42 males and 53 females, age range 20-84 years) were included. Demographics, comorbidities, ASA classification, and subcutaneous fat and muscle thickness with the significant of their association to HBL are presented in Table 1. Table 1 \| Parameter \| Statistics \| P-value \| \| --- \| --- \| --- \| \| Gender (%) \| \| 0.08 \| \| Males \| 44.2 \| \| \| Females \| 55.8 \| \| \| Age (years ± SD) \| 58.23 ± 12.5 \| 0.34 \| \| BMI (kg/m2 ± SD) \| 31.4 ± 4.8 \| 0.67 \| \| Subcutaneous fat thickness (mm) \| 31.76 ± 15.7 \| 0.63 \| \| Muscle thickness (mm) \| 47.61 ± 8 \| 0.51 \| \| Diabetes mellitus (%) \| 42.3 \| 0.19 \| \| Hypertension (%) \| 52.6 \| 0.45 \| \| Chronic kidney disease (%) \| 7.2 \| 0.65 \| \| Coronary artery disease (%) \| 5.2 \| 0.95 \| \| Smoking (%) \| 12.4 \| 0.21 \| \| On antiplatelet (%) \| 14.4 \| 0.79 \| \| ASA classification (%) \| \| 0.9 \| \| I \| 11.3 \| \| \| II \| 72.2 \| \| \| III \| 16.5 \| \| Around $63\%$ of patients had MIS-TLIF, whereas $38\%$ had open-TLIF. Degenerative disc disease was the indication for surgery in $43.2\%$ of patients. Moreover, the most common vertebral level involved was L4-5 (approximately $37.1\%$ of patients). More details on indications of surgeries, levels involved in surgical details, and postoperative parameters are shown in Table 2. Table 2 \| Parameter \| Statistics \| P-value \| \| --- \| --- \| --- \| \| Time since surgery (months ± SD) \| 26 ± 14.6 \| 0.23 \| \| Type of surgery \| \| 0.02 \| \| Open TLIF (%) \| 38.1 \| \| \| MIS TLIF (%) \| 61.9 \| \| \| Indication for surgery (%) \| \| 0.26 \| \| Degenerative disc disease \| 42.3 \| \| \| Spinal stenosis \| 20.6 \| \| \| Spondylolisthesis \| 27.8 \| \| \| Prolapsed intervertebral discs \| 9.3 \| \| \| Levels of surgery (%) \| \| 0.82 \| \| L1-L2 \| 2.1 \| \| \| L2-L3 \| 1 \| \| \| L3-L4 \| 6.2 \| \| \| L4-L5 \| 37.1 \| \| \| L5-S1 \| 14.4 \| \| \| Two levels \| 33 \| \| \| Three levels \| 6.2 \| \| \| Duration of surgery (hours ± SD) \| 3.6 ± 1.1 \| 0.71 \| \| Drainage (ml ± SD) \| 497.8 ± 294.3 \| 0.13 \| \| Hospitalization time (days ± SD) \| 4.49 ± 3.2 \| 0.14 \| \| Time before ambulation (days ± SD) \| 1.54 ± 1.3 \| 0.66 \| The mean EBL was 231.2 ± 166.4 which was less than the mean HBL (265.1 ± 411.7). Patients’ preoperative Hb and Hct were significantly associated with HBL as well as the decrease in them postoperatively. More details in blood loss and pre and postoperative laboratory investigations are shown in Table 3. Table 3 \| Parameter \| Statistics \| P value \| \| --- \| --- \| --- \| \| PBV (L ± SD) \| 4.64 ± 0.77 \| 0.03 \| \| EBL (ml ± SD) \| 231.2 ± 166.4 \| 0.06 \| \| HBL (ml ± SD) \| 265.1 ± 411.7 \| \| \| TBL (ml ± SD) \| 629.7 ± 411.2 \| 0.0 \| \| Preoperative Hct ± SD \| 41.1 ± 4.5 \| 0.0 \| \| Preoperative Hb (g/l ± SD) \| 13.3 ± 1.5 \| 0.0 \| \| Postoperative Hct ± SD \| 35.8 ± 4.8 \| 0.82 \| \| Postoperative Hb (g/l ± SD) \| 11.8 ± 1.6 \| 0.99 \| \| Decrease Hct ± SD \| 5.3 ± 3.3 \| 0.0 \| \| Decrease Hb (g/l ± SD) \| 1.5 ± 1.1 \| 0.0 \| \| aPTT (s ± SD) \| 29.8 ± 4.2 \| 0.39 \| \| PT (s ± SD) \| 11.9 ± 1 \| 0.12 \| \| INR \| 1 ± 0.9 \| 0.13 \| When comparing total blood loss between the MIS-TLIF group and the open-TLIF group, the mean total blood loss in MIS-TLIF was 471.2 ± 316 mL and in open-TLIF was 886.78 ± 421.3 mL (P˂0.00). ## Discussion Spine surgery has seen significant technological advances over the last decades, and various methods are being used to decrease expected surgical blood loss during spinal surgery. Sehat et al. first described HBL in total hip arthroplasty; HBL accounted for $49\%$ of the total blood loss. In their study, they highlight that HBL is thought to be due to hemolysis and extravasation into third spaces [5]. Jiang et al. found a mean of $46.8\%$ HBL in cervical open laminoplasty, whereas Wen et al. reported a mean of $39\%$ HBL in PSF [11,12]. Moreover, Smorgick et al. estimated HBL to be $40\%$ [6], which was similar to our study ($42\%$ of total blood loss was attributed to HBL). In the current study, the mean EBL was 231.2 ± 166.4 mL and the mean HBL was 265.1 ± 411.7 mL. With the HBL being more than the EBL, this shows that a significant amount of bleeding is beyond our scope of estimation, which may result in an inaccurate measurement of peri-operative blood loss. This was also noted by Ogura et al. in their study in which the HBL was 685 mL more than the mean EBL [13]. In the literature, few articles have analyzed the associated risk factors with HBL. In the present study, MIS-TLIF, preoperative low levels of hemoglobin, blood volume, and low Hct were significantly associated with an exceptional increase in HCL. With MIS-TLIF, we aimed to achieve less operative dissection, less surgical time, and better outcomes. The downside of this less invasive technique is that it limits our efforts to achieve precise homeostasis. In a study conducted by Zhang et al., they found the MIS-TLIF technique to be associated with more HBL, which they justified by not using postoperative drainage [7]. Nevertheless, we found that MIS-TLIF showed a significantly lower total blood loss relative to open-TLIF. In our results, total blood loss in the MIS-TLIF group had approximately $50\%$ less total blood loss than the open-TLIF group. In contrast to Lei et al., we found that a higher BMI did not significantly affect HBL, which was supported by a retrospective multicentre study by Miao et al. [ 14,15]. However, Goyal et al. reported that greater muscle thickness, as assessed by magnetic resonance imaging (MRI), is significantly associated with HBL [16]. Greater muscle thickness was analyzed in our study and was found to be non-significant. This could be attributed to the fact that more than half of our cohort had MIS-TLIF, which has been shown not to be associated with significant blood loss, regardless of BMI. Our study also assessed the impact of antiplatelets on HBL. As per our preoperative preparation, all antiplatelet medications were kept on hold for seven days prior to the surgery. In this study, $15\%$ (14 patients) were on lifelong antiplatelet medications and followed these instructions and, thus, the use of antiplatelets showed no significant association with HBL. When we decided to go for one or two levels of TLIF, we opted for the MIS technique. In cases requiring multiple levels of TLIF, we achieved it with the open technique. The number of levels was not significant in the relation to HBL as an independent risk factor. This is due to the fact that the multiple-level surgeries were open technique, thus less HBL. We did not find a significant association between HBL and coagulation profile because all patients had a normal coagulation profile before surgery, while those with abnormal coagulation profiles were medically optimized. Nevertheless, our study has some drawbacks. Our sample size was relatively small; larger numbers are needed for weaker correlations to be significant. Moreover, we measured Hgb and Hct on days 2 and 3 post surgery. However, as fluid shifts would not have been completed in all patients at this time, the HBL obtained might be falsely low. ## Conclusions A significant amount of HBL may occur after TLIF, which is shown to be more than EBL. HBL was found to be correlated with patients’ preoperative Hb and Hct and was found to be significantly more elevated with MIS-TLIF. However, MIS-TLIF resulted in more HBL, and HBL was significantly lower in total blood loss relative to open-TLIF. All these factors should be considered for postoperative management of blood loss ## References 1. Ghogawala Z, Dziura J, Butler WE. Laminectomy plus fusion versus laminectomy alone for lumbar spondylolisthesis. N Engl J Med (2016) 374 1424-1434. PMID: 27074067 2. Chen G, Li LB, Shangguan Z, Wang Z, Liu W, Li J. Clinical effect of minimally invasive microendoscopic-assisted transforaminal lumbar interbody fusion for single-level lumbar disc herniation. Orthop Surg (2022) 14 3300-3312. PMID: 36303440 3. Lavelle WF, Beltran AA, Carl AL, Uhl RL, Hesham K, Albanese SA. Fifteen to twenty-five year functional outcomes of twenty-two patients treated with posterior Cotrel-Dubousset type instrumentation: a limited but detailed review of outcomes. Scoliosis Spinal Disord (2016) 11 18. PMID: 27551692 4. Khashab MA, Alswat MM, Alsofiani BS, Algamedi OA, Shulan MF. Open versus minimally invasive transforaminal lumbar interbody fusion: Intermediate outcomes in overweight and obese patients. J Musculoskelet Surg Res (2020) 4 82-86 5. Sehat KR, Evans RL, Newman JH. Hidden blood loss following hip and knee arthroplasty. Correct management of blood loss should take hidden loss into account. J Bone Joint Surg Br (2004) 86 561-565. PMID: 15174554 6. Smorgick Y, Baker KC, Bachison CC, Herkowitz HN, Montgomery DM, Fischgrund JS. Hidden blood loss during posterior spine fusion surgery. Spine J (2013) 13 877-881. PMID: 23523442 7. Zhou Y, Fu X, Yang M, Ke S, Wang B, Li Z. Hidden blood loss and its possible risk factors in minimally invasive transforaminal lumbar interbody fusion. J Orthop Surg Res (2020) 15 445. PMID: 32993729 8. Harms J, Jeszensky D. The unilateral transforaminal approach for posterior lumbar interbody fusion. Orthop Traumatol (1998) 6 88-99 9. Nadler SB, Hidalgo JH, Bloch T. Prediction of blood volume in normal human adults. Surgery (1962) 51 224-232. PMID: 21936146 10. Gross JB. Estimating allowable blood loss: corrected for dilution. Anesthesiology (1983) 58 277-280. PMID: 6829965 11. Jiang C, Chen TH, Chen ZX, Sun ZM, Zhang H, Wu YS. Hidden blood loss and its possible risk factors in cervical open-door laminoplasty. J Int Med Res (2019) 47 3656-3662. PMID: 31234677 12. Wen L, Jin D, Xie W. Hidden blood loss in posterior lumbar fusion surgery: an analysis of risk factors. Clin Spine Surg (2018) 31 180-184. PMID: 29528878 13. Ogura Y, Dimar Ii JR, Gum JL, Crawford CH 3rd, Djurasovic M, Glassman SD, Carreon LY. Hidden blood loss following 2- to 3-level posterior lumbar fusion. Spine J (2019) 19 2003-2006. PMID: 31326629 14. Lei F, Li Z, He W, Tian X, Zheng L, Kang J, Feng D. Hidden blood loss and the risk factors after posterior lumbar fusion surgery: a retrospective study. Medicine (Baltimore) (2020) 99 0 15. Miao K, Ni S, Zhou X, Xu N, Sun R, Zhuang C, Wang Y. Hidden blood loss and its influential factors after total hip arthroplasty. J Orthop Surg Res (2015) 10 36. PMID: 25889223 16. Goyal A, Elminawy M, Kerezoudis P, Lu VM, Yolcu Y, Alvi MA, Bydon M. Impact of obesity on outcomes following lumbar spine surgery: a systematic review and meta-analysis. Clin Neurol Neurosurg (2019) 177 27-36. PMID: 30583093
--- title: 'FAMetA: a mass isotopologue-based tool for the comprehensive analysis of fatty acid metabolism' authors: - María I Alcoriza-Balaguer - Juan C García-Cañaveras - Marta Benet - Oscar Juan-Vidal - Agustín Lahoz journal: Briefings in Bioinformatics year: 2023 pmcid: PMC10025582 doi: 10.1093/bib/bbad064 license: CC BY 4.0 --- # FAMetA: a mass isotopologue-based tool for the comprehensive analysis of fatty acid metabolism ## Abstract The use of stable isotope tracers and mass spectrometry (MS) is the gold standard method for the analysis of fatty acid (FA) metabolism. Yet, current state-of-the-art tools provide limited and difficult-to-interpret information about FA biosynthetic routes. Here we present FAMetA, an R package and a web-based application (www.fameta.es) that uses 13C mass isotopologue profiles to estimate FA import, de novo lipogenesis, elongation and desaturation in a user-friendly platform. The FAMetA workflow covers the required functionalities needed for MS data analyses. To illustrate its utility, different in vitro and in vivo experimental settings are used in which FA metabolism is modified. Thanks to the comprehensive characterization of FA biosynthesis and the easy-to-interpret graphical representations compared to previous tools, FAMetA discloses unnoticed insights into how cells reprogram their FA metabolism and, when combined with FASN, SCD1 and FADS2 inhibitors, it enables the identification of new FAs by the metabolic reconstruction of their synthesis route. ## Introduction Fatty acids (FAs) are key metabolites that play a central role in cellular biology. FAs act as building blocks for the synthesis of complex lipids or as a source of energy, but also as signaling molecules [1]. Dysregulated FA metabolism has been associated with many of the most prevalent diseases, including obesity [2], type 2 diabetes [3], non-alcoholic fatty liver disease [4], or cancer [5]. FAs can be either synthesized de novo inside cells or imported from external sources. The main product of de novo lipogenesis (DNL) is palmitic acid [FA(16:0)], which results from the condensation of acetyl-CoA molecules through the enzymatic action of acetyl-CoA carboxylase (ACACA/B) and FA synthase (FASN). The acetyl-CoA pool is generated via ATP citrate lyase (ACLY) from citrate that can, in turn, be produced from several carbon sources (i.e. glucose, glutamine, amino acids, FAs), or from acetate via acetyl-CoA synthetases (ACSS$\frac{1}{2}$) [6]. Linoleic [FA(18:2n6)] and γ-linolenic acid [FA(18:3n3)] are essential FAs that must be exogenously acquired. Free FA import occurs by either passive diffusion or the action of translocases like CD36 and FA transport proteins (FATPs). FAs can be elongated via very long-chain FA proteins (ELOVL1-7). They can also be desaturated via the action of stearoyl-CoA desaturases $\frac{1}{5}$ (SCD$\frac{1}{5}$) and FA desaturases $\frac{1}{2}$ (FADS$\frac{1}{2}$) enzymes [1, 5]. The wide variety of FAs required for the cellular functioning results from these transformations. Stable-isotope tracing combined with mass spectrometry (MS)-based detection is a widespread method for interrogating FA metabolism. The total FA synthesis rate can be estimated by using D2O, which labels FAs through direct solvent incorporation and NADPH-mediated hydrogen transfer [7, 8]. Additionally, employing 13C-labeled tracer nutrients (e.g. U-13C-glucose, U-13C-glutamine, U-13C-acetate, etc.) allows the total FA synthesis rate and the relative contribution of a given nutrient to be estimated [9]. The framework for FA synthesis data analysis using 13C-labeled tracers and MS was initially set up by Mass Isotopomer Distribution Analysis (MIDA) [10] and Isotopomer Spectral Analysis (ISA) [11], which model FA synthesis following the incorporation of n 2-carbon units using multinomial distribution fitting. Unfortunately, these mass isotopologue modeling methods only provide information about the DNL of FAs for which the contribution of elongation is minimal (i.e. FAs of 14 or 16 carbons) [10–12]. ConvISA incorporated one elongation to model up to 18-carbon FAs [13]. Recently, Fatty Acid Source Analysis (FASA) included many elongation steps, which extend the FA species that can be properly modeled up to 26 carbons [14]. However, FASA present some limitations as it assumes de novo synthesis up to 26-carbon FAs, and it calculates multiple import-elongation terms (i.e. IEn, which refers to imported and elongated n times), which does not accurately represent the real biological process. Concerning FA desaturation, only a simple strategy for estimating the desaturation of FA(18:0) to FA(18:1n9) has been described [15]. However, this approach is based on the total labeling of precursor and product FAs [15], and its application to the complete array of desaturations has not yet been explored. Despite these valuable advances, reliable FA elongation calculations are still not fully addressed, whereas systematic desaturation estimations remain unresolved. Moreover, the above-mentioned algorithms were developed for platforms that require computational skills and commercial software; thus, they are not readily accessible to the broad metabolism community. To bridge this gap, we developed FAMetA (Fatty Acid Metabolism Analysis), a mass isotopologue-based tool implemented as an R package and a web-based application that aims to analyze all the biosynthetic reactions within the FA metabolic network. FAMetA provides a complete workflow to analyze MS data and returns easy-to-interpret results that facilitate straightforward FA metabolism analyses and the identification of unknown FAs. ## FAMetA overview FAMetA is an R package (https://CRAN.R-project.org/package=FAMetA) and a web-based platform (https://www.fameta.es) that rely on mass isotopologue distributions from GC–MS or LC–MS to estimate the import (I), the synthesis of FA(14:0)/FA(16:0) (S), the fractional contribution of the 13C-tracer (D0, D1 and D2, which represent the acetyl-CoA fraction with 0, 1, or 2 atoms of 13C, respectively), the elongation (E) and the desaturation (Δ) parameters for the expected network of FA synthesis reactions up to 26 carbons [16] (Figure 1, Supplementary Figure 1). The FAMetA workflow comprises the required functionalities needed, from data preprocessing to group-based comparisons and graphical output (Figure 1, Supplementary Figures 2–3). Mass isotopologue distributions usually show overdispersion, which can be attributed to cellular heterogeneity, time-dependent variations that result from changes in nutrient availability, differences between the various intracellular FA pools (e.g. differences between lipid classes or between FA/lipids located in different organelles), among others. FAMetA implements quasi-multinomial modeling that improves the fitting of mass isotopologue distributions compared with formerly used multinomial modeling [10–14, 17, 18] (Figure 2). Furthermore, this fitting provides the parameter Φ that accounts for data overdispersion. For FAs up to 16 carbons, the DNL parameters (I, S, Φ and D0, D1, D2) are estimated. The equations employed to fit the experimental isotopologue distribution are equivalent to those employed by the ISA algorithm [11, 12] if the parameter Φ = 0 and if the ISA equations are modified to take into account that the data have been corrected for the natural abundance of 13C. Unlike the original ISA algorithm [11, 12], which is designed for calculation of the DNL, or ConvISA [13], which only calculates elongation for FA(18:0), for the FAs of 18–26 carbons, apart from the parameters S and I, FAMetA estimates up to five elongation terms (En, $$n = 1$$ for 18-carbon to $$n = 5$$ for 26-carbon FAs). Each elongation term represents the direct estimation of the fraction that comes from the elongation of the total pool of the precursor FA (Figure 3). Compared with previous tools (i.e. FASA, where the synthesis of an FA longer than 16 carbons is described as DNL up to the total length and multiple import-elongation terms are implemented [14]), the way in which elongations are calculated by FAMetA better reflects how FAs are elongated within the cells, which permits the straightforward biological interpretation of the reported elongation parameters. For the FAs that result from the direct desaturation of one precursor FA, Δ is indirectly estimated based on the calculated synthesis parameters of the precursor (S or E) and the FA of interest (S′ or E’) (i.e. Δ = S′/S or Δ = E’/E) (Figure 3). The strategy proposed here is inspired in the simple approach described by Kamphorst et al. [ 15, 19]. The authors calculate desaturation for FA(18:1n9) based on the total labeling in FA(18:0) and FA(18:1n9). We extend the strategy to the complete set of desaturations within the FA metabolic network and refine the calculation by using an approach that uses the estimated synthesis parameter of interest instead of the total labeling. The use of the complete isotopologue distribution to estimate the substrate and product FA synthesis parameters of interest instead of a single summed value may lead to a more robust and accurate estimation of desaturation. However, the key advance of how FAMetA calculates desaturations compared to the calculation proposed by Kamphorst et al. is the possibility of estimating alterations in concrete desaturation steps from the complete set of parameters calculated for a given FA [e.g. identify alterations in SCD activity between two conditions based on the information obtained for FA(18:1n7), where the double bond is introduced at the 16-carbon level]. Finally, the complete metabolic network of FA synthesis is summarized for each sample and group, and comparisons between groups are made and graphically represented (Figure 1). As in previous tools (i.e. ISA, ConvISA and FASA [11–14]) the de novo synthesis parameters (S, E, Δ) are time-dependent. Therefore, at any given time, such parameters correspond to the fraction of a particular FA that has been de novo synthesized up-to-the moment of the sampling, and it corresponds to the actual portion of FA that comes from de novo synthesis only if the steady state has been achieved. Accordingly, the import term ($I = 1$ − S or 1 − En) accounts for both import and pre-existing FAs at any given time and to the actual fraction that is acquired from the exogenous pool when the steady state has been reached. The conditions of metabolic and isotopic steady states are only achieved, or can be closely approximated, if the cells are cultured during a long-enough time to ensure that the pre-existing FA pools can be diluted out while ensuring a nutrient supply that maintains relatively stable concentrations [14, 20]. A key feature of FAMetA compared to previous tools/algorithms is its ease of use and implementation. Although previous tools have been released as Matlab scripts or implemented within complex software that require extra tools for data-preprocessing and graphical representation, FAMetA R package and web-page (www.fameta.es) provide the whole workflow for MS data analysis. A comprehensive comparison of the functions implemented by FAMetA and other available tools is summarized in Supplementary Table 1. Figure 1: FAMetA workflow. FAMetA is an open-access platform-independent software for estimating FA metabolism based on mass isotopologue data that can be executed locally in an R environment (https://CRAN.R-project.org/package=FAMetA) or online (www.fameta.es). FAMetA uses as input the data generated after incubation/infusion with suitable 13C-tracers, based on the LC–MS or GC–MS analysis of FA extracts. Briefly, the FAMetA workflow comprises the following steps: (i) data preprocessing; (ii) analysis of the FA metabolism for each sample and detected FA; and (iii) the combination of these individual results to provide an overview of the FA metabolism network for each condition of interest and to compare them. The analysis of the FA metabolism parameters is based on fitting the experimental mass isotopologue distribution to a quasi-multinomial distribution. For FAs of 14 and 16 carbons, the imported fraction (I) and the DNL parameters (i.e. synthesis (S), fractional contribution of the tracer (D0, D1 and D2), and overdispersion (Φ)) are calculated. For FAs of more than 16 carbons, the D0, D1, D2 and Φ parameters are imported and the sources are described as the import (I) plus the elongation (En) of the total pool of the precursor FA [e.g. for FA(18:0), sources are described as I18:0 + E1 = I18:0 + E1 (I16:0 + S)]. For the FAs that are the result of desaturation, sources are described as the import (I), plus desaturation (Δ), where Δ is indirectly estimated according to the synthesis parameters of the precursor (S or E) and product FAs (S′ or E’), where Δ = S′/S or E’/E. Depending on the experimental design, the most relevant biological outputs to be obtained include the fractional contribution of each tested carbon source, the detailed description of the metabolic origin of each detected FA and the elucidation of an alteration in the FA metabolism between conditions of interest. Figure 2: Fitting experimental mass isotopologue FA data to multinomial and quasi-multinomial distributions. (A–B) FA(16:0) in the A549 cells upon incubation with U-13C-glucose (A) or U-13C-glutamine (B), data obtained from ref. [18]. (C–D) FA(14:0) (C) and FA(16:0) (D) in the H1299 cells upon incubation with U-13C-glucose, data obtained from ref. [14]. (E–F) FA(16:0) (E) and FA(18:0) (F) in the MCF7 cells upon incubation with U-13C-glucose, data obtained from ref. [13]. For each dataset, the experimental data, the fitting done using the FAMetA algorithm with multinomial or quasi-multinomial distributions, and the residuals are shown. The reported P-values correspond to the comparisons between multinomial and quasi-multinomial fitting using a log-likelihood ratio test and right-tailed chi-square distribution. Figure 3: Example of the FAMetA calculations for FA(16:0) to FA(20:1n9). A detailed description of the calculation of FA sources, reported endogenous synthesis and the parameters calculated for the FAs: FA(16:0), FA(18:0), FA(18:1n9), and FA(20:1n9). ## FAMetA validation In silico mass isotopologue distributions are generated to validate the FAMetA algorithm. To simulate experimental distributions, multiple values covering the expected range for each parameter are used. For each theoretical isotopologue distribution, 10 realizations of Gaussian noise are simulated at four noise levels [0, 2, 5, or $10\%$ relative standard deviation (RSD)]. The generated data are used to calculate the RSD and relative error of each modeled synthesis parameter for the following FAs, which comprise an example of all the reactions included in FAMetA: FA(16:0) (Supplementary Figure 4), FA(18:0) (Supplementary Figure 5), FA(20:0) (Supplementary Figure 6), FA(22:0) (Supplementary Figure 7), FA(24:0) (Supplementary Figure 8), FA(16:1n7) (Supplementary Figure 9) and FA(18:1n9) (Supplementary Figure 9). FAMetA accurately determines the complete set of FA synthesis parameters (relative error < $15\%$, RSD < $15\%$) whenever the fractional contribution of the tracer (D2) and the parameters to be calculated for a given FA (i.e. S, E1, E2, E3 and E4) fall within the 0.05–0.9 range. This ensures its applicability in an actual biological scenario. ## FAMetA enables straightforward FA metabolism analyses To evaluate FAMetA performance, a variety of in vitro and in vivo experimental settings are used. First, mouse CD8+ T cells are incubated for 72 h with different uniformly 13C labeled tracers (U-13C-glucose, U-13C-glutamine, U-13C-lactate, or U-13C-acetate) in the presence or absence of well-known inhibitors of FA metabolism enzymes [i.e. FASN (GSK2194069, FASNi) [21], SCD1 (A93572, SCD1i) [22, 23] and FADS2 (SC26196, FADS2i) [24]]. Total lipids are extracted from cell pellets and saponified to release FAs, which are subsequently analyzed by LC–MS. Twenty-seven known FAs are detected in the samples, including a variety of saturated, monounsaturated and polyunsaturated FAs within the range from 14 to 24 carbons. FAMetA accurately models the obtained mass isotopologue distributions for all of them and extracts valuable biological information about nutrient preferences and metabolic origin of each particular FA (Figure 4A–E, Supplementary Results). Figure 4: Biological validation of FAMetA in active mouse CD8+ T cells. Estimation of the FA metabolism parameters in the active mouse CD8+ T cells incubated for 72 h with various U-13C-tracers. (A–D) Estimation of the sources and the DNL parameters for FA(16:0) upon incubation with U-13C-glucose (A), U-13C-glutamine (B), U-13C-lactate (C) or U-13C-acetate (D). (E) Summary of the endogenously synthesized fraction for the 27 known FAs detected in the active mouse CD8+ T cells upon incubation with U-13-Cglucose. (F–L) Analysis of alterations in FA biosynthesis in the active mouse CD8+ T cells incubated for 72 h with U-13C-glucose induced by FASN inhibitor GSK2194069, SCD inhibitor A93572 and FADS2 inhibitor SC26196. (F) Mean proliferation of the active mouse CD8+ T cells during the 72-h incubation period. (G) Heatmap showing for each identified FA the mean value of the log2 fold-of-change (versus untreated) in the relative pool size. (H) Heatmap showing the mean value of the log2 fold-of-change (versus untreated) for each identified FA in the following parameters: endogenously synthesized fraction, calculated S, E1, E2, E3 and E4. For each FA, the parameter reported for the endogenous synthesis is indicated. (I–L) Mass isotopologue distribution, the mean value of the log2 fold-of-change (versus untreated) in the synthesis parameters and synthesis route for FA(18:1n7) (I), FA(18:1n9) (J), FA(20:1n9) (K) and FA(20:3n9) (L). In all cases, n = 3. Individual points are shown for the mass isotopologue distributions, and the mean values are reported elsewhere. The shadowed cells in (B) and (C) indicate the activities (DNS, SCD or FADS2) involved in the synthesis of a particular FA. On the heatmaps, crosses indicate missing or NA values. In (I–L), the horizontal transitions in the synthesis route description denote elongations (enzymes not indicated), and vertical transitions denote desaturations (enzymes indicated). Treatment with FASNi and SCD1i slightly decreases cell proliferation, but FADS2i does not (Figure 4F). Changes in the relative pool size of the detected FAs appear (Figure 4G); e.g. SCD1i lowers the intracellular levels of the n5, n7 and n9 series FAs, and increases the relative abundance of FADS2 products [e.g. sapienic acid, FA(16:1n10)], whereas FADS2i considerably diminishes sapienic acid abundance, which is consistent with previous reports on the complementary and compensatory roles of SCD1 and FADS2 [25] (Figure 4G). When analyzing endogenous synthesis, the changes reveal which enzymes are involved in the synthesis of each identified FA. FASNi decreases the endogenous synthesis of all the FAs that come from FA(16:0), and SCD1i and FADS2i decrease the endogenous synthesis of all the FAs that these enzymes are involved in (e.g. n9 series FAs for SCD1i, n10 series FAs for FADS2i) (Figure 4H). When focusing on each calculated synthesis parameter, identifying the step in which each enzyme acts and mapping synthesis routes are straightforward. For example, for FA(18:1n7) and FA(18:1n9), SCDi differentially affects synthesis parameters. In FA(18:1n9), where SCD acts at the 18-carbon level, the most prominent decrease is in calculated E1 (i.e. E1’ = E1Δ), in FA(18:1n7), where SCD acts at the 16-carbon level, both calculated S (i.e. S′ = SΔ), and E1 decreases upon treatment with SCDi (Figure 4I–J). The SCDi inhibition pattern observed in FA(18:1n9) is mirrored in FA(20:1n9) and FA(20:3n9) (Figure 4H–L). In addition, FADS2i decreases the calculated E2 (i.e. E2’ = E2Δ) for FA(20:3n9), which is indicative of FADS2 introducing a double bond at the 20-carbon level (Figure 4L). Thus, FAMetA allows the identification of both changes in general patterns and particular synthesis parameters induced by FA metabolism inhibitors. Then we move on to analyze previously published data generated using in vitro (H1299 cells incubated with U-13C-glucose and U-13C-glutamine, where the down-regulation of SREBP cleavage activating protein, a key protein in the regulation of FA metabolism, is induced) [14] (Supplementary Figure 10) and in vivo (incorporation of U-13C-fructose into saponified circulating FAs in wild-type and intestine-specific ketohexokinase (KHK-C) knockout mice after drinking normal water for 8 weeks, or 5 or $10\%$ sucrose water) [26] (Supplementary Figure 11) experimental models. FAMetA properly fits the experimental FA distributions and calculates synthesis parameters for the complete array of detected FAs, and in both cases, the more detailed characterization of FA metabolism provided by FAMetA enables to decipher biological insights that were overlooked by the authors of the studies using previously available tools (please see Supplementary Results for a detailed description of the analysis of previously published data using FAMetA, including an in-depth comparison between FAMetA and FASA using the H1299 cell dataset [14]). ## FAMetA enables the identification of unknown FAs in biological samples The analysis of total FAs in the non-small cell lung cancer (NSCLC) cell line A549 reveals high FA diversity (62 species), including several FAs [33] that do not match available standards (Figure 5A–B). We hypothesize that the information provided by the retention time of each FA combined with the FAMetA analysis of the MS-data generated using U-13C-glucose and well-characterized inhibitors (i.e. FASNi, SCDi and FADS2i) would provide a valuable strategy to identify unknown and unexpected FAs by the reconstruction of their metabolic synthesis route. All the detected unknown FAs incorporate 13C from U-13C-glucose, which confirms their endogenous metabolic origin. In all the cases the information provided by the inhibition profile and the retention time allowed us to propose identities for them all (Figure 5C–G, Supplementary Figure 12A–AC). For example, we detect and calculate synthesis parameters for five FA(18:2) (18:2n6, nv, nx, ny, nz). Based on the decision tree depicted in Figure 6, which guides the identification of each double bond position based on the inhibition profile, we identified them as FA(18:2n7)(Δ6,11), FA(18:2n7)(Δ8,11), FA(18:2n9)(Δ6,9) and FA(18:2n10)(Δ5,8), respectively (Figure 5D–G) (please see Supplementary Results for a detailed description of the rationale behind the identification of 18:2 FAs in A549). Of the identities proposed based on the metabolic reconstruction of the biosynthesis route, 11 are confirmed with commercially available standards (Supplementary Figure 12AD–AJ), and 9 of them do not match previously described FAs (Figure 7). Thus, FAMetA and our proposed strategy disclose a more comprehensive FA biosynthetic landscape of A4594 cells, including the description of novel FAs (Figure 7). Figure 5:* Elucidation of the synthesis route of unidentified FA species by combining FAMetA and FA metabolism inhibitors. Analysis of alterations in the FA metabolic network in the human NSCL cell line A549 incubated for 72 h with U-13C-glucose induced by FASN inhibitor GSK2194069, SCD inhibitor A93572, and FADS2 inhibitor SC26196. (A–B) Chromatographic separation of the saponified FAs from the A549 cells in culture. (A) Combined chromatogram showing all the detected FAs. (B) Individual chromatograms for each detected FA. (C) Heatmap showing the mean value of the log2 fold-of-change (versus untreated) for each detected FA in the following parameters: endogenously synthesized fraction, calculated S, E1, E2, E3 and E4. For each FA, the parameter reported for the endogenous synthesis is indicated. The shadowed cells indicate the activities (DNS, SCD or FADS2) involved in the synthesis of a particular FA. Red denotes the FAs whose synthesis route is unknown. On the heatmap, crosses indicate missing or NA values. (D–G) The mass isotopologue distribution, the mean value of the log2 fold-of-change (versus untreated) in the synthesis parameters and the proposed synthesis route for FAs FA(18:2nv) (D), FA(18:2nx) (E), FA(18:2ny) (F) and FA(18:2nz) (G) whose identities do not match any standard employed for the method development. In all cases, n = 3. Individual points are shown for the mass isotopologue distributions. The mean values are reported elsewhere. In the synthesis route description, horizontal transitions denote elongations (enzymes not indicated) and vertical transitions depict desaturations (enzymes indicated). Figure 6: The algorithm employed to identify unknown FAs by the reconstruction of their biosynthesis route. The depicted algorithm is applied to identify the double bond positions for FAs based on the inhibition profile obtained upon incubation with U-13C-glucose, either with or without SCDi or FADS2i. The algorithm applies to FAs whose origin can be tracked to FA(14:0)/FA(16:0). The previous assumptions must be met: (i) the FA incorporates labeling and intensity suffices to obtain values for all/most expected isotopomers; (ii) FASNi decreases parameter S or distribution is consistent with the origin being FA(14:0)/FA(16:0). Based on the chromatographic profile, we expect the FAs to elute by increasing n-series [i.e. RT(n5 series) ≤ RT(n7 series) ≤ RT(n9 series), etc.]. The algorithm allows identifying the initial FA for the FA synthesis routes described in Figure 7; thus, the actual position of the double bonds has to be extrapolated for the FAs of a different carbon length to that indicated in the algorithm. In red, FAs for which we can anticipate the identification and synthesis route based on the described strategy, but were not detected or unambiguously assigned experimentally in the A549 cells because FADS1 inhibitors were lacking. Figure 7: FA biosynthesis routes in the NSCLC cell line A549. Summary of the FA metabolism network in the A549 cells for those FAs that come from DNL. Black arrows denote elongations, blue arrows denote desaturations (the responsible enzyme is indicated) and a red arrow denotes degradation. Red depicts the FAs that have not been previously described. ## Discussion The therapeutic inhibition of specific FA metabolic enzymes/transporters has been proposed in diseases like cancer [21, 23, 27–29], non-alcoholic fatty liver disease [30], autoimmunity [31] or viral infection [32]. Metabolic plasticity in FA desaturation has been recently acknowledged as a relevant phenomenon that supports lipid biosynthesis [25, 33] and confers a metabolic advantage upon SCD inhibition in cancer cells [25]. The expression of particular elongases (e.g. ELOVL2 in glioma [34] or ELOVL5 in prostate cancer [35]) supports cell growth, tumor initiation and metastasis. Despite the wide variety of FAs, their biosynthetic routes and proven functions, current state-of-the-art tools/algorithms do not provide a comprehensive characterization of FA metabolism. The most commonly used algorithm (i.e. ISA) was initially developed for the determination of DNL for FA(14:0) and FA(16:0) [11, 12]. Further developments enabled the estimation of elongations [13, 14] and of the de novo synthesis of odd-chain FAs [36]. Additionally, a simple strategy for the estimation of the desaturation of FA(18:1n9) based on the ratio of the total labeling of FA(18:0) and FA(18:1n9) has been also proposed [15, 19]. The shown relevance of long FAs, the importance of desaturation in cell biology and in the physiopathology of many diseases and the lack of a tool that performed a comprehensive characterization of all the biosynthetic reactions within FA metabolism in a user-friendly platform accessible to the broad lipid metabolism community have motivated us to develop FAMetA. Our results demonstrate that FAMetA deciphers both patterns of global changes and detailed information about alterations in the synthesis route of FAs of interest both in vitro and in vivo (Figure 4, Supplementary Figures 10–11). The use of U-13C-glucose and well-characterized inhibitors of FA metabolism enzymes (i.e. FASN, SCD1 and FADS2), combined with FAMetA data analysis, enables the comprehensive characterization of the FA biosynthetic network in A549 cells (Figure 5, Supplementary Figure 12). Strikingly, it also discloses the identity of 12 novel FAs that belong to already described n-series, which extends the known FA biosynthesis network compared to previous tools (Figure 7). The lack of well-characterized inhibitors of FADS1 or elongases (ELOVL1-7) limits the level of detail that can be achieved when identifying FAs by their metabolic reconstruction. Likely, some detected FAs, which are identified as the product of double desaturation introduced by the consecutive action of SCD1 and FADS2, are instead a mixture in which the products of a double desaturation introduced by SCD1 and FADS1 are also present. So the unambiguous identification of the proposed unknown/novel FAs would require using complementary analytical tools and, if possible, authentic chemical standards. Nevertheless, we demonstrate that FAMetA enables the straightforward mapping of FA biosynthetic pathways by the techniques and reagents routinely used in metabolism studies. Compared to previous tools FAMetA offers (Supplementary Table 1): (i) the characterization of a broader FA biosynthesis network as it includes in a single tool DNL, elongation and desaturation; (ii) the possibility of running the required steps from data preprocessing to analysis of FA metabolism and graphical representation in a single tool; (iii) a user-friendly environment thanks to its implementation as an R package and a web-based app; (iv) better fitting to the experimental data thanks to the implementation of quasi-multinomial fitting that incudes the parameter Φ that accounts for data overdispersion; (v) better description of elongations, thus enabling an easier interpretation of the estimated parameters; (vi) easy-to-interpret parameters and graphical representations that lead to obtain meaningful biological conclusions. Future developments of mass isotopologue data analysis tools, including FAMetA, should address some unresolved issues like the use of labeled-FAs as nutrients, distinguishing the uptake of exogenous FA and the lipolysis of stored lipids, estimating the synthesis rate of the FAs that result from the degradation of a longer FA [e.g. FA(16:1n9), where S′ = SE1degradation], or the resolution of the FA metabolism properties of particular lipid classes of interest or organelles. Additionally, the FAMetA algorithm is exclusively designed to fit the data from 13C-based tracers for even-chain FAs. Thus, future efforts should focus on implementing calculations based on 2H-tracers, such as 2H2O, which contributes to FA synthesis via direct H2O incorporation, and also via NADPH [7, 8], and to expand the reactions to cover odd-chain FAs, in which not only the lipogenic Acetyl-CoA has to be estimated, but also the lipogenic Propionyl-CoA pool [36]. The calculations based on 2H-tracers can be performed using the code by Zhang et al. [ 7], the calculations for odd-chain FAs can be performed using the code by Crown et al. [ 36] and the equations to model FA degradation or the use of labeled-FAs as nutrients could be theoretically implemented within software designed for the 13C-metabolic flux analysis such as METRAN [37], INCA [38], or 13CFLUX2 [39]. Additionally, recent developments in the field of proteomics such as the use of 12C-nutrients as light isotopic tracers [40], or the resolution of the isotope incorporation based on the number of labeling sites and the label enrichment using numerical techniques [41] could be implemented to estimate FA metabolism using isotopic tracers. Finally, the analysis of FA metabolism at a compartmental, lipid class, or single lipid level would require the use of complex fractionation of lipid extracts or the developments of new tools to deal with the complex distributions that arise from the labeling of each structural component of a complex lipid. Despite these limitations, FAMetA constitutes the first tool that enables reliable estimations of FA import, synthesis, elongation and desaturation for the whole FA metabolic network of FAs within the range from 14 to 26 carbons. The FAMetA workflow includes the required functionalities (data preprocessing, FA metabolism analysis, group-based comparisons and graphical representation) to run a complete data analysis on a single platform (Figure 1). Its combination with the systematic genetic manipulation of enzymes/transporters involved in FA metabolism can contribute to the characterization of FA metabolism in unprecedented detail. Finally, to spread its use, FAMetA is freely available as an open-source R package and a web-based application (www.fameta.es). In conclusion, we believe that FAMetA is a valuable addition to existing tools and has the potential to become a key resource to study the complex FA biosynthetic landscape. ## FAMetA implementation FAMetA was developed in an R programming environment. It is available via CRAN (https://CRAN.R-project.org/package=FAMetA). In addition, the web-based implementation of FAMetA was built using the Shiny R package (Shiny: Web Application Framework for R. 2021). It is accessible at www.fameta.es. ## The FAMetA workflow The FAMetA workflow starts with raw MS data files in the mzXML format, which can be obtained with any MS file converter, e.g. msConvert from ProteoWizard [42], and a csv file containing the required metadata (sample name, acquisition mode, sample group or class, and any additional information like external measures for normalization) (Supplementary Figure 2, steps 1–2). Data preprocessing can be performed in the R environment/web-based application using our proposed workflow, which combines functions from FAMetA and our previously described R-package LipidMS [43, 44] (available via CRAN (https://CRAN.R-project.org/package=LipidMS)) (Supplementary Figure 2, steps 2–5). LipidMS is called for the first preprocessing step, which runs peak-picking, alignment and grouping through functions batchdataProcessing, alignmsbatch and groupmsbatch (Supplementary Figure 2, step 2). Then FAMetA is called, and functions annotateFA and curateFAannotations are used to identify any unique FA isomers. Automatic FA annotations can be exported to a csv file and be modified by removing rows of unwanted FA by modifying the initial and end retention times, or by adding new rows with missing compounds. Unique compound names with nomenclature ‘FA(16:1)n7’, where n7 (omega-7) indicates the last double-bond position, are required to differentiate FA isomers. For any unknown positions, letters x, y and z are allowed (i.e. FA(16:1)nx). The internal standards for later normalization can also be added in a new row at this point by indicating IS in the compound name column (Supplementary Figure 2, step 3). Once FAs have been correctly identified, FA isotopes can be extracted using function searchFAisotopes (Supplementary Figure 2, step 4). Finally, data can be corrected and normalized using the dataCorrection function, which runs four different steps (all of which are optional): data correction for natural 13C abundance using the accucor algorithm [45]; data normalization with internal standards; blank subtraction; external normalization (Supplementary Figure 2, step 5). Alternatively, the external data processed by other available software/tools (e.g. data preprocessing using tools as ElMa) can be loaded at this workflow point or before the data correction and normalization steps. Then the actual FA metabolism analysis can be performed by sequentially running the synthesisAnalysis, elongationAnalysis and desaturationAnalysis functions (Supplementary Figure 3, steps 1–3). The first two functions model isotopologue distributions by non-linear regression (https://CRAN.R-project.org/package=minpack.lm) with many initial values [46] to ensure that the best fits are found. By default, a maximum of 1,000 iterations for synthesis and 10,000 for elongation are performed for each set of initial values to fit the isotopologue distributions (maxiter parameter) or until the model has converged 100 times (maxconvergence parameter). If no results are obtained or parameters come close to the limits of the confidence intervals, these parameters can be increased to improve the results. The third function employs the previous results to estimate the desaturation values. Finally, the summarized results tables and heatmaps are obtained using the summarizeResults function to export and explore the results (Supplementary Figure 3, step 4). ## Model assumptions (i) The acetyl-CoA pool contributing to lipogenesis has a uniform labeling pattern. ( ii) The lipogenic acetyl-CoA pool reaches isotopic steady state quickly compared with the total labeling time. ( iii) For FAs of 16 or more carbons the final product of FASN is FA(16:0). ( iv) For the FAs belonging to the n3 and n6 series, $S = 0.$ ( v) At any given time point I = import + pre-existing FAs, and only when the pre-existing FAs have been completely replaced (the actual steady state has been achieved) I = import. ( vi) There is a single FA pool. ( vii) The data have been corrected to account for the natural abundance of the 13C isotopes. ## Data requirements for FA modeling Before FA metabolism analysis, the user should check that the FAs of interest have been labeled enough to obtain isotopologue distributions of good quality (avoid missing isotopologues) that guarantee the calculated parameters fall within the ranges that allow their accurate estimation. When curating FA annotations, FA names must follow the nomenclature FA(C:d)ns, where C is the total number of carbon, d is the number of unsaturations and ns refers to the omega series, which indicates the position of the last double bound starting from the end of the chain. Duplicated identities are not allowed and the series must belong either to known series [i.e. 3, 5, 6, 7, 7a (i.e. second double bond introduced by FADS2 at 16C), 7b (i.e. second double bond introduced by FADS2 at 18C), 9, 10, 12, 13], or use the letters x, y and z for an unknown series. For the estimation of synthesis parameters, data must have been corrected to account for the natural abundance of the 13C isotopes. ## Estimation of the DNL parameters We considered FA(16:0) the final DNL product. Thus, FAMetA can estimate the DNL parameters for FAs up to 16 carbons. For these species, I and S represent the fraction of the FA pool that is imported and synthesized, respectively, and sums 1:[1]\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}\begin{equation} {I}_{16:0}+{S}_{16:0}=1 \end{equation}\end{document} For the DNL analysis, FA isotopologue distributions (previously corrected for the natural abundance of the 13C isotopes) are modeled with the following sum of the weighted quasi-multinomial distributions adapted from [47]:[2]\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}\begin{equation} P\left($m = 0$\right)=I+S\ast \left(1+N\ast \varPhi \right)\ast \frac{\ {D}_0}{1+N\ast \varPhi}\ast{\left(\frac{D_0+N\ast \varPhi }{1+N\ast \varPhi}\right)}^{N-1} \end{equation}\end{document}[3]\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}\begin{equation} P(m)=\sum_{$j = 1$}^kP\left({X}_0={x}_{0,j},{X}_1={x}_{1,j},{X}_2={x}_{2,j}\right);\mathrm{for}\ 1\le m\le M \end{equation}\end{document}where[4]\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}\begin{align} &P\left({X}_0={x}_{0,j},{X}_1={x}_{1,j},{X}_2={x}_{2,j}\right)\nonumber\\\nonumber&=S\ast \frac{N!}{x_{0,j}!{x}_{1,j}!{x}_{2,j}!}\ast \left(1+N\ast \varPhi \right)\ast \frac{D_0}{1+N\ast \varPhi}\ast{\left(\frac{D_0+{x}_{0,j}\ast \varPhi }{1+N\ast \varPhi}\right)}^{x_{0,j}-1}\\ &\ast \frac{D_1}{1+N\ast \varPhi}\ast{\left(\frac{D_1+{x}_{1,j}\ast \varPhi }{1+N\ast \varPhi}\right)}^{x_{1,j}-1}\!\ast \frac{D_2}{1+N\ast \varPhi}\ast{\left(\frac{D_2+{x}_{2,j}\ast \varPhi }{1+N\ast \varPhi}\right)}^{x_{2,j}-1} \end{align}\end{document}given that[5]\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}\begin{equation} {\displaystyle \begin{array}{c}{x}_{i,j}=0,1,\dots, N\\{}\sum_{$i = 1$}^2{x}_{i,j}={x}_{0,j}+{x}_{1,j}+{x}_{2,j}=N\end{array}} \end{equation}\end{document}[6]\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}\begin{equation} \sum_{$i = 1$}^2i\ast{x}_{i,j}=0\ast{x}_{0,j}+1\ast{x}_{1,j}+2\ast{x}_{2,j}=m \end{equation}\end{document}[7]\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}\begin{equation} 0\le \varPhi \le \frac{1-\max \left({D}_0,{D}_1,{D}_2\right)}{N} \end{equation}\end{document} M is the total number of carbons in the FA molecule and N = M/2. This represents the number of acetyl-CoA molecules used for the synthesis of an FA of length M. m is the number of 13C atoms incorporated into the FA molecule. D0, D1 and D2 represent the fraction of acetyl-CoA with 0, 1, or 2 atoms of 13C, respectively, and sum 1. x0, x1 and x2 represent the number of acetyl-CoA units with 0, 1, or 2 13C atoms that provide an M-carbon FA with an m label. For a given pair of N and m values, up to k combinations of the x0, x1 and x2 values fulfill equations [5] and [6]. Φ accounts for overdispersion and can be set at 0 to reduce quasi-multinomial distribution to multinomial distribution. The in silico validation of the above-described equations demonstrates an overestimation of Φ and an underestimation of S and D2 for values of D2 ≥ 0.75. In these situations, the upper limit of Φ is set at 0.5(1 − max(D0, D1, D2)/N). Note that overdispersion parameter Φ modifies D0, D1 and D2 for each synthesis step, which allows distribution to widen or narrow. Based on this model, non-linear regression (https://CRAN.R-project.org/package=minpack.lm) with many sets of plausible initial values (adapted from ref [46]) is used to fit the observed isotopologue distributions of FAs up to 16 carbons, and to estimate parameters D1, D2, Φ and S. When analyzing multiple samples per group, S and D2 values can be checked to ensure homogeneity within each group. If not, we can assume D2 should remain within a narrow range for a given condition and thus fix D2 by the mean of the rest of the samples in the group for the outlier sample and repeat the analysis to improve the calculation of the S value. To improve the analysis results, the D1, D2 and Φ values obtained for FAs up to 16C are used to model the distribution of FAs of 18-to-26C. ## Elongation The main product of the DNL of FA is FA(16:0) [1]. Therefore, the main DNL route, plus elongation, starts at 16 carbons and then adds blocks of two carbons. Elongation from FA(14:0) is a minor route [14] and is omitted for simplicity. For the FAs ranging from 18 to 26 carbons, the following equations are considered:[8]\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}\begin{equation} {I}_{18:0}+{E}_1\left({I}_{16:0}+{S}_{16:0}\right)={I}_{18:0}+{E}_1=1 \end{equation}\end{document}[9]\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}\begin{equation} {I}_{20:0}+{E}_2\ast \left({I}_{18:0}+{E}_1\ast \left({I}_{16:0}+{S}_{16:0}\right)\right)={I}_{20:0}+{E}_2=1 \end{equation}\end{document}[10]\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}\begin{equation} {I}_{22:0}+{E}_3\ast \left({I}_{20:0}+{E}_2\ast \left({I}_{18:0}+{E}_1\ast \left({I}_{16:0}+{S}_{16:0}\right)\right)\right)={I}_{22:0}+{E}_3=1 \end{equation}\end{document}[11]\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}\begin{align} {I}_{24:0}+{E}_4 &\ast \left({I}_{22:0}+{E}_3\ast \left({I}_{20:0}+{E}_2\ast \left({I}_{18:0}+{E}_1\ast \left({I}_{16:0}+{S}_{16:0}\right)\right)\right)\right)\nonumber\\ &={I}_{24:0}+{E}_4=1 \end{align}\end{document}[12]\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}\begin{align} {I}_{26:0}+{E}_5 &\ast ({I}_{24:0}+{E}_4\ast ({I}_{22:0}+{E}_3\ast ({I}_{20:0}+{E}_2\ast ({I}_{18:0}+{E}_1 \nonumber\\ &\ast ({I}_{16:0}+{S}_{16:0})))))={I}_{26:0}+{E}_5=1 \end{align}\end{document} For the elongation analysis of endogenous FA, isotopologue distributions are modeled using equation [2] for synthesis until FA(16:0), followed by single independent elongation steps (E1, E2 …, En). The probability of incorporating 0, 1, or 2 13C atoms into the FA to be elongated equals EiD0, EiD1 and EiD2, respectively. For FA longer than 16C, only synthesis and elongation terms are estimated (S, E1, E2 …, En), whereas the rest (D0, D1, D2 and Φ) are inherited from the results obtained for the FA(16:0). In case no results are available for FA(16:0), FAMetA uses FA(14:0), mean of all FA of 16C (FA(16:X)), or mean of all FA of 14C (FA(14:X)) in this order of priority. For FA(18:0), FA isotopologue distributions (previously corrected for natural 13C isotopes abundance) are modeled with the following equations:[13]\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}\begin{equation} {P}_{18:0}\left($m = 0$\right)={I}_{18:0}+{E}_1\ast{D}_0\ast{P}_{16:0}\left($m = 0$\right) \end{equation}\end{document}[14]\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}\begin{equation} {P}_{18:0}\left($m = 1$\right)={E}_1\ast{D}_0\ast{P}_{16:0}\left($m = 1$\right)+{E}_1\ast{D}_1\ast{P}_{16:0}\left($m = 0$\right) \end{equation}\end{document}[15]\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}\begin{equation} {\displaystyle \begin{array}{c}{P}_{18:0}(m)={E}_1\ast{D}_0\ast{P}_{16:0}\left(m=m\right)+{E}_1\ast{D}_1\ast{P}_{16:0}\left(m=m-1\right)+{E}_1\\\ast{D}_2\ast{P}_{16:0}\left(m=m-2\right) {}\mathrm{for}\ 2\le m\le M;{P}_{16:0}\left(m>16\right)=0\end{array}} \end{equation}\end{document} Analogous equations can be obtained for FA with M > 18 by adding elongation terms to previously existing distributions. For series n6 and n3 (Supplementary Figure 1), elongation is usually expected from FA(18:2)n6 and FA(18:3)n3. Thus, synthesis (S) and the first elongation step (E1) are set at 0. If isotopologue M + 2 is observed, given the degradation of FA(18:2)n6 or FA(18:3)n3, followed by one elongation step, then E1 is estimated. However, the endogenously synthesized fraction remains at NA. In addition, isotopologue distributions of FA longer than 16C are checked to decide if any parameter can be fixed to 0 (for those parameters selected based on the omega series). At least two or three even isotopologues (M + 2, M + 4, M + 6, …), with a relative intensity greater than 0.1 or $0.01\%$, respectively, along the whole distribution, are required to estimate S. Similarly, for elongation terms, specific isotopologues are checked to ensure how many elongation steps have occurred (M + x > $0.1\%$). Once again, non-linear regression (https://CRAN.R-project.org/package=minpack.lm) with multiple initial values [46] is used to fit the observed isotopologue distributions of the elongated FAs. ## Desaturation After estimating the synthesis and elongation parameters, these results can be used to calculate the FA fraction that comes from desaturation in the unsaturated FA. For a given unsaturated FA (e.g. FA(18:1n9)), we can conceptually consider a one-step elongation-desaturation reaction (in this example, directly from FA(16:0) to FA(18:1n9)), or a two-step elongation followed by a desaturation process (in this example, FA(16:0) is elongated to FA(18:0) and then desaturated to FA(18:1n9)) (Figure 3). By using FAMetA, we can directly estimate both E1 and E1’ from the isotopologue distributions of FA(18:0) and FA(18:1n9), respectively. From alternative paths, the relative import and endogenous synthesis pathways of FA(18:1n9) can be written as[16]\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}\begin{equation} {I_{18:1n9}}^{\prime }+{E_1}^{\prime}\ast \left({S}_{16:0}+{I}_{16:0}\right)=1 \end{equation}\end{document}[17]\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}\begin{equation} {I}_{18:1n9}+\varDelta \ast{E}_1\ast \left({S}_{16:0}+{I}_{16:0}\right)+\varDelta \ast{I}_{18:0}=1 \end{equation}\end{document} By combining both equations, we can define that[18]\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}\begin{equation} {I_{18:1n9}}^{\prime }={I}_{18:0}\ast \varDelta +{I}_{18:1n9} \end{equation}\end{document}and, thus, calculate desaturation parameter Δ as[19]\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}\begin{equation} \varDelta =\frac{{E_1}^{\prime }}{E_1} \end{equation}\end{document} If both Ei’ and Ei are below the confidence interval, which is set at 0.05, by default, for desaturation, parameter Δ is not calculated, and Ei’ remains as the endogenously synthesized fraction. If the stationary state is not reached, values >1 can be obtained for the desaturation parameter, that is, in this case, replaced with 1. This same model can be used for all the known desaturation steps, provided that the precursor and product FA isomers are correctly and uniquely identified, and the stationary state is reached. For the FA synthesized from desaturation activities, Δ is considered the fraction from endogenous synthesis. So the imported fraction is calculated as 1 − Δ. With unknown isomers or missing precursors, S′ or E’ is returned for the DNS of FAs until 16 carbons or the elongation of longer FAs, respectively. The reactions included in FAMetA are described in Supplementary Figure 1 [14, 16, 48, 49]. However, additional reactions (desaturations) can be included for unknown/additional FAs by modifying desaturationdb in FAMetA. ## In silico tests of FAMetA To test FAMetA’s performance with different FA isotopologue distributions and noise levels, in silico tests on models are run. To evaluate FAMetA’s performance to estimate parameters for the DNS analysis, realistic values for D1 (5 values from 0 to 0.2), D2 (15 values from 0 to 1), Φ (10 values from 0 to 0.1) and S (15 values from 0 to 1) are combined to simulate 3,945 theoretical FA(16:0) distributions to which 0, 2, 5 and $10\%$ noise levels are added to obtain 10 different noised distributions for each set of parameters. Bias (evaluated as an absolute or relative error) and dispersion (evaluated as RSD) are calculated and graphically represented for parameters D2, S and Φ (Supplementary Figure 4). To evaluate FAMetA’s performance to estimate the parameters for the elongation analysis, the mass isotopologue distributions for FA(18:0), FA(20:0), FA(22:0) and FA(24:0) are generated. To evaluate the elongation of FA(16:0) to FA(18:0), D1 and Φ are set at 0.05 and 0.01, respectively. The realistic values for D2 (9 values between 0.1 and 0.9), S (19 values between 0.05 and 1 and E1 (19 values between 0.05 and 1) are employed to generate 3,249 theoretical FA(18:0) distributions. For FA(20:0), FA(22:0) and FA(24:0), the synthesis parameters for FA(16:0) are set at D1 = 0.05, Φ = 0.01 and $S = 0.6.$ Nine values within the 0.1–0.9 range and 10 values within the 0.1–1 range are generated for D2 and En, respectively. Bias (evaluated as a relative error) and dispersion (evaluated as RSD) are calculated and graphically represented for all the estimated parameters (Supplementary Figures 5–8). To evaluate FAMetA’s performance to estimate the parameters for the desaturation analysis, the mass isotopologue distributions for FA(16:1n7) and FA(18:1n9) are generated. D1 and Φ are set at 0.05 and 0.01, respectively. For FA(16:1n7), 13 values within the 0.1–0.87 range and 14 values within the 0.07–1 range are generated for D2 and S, respectively. For FA(18:1n8), 13 values within the 0.1–0.87 range and 14 values within the 0.07–1 range are generated for D2 and E1, respectively. In both cases, 14 values within the 0.07–1 range are generated for Δ. Bias (evaluated as a relative error) and dispersion (evaluated as RSD) are calculated and graphically represented for parameter Δ for both FAs (Supplementary Figure 9). ## Reagents, biological sources and experimental details Detailed description of reagents, cell isolation and culture, animal models, cell lines and methods to extract and analyze FAs is provided in the Supplementary Information. ## Funding Carlos III Health Institute of the Spanish Ministry of Economy and Competitiveness (FI$\frac{18}{00224}$ to M.I.A.-B.); Conselleria de Sanidad Universal y Salud Pública, Generalitat Valenciana, as part of Plan GenT, Generació Talent (DEI-$\frac{01}{20}$-C to J.C.G.-C.); European Regional Development Fund (FEDER) and the Carlos III Health Institute of the Spanish Ministry of Economy and Competitiveness (PI$\frac{20}{00580}$ and DTS$\frac{2019}{0143}$ to A.L.); Generalitat Valenciana and European Regional Development Fund (FEDER) funds (PO FEDER of Comunitat Valenciana 2014-2020). ## Data and materials availability The input mzXML LC–MS data files used to generate Figures 2–4, the metadata associated with the studies, the FA identities for each dataset, the preprocessed data ready to be used to estimate the FA metabolism parameters for each study and an R script are available at Zenodo with accession number 6511248. FAMetA’s source code is offered to the public as a freely accessible software package under the GNU GPL license, version 3. It is available at https://github.com/maialba3/FAMetA and Zenodo with accession number 6511248. FAMetA R package available at CRAN (https://CRAN.R-project.org/package=FAMetA) will be maintained at least for the following 5 years. ## Author contributions J.C.G.-C. and A.L. conceived the method. A.L. supervised study development. M.I.A.-B. designed and programmed the algorithms for FAMetA and its online implementation. M.I.A.-B. and J.C.G.-C. optimized the LC–MS-based method for the FA analysis. M.I.A.B. and J.C.G.-C. performed the FA analyses. J.C.G.-C. and M.B. performed the labeling experiments. M.I.A.-B., J.C.G.-C., M.B., O.J. and A.L. analyzed the data and discussed the results. J.C.G.-C. and A.L. wrote the manuscript in collaboration with all the authors. ## References 1. Batchuluun B, Pinkosky SL, Steinberg GR. Lipogenesis inhibitors: therapeutic opportunities and challenges. Nat Rev Drug Discov* (2022) 21 283-305. PMID: 35031766 2. Klop B, Elte JWF, Cabezas MC. Dyslipidemia in obesity: mechanisms and potential targets. Nutrients (2013) 5 1218-40. PMID: 23584084 3. Taskinen MR, Borén J. New insights into the pathophysiology of dyslipidemia in type 2 diabetes. Atherosclerosis (2015) 239 483-95. PMID: 25706066 4. Ipsen DH, Lykkesfeldt J, Tveden-Nyborg P. Molecular mechanisms of hepatic lipid accumulation in non-alcoholic fatty liver disease. Cell Mol Life Sci (2018) 75 3313-27. PMID: 29936596 5. Broadfield LA, Pane AA, Talebi A. Lipid metabolism in cancer: new perspectives and emerging mechanisms. Dev Cell (2021) 56 1363-93. PMID: 33945792 6. Pietrocola F, Galluzzi L, Bravo-San Pedro JM. Acetyl coenzyme A: a central metabolite and second messenger. Cell Metab (2015) 21 805-21. PMID: 26039447 7. Zhang Z, Chen L, Liu L. Chemical basis for deuterium labeling of fat and NADPH. J Am Chem Soc (2017) 139 14368-71. PMID: 28911221 8. Fu X, Deja S, Fletcher JA. Measurement of lipogenic flux by deuterium resolved mass spectrometry. Nat Commun (2021) 12 3756. PMID: 34145255 9. Antoniewicz MR. A guide to 13C metabolic flux analysis for the cancer biologist. Exp Mol Med (2018) 50 1-13 10. Hellerstein MK, Neese RA. Mass Isotopomer Distribution Analysis: a technique for measuring biosynthesis and turnover of polymers. Am J Physiol Endocrinol Metab (1992) 263 E988-1001 11. Kelleher JK, Masterson TM. Model equations for condensation biosynthesis using stable isotopes and radioisotopes. Am J Physiol Endocrinol Metab (1992) 262 E118-25 12. Kelleher JK, Nickol GB. Isotopomer spectral analysis: utilizing nonlinear models in isotopic flux studies. Methods Enzymol (2015) 561 13. Tredwell GD, Keun HC. ConvISA: a simple, convoluted method for isotopomer spectral analysis of fatty acids and cholesterol. Metab Eng (2015) 32 125-32. PMID: 26432945 14. Argus JP, Wilks MQ, Zhou QD. Development and application of FASA, a model for quantifying fatty acid metabolism using stable isotope labeling. Cell Rep (2018) 25 2919-2934.e8. PMID: 30517876 15. Kamphorst JJ, Cross JR, Fan J. Hypoxic and Ras-transformed cells support growth by scavenging unsaturated fatty acids from lysophospholipids. Proc Natl Acad Sci USA (2013) 110 8882-7. PMID: 23671091 16. Guillou H, Zadravec D, Martin PGP. The key roles of elongases and desaturases in mammalian fatty acid metabolism: insights from transgenic mice. Prog Lipid Res (2010) 49 186-99. PMID: 20018209 17. Antoniewicz MR, Kelleher JK, Stephanopoulos G. Elementary metabolite units (EMU): a novel framework for modeling isotopic distributions. Metab Eng (2007) 9 68-86. PMID: 17088092 18. Metallo CM, Gameiro PA, Bell EL. Reductive glutamine metabolism by IDH1 mediates lipogenesis under hypoxia. Nature (2012) 481 380-4 19. Tumanov S, Bulusu V, Kamphorst JJ. Analysis of fatty acid metabolism using stable isotope tracers and mass spectrometry. Methods Enzymol (2015) 561 20. Buescher JM, Antoniewicz MR, Boros LG. A roadmap for interpreting13C metabolite labeling patterns from cells. Curr Opin Biotechnol (2015) 34 189-201. PMID: 25731751 21. Hardwicke MA, Rendina AR, Williams SP. A human fatty acid synthase inhibitor binds β-ketoacyl reductase in the keto-substrate site. Nat Chem Biol (2014) 10 774-9. PMID: 25086508 22. Xin Z, Zhao H, Serby MD. Discovery of piperidine-aryl urea-based stearoyl-CoA desaturase 1 inhibitors. Bioorg Med Chem Lett (2008) 18 4298-302. PMID: 18632269 23. Von Roemeling CA, Marlow LA, Wei JJ. Stearoyl-CoA desaturase 1 is a novel molecular therapeutic target for clear cell renal cell carcinoma. Clin Cancer Res (2013) 19 2368-80. PMID: 23633458 24. Obukowicz MG, Raz A, Pyla PD. Identification and characterization of a novel Δ6/Δ5 fatty acid desaturase inhibitor as a potential anti-inflammatory agent. Biochem Pharmacol (1998) 55 1045-58. PMID: 9605428 25. Vriens K, Christen S, Parik S. Evidence for an alternative fatty acid desaturation pathway increasing cancer plasticity. Nature (2019) 566 403-6. PMID: 30728499 26. Jang C, Wada S, Yang S. The small intestine shields the liver from fructose-induced steatosis. Nat Metab (2020) 2 586-93. PMID: 32694791 27. Pascual G, Avgustinova A, Mejetta S. Targeting metastasis-initiating cells through the fatty acid receptor CD36. Nature (2017) 541 41-5. PMID: 27974793 28. Veglia F, Tyurin VA, Blasi M. Fatty acid transport protein 2 reprograms neutrophils in cancer. Nature (2019) 569 73-8. PMID: 30996346 29. Ferraro GB, Ali A, Luengo A. Fatty acid synthesis is required for breast cancer brain metastasis. Nat Cancer (2021) 2 414-28. PMID: 34179825 30. Loomba R, Mohseni R, Lucas KJ. TVB-2640 (FASN inhibitor) for the treatment of nonalcoholic steatohepatitis: FASCINATE-1, a randomized, placebo-controlled phase 2a trial. Gastroenterology (2021) 161 1475-86. PMID: 34310978 31. Berod L, Friedrich C, Nandan A. De novo fatty acid synthesis controls the fate between regulatory T and T helper 17 cells. Nat Med (2014) 20 1327-33. PMID: 25282359 32. Munger J, Bennett BD, Parikh A. Systems-level metabolic flux profiling identifies fatty acid synthesis as a target for antiviral therapy. Nat Biotechnol (2008) 26 1179-86. PMID: 18820684 33. Young RSE, Bowman AP, Williams ED. Apocryphal FADS2 activity promotes fatty acid diversification in cancer. Cell Rep (2021) 34 108738. PMID: 33567271 34. Gimple RC, Kidwell RL, Kim LJY. Glioma stem cell–specific superenhancer promotes polyunsaturated fatty-acid synthesis to support EGFR signaling. Cancer Discov (2019) 9 1248-67. PMID: 31201181 35. Centenera MM, Scott JS, Machiels J. ELOVL5 is a critical and targetable fatty acid elongase in prostate cancer. Cancer Res (2021) 81 1704-18. PMID: 33547161 36. Crown SB, Marze N, Antoniewicz MR. Catabolism of branched chain amino acids contributes significantly to synthesis of odd-chain and even-chain fatty acids in 3T3-L1 adipocytes. PLoS One (2015) 10 1-22 37. Yoo H, Antoniewicz MR, Stephanopoulos G. Quantifying reductive carboxylation flux of glutamine to lipid in a brown adipocyte cell line. J Biol Chem (2008) 283 20621-7. PMID: 18364355 38. Young JD. INCA: a computational platform for isotopically non-stationary metabolic flux analysis. Bioinformatics (2014) 30 1333-5. PMID: 24413674 39. Weitzel M, Nöh K, Dalman T. 13CFLUX2 - high-performance software suite for 13C-metabolic flux analysis. Bioinformatics (2013) 29 143-5. PMID: 23110970 40. Sénécaut N, Alves G, Weisser H. Novel insights into quantitative proteomics from an innovative bottom-up simple light isotope metabolic (bSLIM) Labeling data processing strategy. J Proteome Res (2021) 20 1476-87. PMID: 33573382 41. Sadygov RG. Partial isotope profiles are sufficient for protein turnover analysis using closed-form equations of mass isotopomer dynamics. Anal Chem (2020) 92 14747-53. PMID: 33084301 42. Chambers MC, Maclean B, Burke R. A cross-platform toolkit for mass spectrometry and proteomics. Nat Biotechnol (2012) 30 918-20. PMID: 23051804 43. Alcoriza-Balaguer MI, García-Cañaveras JC, López A. LipidMS: an R package for lipid annotation in untargeted liquid chromatography-data independent acquisition-mass spectrometry lipidomics. Anal Chem (2019) 91 836-45. PMID: 30500173 44. Alcoriza-Balaguer MI, García-Cañaveras JC, Ripoll-Esteve FJ. LipidMS 3.0: an R-package and a web-based tool for LC-MS/MS data processing and lipid annotation. Bioinformatics (2022) 38 4826-8. PMID: 36005855 45. Su X, Lu W, Rabinowitz JD. Metabolite spectral accuracy on orbitraps. Anal Chem (2017) 89 5940-8. PMID: 28471646 46. Padfield D, Yvon-Durocher G, Buckling A. Rapid evolution of metabolic traits explains thermal adaptation in phytoplankton. Ecol Lett (2016) 19 133-42. PMID: 26610058 47. Consul PC, Mittal SP. Some discrete multinomial probability models with predetermined strategy. Biom J (1977) 19 161-73 48. Purdy JG, Shenk T, Rabinowitz JD. Fatty acid elongase 7 catalyzes lipidome remodeling essential for human cytomegalovirus replication. Cell Rep (2015) 10 1375-85. PMID: 25732827 49. Deák F, Anderson RE, Fessler JL. Novel cellular functions of very long chain-fatty acids: insight from ELOVL4 mutations. Front Cell Neurosci (2019) 13
--- title: Clinical Efficacy of Chemotherapy Regimen Combined with Levofloxacin in Patients with Pulmonary Tuberculosis Complicated with Type-2 Diabetes authors: - Pei-pei Luo - Li Wu - Fang Liu - Yan-hong Tian - Lai-yin Chen - Ya-lin Liu journal: Pakistan Journal of Medical Sciences year: 2023 pmcid: PMC10025689 doi: 10.12669/pjms.39.2.6364 license: CC BY 3.0 --- # Clinical Efficacy of Chemotherapy Regimen Combined with Levofloxacin in Patients with Pulmonary Tuberculosis Complicated with Type-2 Diabetes ## Abstract ### Objective: To evaluate the clinical efficacy of a chemotherapy regimen combined with levofloxacin in patients with pulmonary tuberculosis complicated with Type-2 diabetes. ### Methods: Total 80 patients with pulmonary tuberculosis complicated with Type-2 diabetes admitted to Baoding People’s Hospital from January, 2019 to January, 2022 were randomly divided into two groups: the experimental group and the control group, with 40 cases in each group. Patients in the control group were given the conventional 2HRZE/10HRE regimen, while those in the experimental group were given the chemotherapy regimen 2HRZEL/6HRE combined with levofloxacin. Sixty four slice spiral CT was used for chest plain scan before and after treatment, respectively, to evaluate the absorption of lesions based on the range of lung lesions; Venous blood was drawn to detect the changes of oxidative stress indicators, the incidence of adverse drug reactions and the negative conversion rate of sputum tuberculosis bacteria in the two groups. ### Results: After treatment, the efficacy of the experimental group was $90\%$, which was significantly higher than that of the control group ($67.5\%$), with a statistically significant difference ($$p \leq 0.01$$). After treatment, CD3+, CD4+, CD4+/CD8+ and other indicators in the experimental group were significantly higher than those in the control group, with a statistically significant difference (CD3+, $$p \leq 0.01$$; CD4+, $$p \leq 0.01$$; CD4+/CD8+, $$p \leq 0.00$$), while CD8+ did not change significantly ($$p \leq 0.92$$); The incidence of adverse reactions was $52.5\%$ in the experimental group and $47.5\%$ in the control group, with no statistically significant difference ($$p \leq 0.66$$); The negative conversion rate of patients in the experimental group was significantly higher than that in the control group at one month, three months and six months after treatment, with a statistically significant difference ($p \leq 0.05$). ### Conclusion: Chemotherapy combined with levofloxacin is a safe and effective regimen for patients’ pulmonary tuberculosis complicated with Type-2 diabetes, boasting a variety of benefits such as improved clinical efficacy, ameliorated cellular immune status, a high negative conversion rate of sputum tuberculosis bacteria, and no significant increase in adverse reactions. ## INTRODUCTION Tuberculosis (TB) has been coexisting with the development of human history since the Stone Age. It is still one of the chronic infectious diseases that threaten human health and is the top 10 leading causes of human death worldwide.1 TB is more common in developing countries and poverty-stricken areas, especially in immunocompromised patients.2 Studies have shown that3 the incidence of pulmonary TB combined with diabetes is increasing year by year, which has become a global health problem. It was suggested in the study of Deshmukh et al.4 that compared with the normal population, patients with pulmonary TB have a higher probability of suffering from diabetes, while patients with pulmonary TB complicated with diabetes have significantly increased blood and tissue glucose levels due to abnormal glucose metabolism caused by diabetes so that the normal immune function of cells is inhibited, thus accelerating the proliferation of mycobacterium TB and aggravating pulmonary TB. It has been reported in the literature5 that patients with pulmonary TB complicated with diabetes have rapid disease progression, a high positive rate of sputum culture, more cavities, difficult treatment and poor clinical efficacy. To this end, correct anti-TB drugs should be selected and appropriate chemotherapy regimens should be developed to treat pulmonary TB complicated with diabetes. Studies have shown that6 levofloxacin is a broad-spectrum antibacterial drug with a strong antibacterial effect. Levofloxacin can bind to bacterial DNA gyrase subunit A for rapid bactericidal action by inhibiting bacterial DNA replication and DNA helix activity. In this study, a short-course chemotherapy regimen combined with levofloxacin was used for patients with pulmonary TB complicated with Type-2 diabetes, and certain clinical effects were achieved. ## METHODS Eighty patients with pulmonary TB complicated with Type-2 diabetes admitted to Baoding People’s Hospital from January 2019 to January 2022 were randomly divided into two groups: the experimental group and the control group, with 40 cases in each group. Among them, there were 27 males and 13 females in the experimental group, aged 32-68 years, with an average of 52.70±12.18 years, and 25 males and 15 females in the control group, aged 27-70 years, with an average of 51.86±12.84 years. No significant difference was observed in the comparison of general data between the two groups, which was comparable (Table-I). The study was approved by the Institutional Ethics Committee of Baoding people’s Hospital on September 17, 2020(No.:2020020), and written informed consent was obtained from all participants. Table-I \| Indicators \| Experimental group \| Control group \| t/χ2 \| p \| \| --- \| --- \| --- \| --- \| --- \| \| Age (years old) \| 52.70±12.18 \| 51.86±12.84 \| 0.3 \| 0.76 \| \| Male (cases %) \| 27 (67.5%) \| 25 (62.5%) \| 0.22 \| 0.64 \| \| Course of diabetes (years) \| 7.12±2.33 \| 7.31±2.74 \| 0.33 \| 0.74 \| \| Accompanied symptoms \| \| \| \| \| \| Cough \| 24 (60%) \| 22 (55%) \| 0.2 \| 0.65 \| \| Chest pain \| 6 (15%) \| 7 (17.5%) \| 0.09 \| 0.76 \| \| Fever \| 23 (57.5%) \| 25 (62.5%) \| 0.21 \| 0.65 \| \| Other \| 9 (22.5%) \| 7 (17.5%) \| 0.31 \| 0.58 \| \| Smoking history (%) \| 24 (60%) \| 27 (67.5%) \| 0.49 \| 0.49 \| ## Inclusion criteria: Patients who met the diagnostic criteria for TB7 and Type-2 diabetes8 and whose chest imaging examinations (CT, X-ray) could accurately calculate the size of lung lesions;Patients under the age of 70;Patients who have not received regular anti-TB treatment in the past;Patients without serious cardiovascular and cerebrovascular diseases, liver and kidney disease history, and without extrapulmonary TB;Patients without HIV/AIDS;Patients without obvious disturbance of consciousness and able to cooperate with the completion of the study;Patients who have not used drugs that affect the study recently, such as immunosuppressants and hormone drugs;Patients who are not allergic to the drugs involved in the study;Patients who signed the consent form by themselves and their family members and were able to cooperate with the study. ## Exclusion Criteria: Children and pregnant women;Patients with drug-resistant TB;Patients who cannot tolerate the drugs used in the study protocol for various reasons;Patients with long QT interval (>480 ms);Patients with mental or consciousness disorders, cognitive disorders, and unable to cooperate with the study;Patients with allergies, intolerance or contraindications to the relevant drugs involved in the study. Patients in the control group were given the conventional 2HRZE/10HRE regimen: oral isoniazid tablets 0.3g, qd; rifampicin 0.45g, qd; pyrazinamide 1.5g, qd; ethambutol hydrochloride Tablet 0.75g, qd. After two months of treatment, patients were switched to the HRE regimen for ten months,9 with the drug application method and dose remaining unchanged. Patients in the experimental group were given the chemotherapy regimen combined with levofloxacin: 2HRZEL/6HRE, isoniazid 0.3g daily at a draught; rifampicin 0.45g daily at a draught; pyrazinamide 0.5g daily, tid; ethambutol 0.75 g daily at a draught; levofloxacin 0.8g daily, orally.10 All patients underwent sputum culture + drug sensitivity test before treatment. After two months of treatment, patients were switched to the HRE regimen for six months, with the drug application method and dose as above. ## Evaluation of clinical efficacy: Siemens 64-slice spiral CT was used for chest plain scan before and 3 months after treatment to evaluate the absorption of lesions based on the range of lung lesions.11 Remarkably effective: ≥$50\%$ reduction in the range of lesions compared with before treatment; Effective: $20\%$-$50\%$ reduction in the range of lesions compared with before treatment; Invalid: ≤$20\%$ reduction in the range of lesions compared with before treatment; Deterioration: Enlarged or disseminated range of lesions. Total effective rate = (remarkably effective + effective)/total number of cases; ## Analysis of immune status: Fasting blood was taken in the morning before and after treatment, respectively, to detect the levels of T lymphocyte subsets CD3+, CD4+, CD8+, CD4+/CD8+, and the differences between the two groups before and after treatment were compared ## Assessment of adverse drug reactions: Adverse drug reactions of the two groups after treatment were recorded, including: abnormal liver function, neuritis, abnormal renal function, rash, leukopenia, gastrointestinal reactions, etc. ## Comparative analysis of the negative conversion rate of sputum mycobacterium TB: After treatment, patients were examined for sputum bacteria every month. If the examination for more than two consecutive months shows that sputum bacteria turn negative and there is no recurrence, it is regarded as successful negative conversion. ## Statistical Analysis: All data in this study were analyzed with SPSS 20.0 software, and measurement data were expressed as (±S). Data between the experimental group and the control group were analyzed using two independent t test. Paired t test was utilized for the comparative analysis of each indicator in the experimental group before and after treatment. $P \leq 0.05$ indicates a statistically significant difference. ## RESULTS The efficacy of the experimental group was $90\%$, which was significantly higher than that of the control group ($67.5\%$), with a statistically significant difference ($$p \leq 0.01$$, see Table-II). No significant difference was observed in the levels of CD3+, CD4+, CD8+ and CD4+/CD8+ before treatment of the two groups ($p \leq 0.05$). After treatment, CD3+, CD4+, CD4+/CD8+ and other indicators in the experimental group were significantly higher than those in the control group, with a statistically significant difference (CD3+, $$p \leq 0.01$$; CD4+, $$p \leq 0.01$$; CD4+/CD8+, $$p \leq 0.00$$), while CD8+ did not change significantly ($$p \leq 0.92$$) (Table-III). Comparative analysis of the incidence of adverse drug reactions between the two groups after treatment showed that the incidence of adverse reactions was $52.5\%$ in the experimental group, which was higher than that of the control group ($47.5\%$), with no statistically significant difference ($$p \leq 0.66$$) (Table-IV). Table-IV \| Group \| Rash \| Gastrointestinal reaction \| WBC decrease \| Renal impairment \| Neuritis \| Liver damage \| Incidence \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Experimental group \| 2.0 \| 3.0 \| 4.0 \| 3.0 \| 3.0 \| 6.0 \| 21 (52.5%) \| \| Control group \| 4.0 \| 3.0 \| 5.0 \| 0.0 \| 3.0 \| 4.0 \| 19 (47.5%) \| \| c2 \| \| \| \| \| \| \| 0.20 \| \| p \| \| \| \| \| \| \| 0.66 \| The comparative analysis of the negative conversion rate of sputum tuberculosis bacteria between the experimental group and the control group after treatment suggested that the negative conversion rate of patients in the experimental group was significantly higher than that in the control Group at one, three and six months after treatment, with a statistically significant difference ($p \leq 0.05$). Table-V Table-V \| Group \| 1 month* \| 3 months* \| 6 months* \| \| --- \| --- \| --- \| --- \| \| Experimental group \| 19.0 \| 34.0 \| 38.0 \| \| Control group \| 10.0 \| 25.0 \| 31.0 \| \| c2 \| 4.38 \| 5.23 \| 5.16 \| \| p \| 0.04 \| 0.02 \| 0.03 \| ## DISCUSSION Tuberculosis (TB), caused by Mycobacterium tuberculosis, is one of the oldest diseases known to affect humans and a leading cause of death worldwide. It is the leading cause of death in humans after HIV/AIDS.12 TB causes more deaths than any other infectious disease worldwide, with tuberculosis being the most frequent form.13 Delayed diagnosis remains a major challenge for tuberculosis control and prevention.14 Delayed initiation of treatment in patients with TB leads to increased infectivity, poor treatment outcomes, and increased mortality.15 Patients with diabetes are in a hyperglycemic state, which provides conditions for the survival of mycobacterium tuberculosis and promotes a higher incidence of tuberculosis.16 Patients with diabetes have a reduced ability to resist the invasion of external bacteria, aggravated islet B cell burden, and poor cellular immune function, resulting in an unsatisfactory prognosis. Given their clinical characteristics, patients with pulmonary tuberculosis complicated with diabetes have a lengthy course of treatment and suboptimal clinical efficacy than those with pulmonary TB alone. Currently, 2HRZE/10HRE is a recommended chemotherapy regimen17 promoted as stabilizing the condition of patients with retreatment smear-positive pulmonary tuberculosis. It has a certain therapeutic effect but is still not ideal.18 Each first-line antituberculosis drug reacts adversely in a separate way. Specifically, isoniazid interferes with carbohydrate metabolism and aggravates peripheral neuritis in patients with diabetes, while rifampicin reduces the hypoglycemic effects of sulfonylureas. Ethambutol and diabetes have double adverse reactions to the eyes, which can aggravate damage to the optic nerve. Hyperglycemia can affect the blood concentration of pyrazinamide.19 Given the drug resistance of TB bacteria and patient dependence, there is an urgent need for a new treatment strategy to simplify and shorten the course of treatment and improve the sensitivity of anti-tuberculosis drugs.20 It is strongly recommended in the 2018 WHO Treatment Guidelines for Multidrug- and rifampicin-Resistant Tuberculosis (MDR/RR-TB) that levofloxacin (or moxifloxacin)21 be used in combination to increase treatment efficacy and reduce the development of drug resistance. Ahmad et al.22 believed that the combination of a new generation of fluoroquinolones had better results in the treatment of tuberculosis. Grace et al.23 concluded that the use of levofloxacin-containing combination chemotherapy increased treatment efficacy with no difference in adverse events compared with standard chemotherapy. Jhun et al.24 suggested a 6-month regimen of isoniazid, rifampicin, ethambutol, pyrazinamide, and levofloxacin (LFX), which could reduce drug resistance and increase treatment success. Lan et al.25 considered that a short-course treatment regimen combined with fluoroquinolones could reduce the adverse reactions of first-line anti-tuberculosis drugs. According to the study of Serebryakova et al.26 It was believed that the fluoroquinolone levofloxacin had an effect on peripheral blood lymphocytes in patients with invasive pulmonary tuberculosis, and levofloxacin could increase the number of CD3 lymphocytes in patients with tuberculosis. Serebryakova et al.27 suggested that levofloxacin (fluoroquinolone) could inhibit the production of TNF-a in drug-resistant tuberculosis and the production of IL-12 and IFNγ in drug-sensitive tuberculosis, thereby enhancing the anti-inflammatory effect. Shah et al.28 confirmed that the drug concentration of levofloxacin in lung tissue was 1.71 times that of other tissues and continued to exert its efficacy for up to 120 hours. In other words, a smaller dose of levofloxacin could exert a larger therapeutic effect.29 It was finally confirmed in this study that the short-course chemotherapy regimen combined with levofloxacin had an efficacy of $90\%$, which was significantly higher than that of the control group ($67.5\%$), with a statistically significant difference ($$p \leq 0.01$$). After treatment, CD3+, CD4+, CD4+/CD8+ and other indicators in the experimental group were significantly higher than those in the control group, with a statistically significant difference (CD3+, $$p \leq 0.01$$; CD4+, $$p \leq 0.01$$; CD4+/CD8+, $$p \leq 0.00$$); The incidence of adverse reactions was $52.5\%$ in the experimental group and $47.5\%$ in the control group, with no statistically significant difference ($$p \leq 0.66$$); The negative conversion rate of patients in the experimental group was significantly higher than that in the control group at one month, three months and six months after treatment, with a statistically significant difference ($p \leq 0.05$). ## Limitations of the study: Nevertheless, shortcomings can still be seen in this study: fewer cases and short follow-up time. Moreover, no other treatment options are included for comparative analysis with this study. In response to this, more cases will be included and follow-up will continue to be extended, and other treatment regimens will be included in the study, in order to further elaborate the benefits of chemotherapy combined with levofloxacin on patients with pulmonary tuberculosis complicated with Type-2 diabetes. ## CONCLUSIONS Chemotherapy combined with levofloxacin is a safe and effective regimen for patients with pulmonary tuberculosis complicated with Type-2 diabetes, boasting a variety of benefits such as improved clinical efficacy, ameliorated cellular immune status, a high negative conversion rate of sputum tuberculosis bacteria, and no significant increase in adverse reactions. ## Author’s Contributions: PL and LW: Designed this study, prepared this manuscript, are responsible and accountable for the accuracy and integrity of the work. FL and YT: Collected and analyzed clinical data, and made important contributions to the design and thinking of the study. LC and YL: Data analysis, significantly revised this manuscript. ## References 1. Khan R, Malik NI, Razaque A. Imaging of Pulmonary Post-Tuberculosis Sequelae. Pak J Med Sci (2020) 36 S75-S82. PMID: 31933611 2. Nachiappan AC, Rahbar K, Shi X, Guy ES, Mortani Barbosa EJ, Shroff GS. Pulmonary Tuberculosis: Role of Radiology in Diagnosis and Management. Radiographics (2017) 37 52-72. PMID: 28076011 3. Sahakyan S, Petrosyan V, Abrahamyan L. Diabetes mellitus and treatment outcomes of pulmonary tuberculosis: A cohort study. Int J Public Health (2020) 65 37-43. PMID: 31240332 4. Deshmukh S, Atre S, Chavan A, Raskar S, Sawant T, Mave V. Assessment of the Xpert assay among adult pulmonary tuberculosis suspects with and without diabetes mellitus. Int J Tuberc Lung Dis (2020) 24 113-117. PMID: 32005314 5. Rao S, Rahim M, Iqbal K, Haroon F, Hasan Z. Impact of diabetes on mechanisms of immunity against Mycobacterium tuberculosis. J Pak Med Assoc (2019) 69 94-98. PMID: 30623920 6. Qiujing F, Weiwei W. Efficacy of moxifloxacin combined with levofloxacin in the treatment of drug-resistant tuberculosis. Pak J Pharm Sci (2020) 33 1361-1366. PMID: 33361023 7. Suarez I, Funger SM, Kroger S, Rademacher J, Fatkenheuer G, Rybniker J. The Diagnosis and Treatment of Tuberculosis. Dtsch Arztebl Int (2019) 116 729-735. PMID: 31755407 8. Caruso R, Magon A, Baroni I, Dellafiore F, Arrigoni C, Pittella F. Health literacy in type 2 diabetes patients: A systematic review of systematic reviews. Acta Diabetol (2018) 55 1-12. PMID: 29129000 9. Mirzayev F, Viney K, Linh NN, Gonzalez-Angulo L, Gegia M, Jaramillo E. World Health Organization recommendations on the treatment of drug-resistant tuberculosis, 2020 update. Eur Respir J (2021) 57 2003300. PMID: 33243847 10. Sidamo T, Shibeshi W, Yimer G, Aklillu E, Engidawork E. Explorative Analysis of Treatment Outcomes of Levofloxacin- and Moxifloxacin-Based Regimens and Outcome Predictors in Ethiopian MDR-TB Patients: A Prospective Observational Cohort Study. Infect Drug Resist (2021) 14 5473-5489. PMID: 34984005 11. Kulkarni S, Jha S. Artificial Intelligence, Radiology, and Tuberculosis: A Review. Acad Radiol (2020) 27 71-75. PMID: 31759796 12. Natarajan A, Beena PM, Devnikar AV, Mali S. A systemic review on tuberculosis. Indian J Tuberc (2020) 67 295-311. PMID: 32825856 13. Gallardo CR, Rigau Comas D, Valderrama Rodriguez A, Roquei Figuls M, Parker LA, Cayla J. Fixed-dose combinations of drugs versus single-drug formulations for treating pulmonary tuberculosis. Cochrane Database Syst Rev (2016) 2016 CD009913. PMID: 27186634 14. Getnet F, Demissie M, Assefa N, Mengistie B, Worku A. Delay in diagnosis of pulmonary tuberculosis in low-and middle-income settings: Systematic review and meta-analysis. BMC Pulm Med (2017) 17 202. PMID: 29237451 15. Tedla K, Medhin G, Berhe G, Mulugeta A, Berhe N. Factors associated with treatment initiation delay among new adult pulmonary tuberculosis patients in Tigray, Northern Ethiopia. PLoS One (2020) 15 e0235411. PMID: 32822368 16. Atif M, Anwar Z, Fatima RK, Malik I, Asghar S, Scahill S. Analysis of tuberculosis treatment outcomes among pulmonary tuberculosis patients in Bahawalpur, Pakistan. BMC Res Notes (2018) 11 370. PMID: 29884241 17. Martinez N, Kornfeld H. Tuberculosis and diabetes: from bench to bedside and back. Int J Tuberc Lung Dis (2019) 23 669-677. PMID: 31315698 18. Pranger AD, van der Werf TS, Kosterink JGW, Alffenaar JWC. The Role of Fluoroquinolones in the Treatment of Tuberculosis in 2019. Drugs (2019) 79 161-171. PMID: 30617959 19. Kumar AK, Chandrasekaran V, Kannan T, Murali AL, Lavanya J, Sudha V. Anti-tuberculosis drug concentrations in tuberculosis patients with and without diabetes mellitus. Eur J Clin Pharmacol (2017) 73 65-70. PMID: 27651240 20. Yew WW, Koh WJ. Emerging strategies for the treatment of pulmonary tuberculosis: promise and limitations?. Korean J Intern Med (2016) 31 15-29. PMID: 26767853 21. Van Deun A, Decroo T, Tahseen S, Trebucq A, Schwoebel V, Ortuno-Gutierrez N. World Health Organization 2018 treatment guidelines for rifampicin-resistant tuberculosis: uncertainty, potential risks and the way forward. Int J Antimicrob Agents (2020) 55 105822. PMID: 31626907 22. Ahmad N, Ahuja SD, Akkerman OW, Alffenaar JC, Anderson LF. Treatment correlates of successful outcomes in pulmonary multidrug-resistant tuberculosis: an individual patient data meta-analysis. Lancet (2018) 392 821-834. PMID: 30215381 23. Grace AG, Mittal A, Jain S, Tripathy JP, Satyanarayana S, Tharyan P. Shortened treatment regimens versus the standard regimen for drug-sensitive pulmonary tuberculosis. Cochrane Database Syst Rev (2019) 12 CD012918. PMID: 31828771 24. Jhun BW, Koh WJ. Treatment of Isoniazid-Resistant Pulmonary Tuberculosis. Tuberc Respir Dis (Seoul) (2020) 83 20-30. PMID: 31905429 25. Lan Z, Ahmad N, Baghaei P, Barkane L, Benedetti A, Brode SK. Drug-associated adverse events in the treatment of multidrug-resistant tuberculosis: An individual patient data meta-analysis. Lancet Respir Med (2020) 8 383-394. PMID: 32192585 26. Serebryakova VA, Urazova OI, Novitsky VV, Vengerovskii AI, Kononova TE, Vasil'eva OA. Effects of Levofloxacin on Blood Lymphocyte Apoptosis in Patients with Pulmonary Tuberculosis: an In Vitro Study. Bull Exp Biol Med (2019) 168 109-112. PMID: 31773352 27. Serebryakova VA, Urazova OI, Novitsky VV, Vengerovskii AI, Kononova TE. In Vitro Study of the Modulatory Effects of Levofloxacin and BCG on Secretion of Proinflammatory Cytokines in Infiltrative Pulmonary Tuberculosis. Bull Exp Biol Med (2018) 166 225-228. PMID: 30488211 28. Shah SR, Prajapati HR, Sheth DB, Gondaliya EM, Vyas AJ, Soniwala MM. Pharmacokinetics and in vivo distribution of optimized PLGA nanoparticles for pulmonary delivery of levofloxacin. J Pharm Pharmacol (2020) 72 1026-1037. PMID: 32337714 29. Deshpande D, Pasipanodya JG, Mpagama SG, Bendet P, Srivastava S, Koeuth T. Levofloxacin Pharmacokinetics/Pharmacodynamics, Dosing, Susceptibility Breakpoints, and Artificial Intelligence in the Treatment of Multidrug-resistant Tuberculosis. Clin Infect Dis (2018) 67 S293-S302. PMID: 30496461
--- title: Liver and Renal Injury with Remdesivir Treatment in SARS-CoV-2 Patients authors: - Rabiah Sadaf - Faiza Sadaqat Ali - Tazeen Rasheed - Bader Faiyaz Zuberi journal: Pakistan Journal of Medical Sciences year: 2023 pmcid: PMC10025697 doi: 10.12669/pjms.39.2.6236 license: CC BY 3.0 --- # Liver and Renal Injury with Remdesivir Treatment in SARS-CoV-2 Patients ## Abstract ### Objective: To determine the effect of Remdesivir on liver enzymes and renal functions in SARS-CoV-2 patients. ### Methods: This prospective cohort study was conducted at Dr. Ruth KM Pfau, Civil Hospital Karachi between 1st December 2021 to 31st January, 2022. All patients of severe SARS-CoV-2 infection who received Inj. Remdesivir for five days as per protocol of SARS-CoV-2 management were included. Biodata of selected patients including age, gender, diabetic, hypertensive status was recorded. Patients Liver Function Tests and Serum Creatinine were performed on days 0, 3, 5, 7 and 14. ### Result: This study included 85 patients, out of which 55 ($64.7\%$) were males and 30 ($35.3\%$) were females. Out of 85 patients, Remdesivir was stopped in 3 ($3.5\%$) patients. Among these three patients Remdesivir was stopped in one patient on day three because of decrease in CrCl to <30 ml/min. His CrCl improved after stopping Remdesivir. In the remaining two patients, Remdesivir was stopped due to increase in ALT to greater than 10 times from normal values on day three. Similarly, in these two patients the ALT improved after stopping Remdesivir. ### Conclusion: Only three patients developed adverse effects resulting in stopping of Remdesivir, however these were reversible on stopping the drug. Therefore, Remdesivir is a relatively safe drug and well tolerated in SARS-CoV-2 patients. ## INTRODUCTION Since the emergence of highly contagious respiratory illness caused by SARS-CoV-2 virus in December, 2019 from Wuhan, China SARS-CoV-2 has become a global pandemic affecting almost every aspect of life. This pandemic has devastated economies and given surpassing challenges to the healthcare systems around the world.1 To counteract this threat several drugs have been investigated, one of them is Remdesivir which got FDA approval in 2020 for the treatment of Corona virus disease.2 Remdesivir is an antiviral which was first developed in 2016 to control Ebola Virus disease outbreak in Africa.3 Several in vivo and in vitro studies in animal models suggested its potential broad spectrum activity against other viruses as well such as SARS, MERS, Marburg and SARS-CoV-2.4 Remdesivir and its active metabolite are mostly ($74\%$) eliminated by kidneys. The plasma half-life of the parent drug is short around one to two hours, but the half-life of its active metabolite Remdesivir triphosphate is about 20–25 hours. Remdesivir has limited water solubility, largely excreted through glomerular filtration, hence its elimination becomes slow in patients with abnormal renal functions. In the liver, Remdesivir is metabolized by CYP3A4.5 There is limited local data regarding the hepatic and renal safety profile of this drug in our population, making treatment decisions difficult for physicians, therefore we planned this study to assess the effects of this only approved antiviral drug till date on liver and renal profile in our population. This may guide how frequently liver enzymes and serum creatinine should be monitored in SARS-CoV-2 patients taking Remdesivir and when the drug should be stopped. The objective was to determine the effect of Remdesivir on liver enzymes and renal functions in SARS-CoV-2 patients. ## Severe SARS-CoV-2 Infection: Individuals with SpO2 < $94\%$ on room air, respiratory rate of > 25 breaths/min, or lung infiltrates on CXR > $50\%$, requiring supplemental oxygen.6 ## Normal values of Liver Function Tests: ALT: 19-25 IU/L for females; 29-33 IU/L for males7 ## Alkaline phosphatase: 50-100 U/L, Total bilirubin: 0.3-1.0 mg/dl Normal value of CrCl: Male: 110 to 150 ml/min, Female: 100 to 130 ml/min8 ## Creatinine Clearance (mL/min): “calculated using Cockcroft-Gault formula”9 For females result of the equation was multiplied by 0.85. (Age in years; weight in Kg; Serum Creatinine in mg/dl) ## METHODS This Prospective Cohort study was conducted at Dr. Ruth KM Pfau, Civil Hospital Karachi between 1st December,2021 to 31st January,2022 satisfying inclusion/exclusion criteria were included after informed consent and ethical approval (IRB-2268/DUHS/Approval/$\frac{2021}{619}$ dated 29th November, 2021) was taken from Institutional Review Board of Dow University of Health Sciences. A total of 85 patients were included. Non-probability consecutive sampling was used for selection of patients. ## Inclusion criteria: All patients of severe SARS-CoV-2 infection who received Inj. Remdesivir for 5 days as per protocol of SARS-CoV-2 management.11 ## Exclusion criteria: Patients having ALT > 5 times the ULN prior giving Remdesivir, known patients of HBV, HDV & HCV, patients having eGFR < 30 ml/min, pregnant or breast-feeding females, patients with known hypersensitivity to Remdesivir, those who expired before 14th day when last sample was collected were excluded. Given frequencies from previous study of liver $35\%$ involvements in SARS-CoV-2 patients receiving Remdesivir. Using power of $80.0\%$ to detect difference of P0-P1 of -0.15 and alpha of 0.05, the sample size was calculated as 85. Calculation was done using PASS software.12 All admitted patients meeting inclusion criteria were included after taking informed consent by non-parametric consecutive sampling method. Biodata of selected patients including age, gender, diabetic, hypertensive status was recorded. After all aseptic measures blood was drawn by a trained phlebotomist for Liver Function Tests and Serum Creatinine on days 0, 3, 5, 7 and 14. On day zero tests for HBs Ag and HCV Ab were also done. LFTs and serum creatinine were done by photometric method using Cobas C501 analyzer. ## Data analysis: Frequencies of gender, diabetes and hypertension were reported and compared by χ2-test. Frequencies of different adverse effects on liver and renal functions were reported. Means ±SD of quantitative variables like age, duration, liver enzymes values and renal parameters were determined and compared by Student’s t-test. Significant level was set at ≤.05. Data was analyzed using SPSS version 26. ## RESULTS In this study eighty-five patients admitted in SARS-CoV-2 treatment facility of Dr Ruth KM Pfau, Civil Hospital Karachi were enrolled. Mean ±SD of age of patients was 51.47 ±13.39. Out of the 85 patients 55 ($64.7\%$) were males and 30 ($35.3\%$) were females. Details of frequencies of Hypertension, Diabetes Mellitus and age and their statistical comparison is given in Table-II. The Mean ± SD of CrCl & ALT on days 0, 3, 7 & 14 are shown in Table-III. Out of 85 patients, Remdesivir was stopped in 3 ($3.5\%$) patients. Among these three patients Remdesivir was stopped in one male patient because of decrease in CrCl to <30 ml/min. In the remaining two patients Remdesivir was stopped on day three due to increase in the ALT levels. Follow up of three patients in whom Remdesivir was stopped due to decrease in CrCl (Patient one) & increased ALT (Patients two & three) on Day three is given in Table-IV. Table-IV \| Unnamed: 0 \| Parameter \| Day 0 \| Day 3 \| Day 7 \| Day 14 \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| Patient 1 \| CrClµ \| 88.5 \| 29.5 \| 40.3 \| 55.3 \| \| Patient 2 \| Bilirubin (mg/dl) \| 0.5 \| 0.7 \| 0.7 \| 0.6 \| \| Patient 2 \| ALT* (IU/L) \| 100.0 \| 343.0 \| 280.0 \| 62.0 \| \| Patient 2 \| Alk Phos \| 125.0 \| 127.0 \| 122.0 \| 100.0 \| \| Patient 3 \| Bilirubin(mg/dl) \| 0.5 \| 0.9 \| 0.7 \| 0.4 \| \| Patient 3 \| ALT (IU/L) \| 67.0 \| 354.0 \| 247.0 \| 54.0 \| \| Patient 3 \| Alk Phos** \| 135.0 \| 194.0 \| 132.0 \| 128.0 \| There was no significant gender difference in patients who developed derangement in hepatic and renal function while on Remdesivir. [ χ2 (df=1, $$n = 85$$) = 0.005; $$p \leq 0.942$$]. There was no significant difference in age in patients who developed derangement in hepatic and renal function while on Remdesivir. t [83] = 1.080, $$p \leq 0.28.$$ In one-way ANOVA, there was significant difference in ALT on days three and seven between patients in whom Remdesivir was stopped and not stopped. Day three [F [1, 83] = 67.46, $p \leq 0.001$)]. Day three [(F [1,83] =48.56, $p \leq 0.001$)]. ## DISCUSSION Our study demonstrated safety of Remdesivir on liver and kidney function. Majority of patients in our study who were hospitalized and received maximum five days of Remdesivir, did not develop significant hepatotoxicity or nephrotoxicity. Only three out of eighty-five ($3.53\%$) patients developed drug dependent adverse effects to the extent of stopping the drug. Out of these three, only one patient developed significant derangement of creatinine clearance to less than 30 ml/min, and two patients developed significant elevation in ALT to almost 10 times upper limit of normal, necessitating stoppage of Remdesivir on day 3, with gradual improvement in renal and liver functions after stopping the drug. Mild derangement of liver enzymes, bilirubin and creatinine clearance due to Remdesivir, was noted in some case studies but data regarding severe hepatic and renal function derangement by this drug causing stopping it, is scarce.13 It has been documented that abnormalities in liver functions correlates with severity of Covid infection.14 Covid infection also had impact on eye and GI hemorrhage and its management.15,16 Very few studies demonstrated the safety of Remdesivir in hospitalized SARS-CoV-2 patients on renal and hepatic functions. Although SARS-CoV-2 infection can cause aminotransferase elevation, patient 2 had raised ALT levels upto 2.7 times prior to initiating Remdesivir and patient 3 had ALT levels 1.8 times prior to initiating Remdesivir. Their ALT levels increased to more than 9 times after Remdesivir, suggesting a direct role of Remdesivir in hepatocellular toxicity. van Laar SA et al. in their study on 103 hospitalized patients, on oxygen who received five days treatment of Remdesivir, revealed $11\%$ of the patients had a decline in estimated glomerular filtration rate >10 mL/min/1.73m2, $25\%$ had raised ALT and $35\%$ had raised AST levels from their baseline values.17 However similar to our study, severe derangements were less, as six out of 103 ($5.82\%$) patients developed eGFR less than 30ml/min and $\frac{5}{103}$ ($4.85\%$) developed >5 times ULN of ALT and AST.17 World health organization in its Database on 439 individual case reports, highlighted suspected adverse drug events due to Remdesivir and reported elevation in liver enzymes ($32.1\%$), renal injury ($14.4\%$) and increase serum creatinine ($11.2\%$) of patients.18 They also reported majority of drug related adverse effects were seen in Americans ($67.7\%$) and mostly in males > 45 years of age,18 but in our study we did not find significant age or gender difference in patients who developed derangement in hepatic and renal function while on Remdesivir. Pettit NN et al. in their comparative study at academic medical center in Chicago, Illinois, gave intravenous solution of Remdesivir.19 Out of 137 patients whom they gave Remdesivir, 20 ($14.8\%$) patients already had severe renal impairment as 15 patients had CrCl <30 mL/min and five patients had ESRD that required intermittent hemodialysis, where they concluded that Remdesivir related toxic effects in patients with SARS-COV-2 and severe renal impairment were same as in SARS-COV-2 patients without severe renal impairment as only $\frac{4}{20}$ ($20\%$) patients had further serum Cr elevations following Remdesivir, of which three of four patients were already suffering from AKI, before the initiation of RDV.19 A study in Pakistan had earlier showd that RDV did not show any difference in in-hospital mortality in Covid19 patients,, more patients had severe ARDS in the RDV group The authors also reported that the length of stay was longer in patients receiving Remdesivir therapy. 20 One of the major strengths of our study is its study design being prospective cohort so provides clarity of the temporal sequence and the selection bias is minimum. We assessed clinical safety of Remdesivir and evaluated the actual clinical need of stopping treatment from severe adverse effect versus theoretical knowledge of Remdesivir adverse effects. There is very limited data available in Pakistan and worldwide on this clinical assessment in cohort of relatively large number of patients. ## Limitations: We excluded patients with ALT or AST > 5 times the ULN prior starting Remdesivir, known HBV, HDV & HCV or any known liver disease, eGFR < 30 ml/min so more studies with larger sample size would be beneficial to evaluate the effect of Remdesivir in these group of patients in more detail. Another limitation was that this was a single center study and patients were not followed later. ## CONCLUSION Majority of our patients did not develop any significant liver of renal adverse effects due to Remdesivir. Only three patients developed adverse effects necessitating the stoppage of Remdesivir, however these were reversible on stopping the drug. Therefore, Remdesivir is a relatively safe drug and well tolerated in SARS-CoV-2 patients. ## Authors’ Contribution: RS, TR, BFZ: Substantial contributions to conception and design, or acquisition of data, or analysis and interpretation of data. RS, FSA: Drafting the article or revising it critically for important intellectual content. BFZ: Final approval of the version to be published. TR, FSA: Statistical Analysis. All Authors: Agree to be 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. ## References 1. Lone SA, Ahmad A. Covid-19 Pandemic–an African Perspective. Emerg Microbes Infect (2020.0) 9 1300-1308. PMID: 32458760 2. Aleem A, Kothadia J. Remdesivir. StatPearls (2021.0) 3. Nili A, Farbod A, Neishabouri A, Mozafarihashjin M, Tavakolpour S, Mahmoudi H. Remdesivir: A Beacon of Hope from Ebola Virus Disease to Covid-19. Rev Med Virol (2020.0) 30 1-13 4. Elsawah HK, Elsokary MA, Abdallah MS, ElShafie AH. Efficacy and Safety of Remdesivir in Hospitalized Covid-19 Patients: Systematic Review and Meta-Analysis Including Network Meta-Analysis. Rev Med Virol (2020.0) e2187. PMID: 33128490 5. Yang K. What Do We Know About Remdesivir Drug Interactions?. Clin Translational Sci (2020.0) 13 842 6. Wang YY, Huang Q, Shen Q, Zi H, Li BH, Li MZ. Quality of and Recommendations for Relevant Clinical Practice Guidelines for Covid-19 Management: A Systematic Review and Critical Appraisal. Front Med (Lausanne) (2021.0) 8 630765. PMID: 34222270 7. Kwo PY, Cohen SM, Lim JK. Acg Clinical Guideline: Evaluation of Abnormal Liver Chemistries. Offi J Am Coll Gastroenterol (2017.0) 112 18-35 8. Shahbaz HGM. Creatinine Clearance. StatPearls [Internet] 9. Michels WM, Grootendorst DC, Verduijn M, Elliott EG, Dekker FW, Krediet RT. Performance of the Cockcroft-Gault, Mdrd, and New Ckd-Epi Formulas in Relation to Gfr, Age, and Body Size. Clin J Am Soc Nephrol (2010.0) 5 1003-1009. PMID: 20299365 10. 10U.S. Department of Health and Human Services, National Institutes of Health, National Institute of Allergy and Infectious Diseases, Division of Aids. Division of Aids (Daids) Table for Grading the Severity of Adult and Pediatric Adverse Events, Corrected Version 2.1Available from: https://Rsc.Niaid.Nih.Gov/Sites/Default/Files/Daidsgradingcorrectedv21.Pdf (July 2017) 11. 11Ministry of National Health Services raCClinical Managment Guidelines for Covid-19 InfectionsAccessed August 5 2021https://storage.covid.gov.pk/new_guidelines/11December2020_20201211_Clinical_Management_Guidelines_for_COVID-19_infection_1204.pdf. Clinical Managment Guidelines for Covid-19 Infections 12. van Laar SA, de Boer MGJ, Gombert-Handoko KB, Guchelaar HJ, Zwaveling J. Liver and Kidney Function in Patients with Covid-19 Treated with Remdesivir. Br J Clin Pharmacol (2021.0) 87 4450-4454. PMID: 33763917 13. Zampino R, Mele F, Florio LL, Bertolino L, Andini R, Galdo M. Liver Injury in Remdesivir-Treated Covid-19 Patients. Hepatol Int (2020.0) 14 881-883. PMID: 32725454 14. Zuo G, Liu W. The Clinical Characteristics and Underlying Causes of Liver Damage in Patients with Covid-19 Infection: Retrospective Analysis. Pak J Med Sci (2021.0) 37 1282-1287. PMID: 34475899 15. Awan MA, Shaheen F, Mohsin F. Impact of Covid-19 Lockdown on Retinal Surgeries. Pak J Med Sci (2021.0) 37 1808-1812. PMID: 34912399 16. Chen J, Hang Y. Characteristics, Risk Factors and Outcomes of Gastrointestinal Hemorrhage in Covid-19 Patients: A Meta-Analysis. Pak J Med Sci (2021.0) 37 1524-1531. PMID: 34475942 17. van Laar SA, de Boer MGJ, Gombert-Handoko KB, Guchelaar HJ, Zwaveling J. Liver and Kidney Function in Patients with Covid-19 Treated with Remdesivir. Br J Clin Pharmacol (2021.0) 87 4450-4454. PMID: 33763917 18. Charan J, Kaur RJ, Bhardwaj P, Haque M, Sharma P, Misra S. Rapid Review of Suspected Adverse Drug Events Due to Remdesivir in the Who Database;Findings and Implications. Expert Rev Clin Pharmacol (2021.0) 14 95-103. PMID: 33252992 19. Pettit NN, Pisano J, Nguyen CT, Lew AK, Hazra A, Sherer R. Remdesivir Use in the Setting of Severe Renal Impairment: A Theoretical Concern or Real Risk?. Clin Infect Dis (2021.0) 73 e3990-e3995. PMID: 33315065 20. Shaikh Q, Sarfaraz S, Rahim A, Hussain M, Shah R, Soomro S. Effect of Remdesivir on mortality and length of stay in hospitalized COVID-19 patients: A single center study. Pak J Med Sci (2022.0) 38 405-410. PMID: 35310809
--- title: Effects of body mass index on propofol-induced cardiovascular depression in the Pakistani population authors: - Uzma Naeem - Akbar Waheed - Yasmeen Azeem - Muhammad Nazir Awan journal: Pakistan Journal of Medical Sciences year: 2023 pmcid: PMC10025699 doi: 10.12669/pjms.39.2.6787 license: CC BY 3.0 --- # Effects of body mass index on propofol-induced cardiovascular depression in the Pakistani population ## Abstract ### Objective: To determine the relationship between the patient’s Body Mass Index (BMI) and the cardiovascular effects produced by propofol at a dose of 1.5 mg/kg in the Pakistani population. ### Methods: This descriptive cross-sectional study was conducted in the Holy Family Hospital Rawalpindi from August 2021 to January 2022. According to their BMI, one hundred twenty Pakistani individuals 18 to 60 years of age were equally divided into three groups. Group N ($$n = 40$$) with a BMI of 18 to 24.9, group OW ($$n = 40$$) with a BMI of 25 to 29.5, and group O ($$n = 40$$) with a BMI of 30 to 34.9 were randomized to receive propofol injections at a 1.5 mg/kg dose for induction of anesthesia. We measured mean blood pressure before the propofol and then at one, three, and ten minutes after the injection. Data were analyzed by using SPSS 22. ### Results: Mean blood pressure decreases significantly in all groups, as shown by p-values of <0.001 for the first two readings. In group N, blood pressure returned to near normal within ten minutes (p-value 0.061), but in groups, OW and O, mean blood pressure was significantly lower even after ten minutes (p-values 0.005 and 0.001, respectively). Individual variations in propofol response were also observed. ### Conclusion: In the Pakistani population, propofol at an induction dose of 1.5 mg/kg to patients with different body weights produces cardiovascular effects with marked standard deviations in each group, which indicate different individual responses. Clinical Trial Number: NCT05383534 https://register.clinicaltrials.gov/ ## INTRODUCTION Propofol is a phenolic by-product, available as an oil-in-water intravenous emulsion for calming and mesmerizing purposes. It is a broad-spectrum ultrashort-acting anesthetic, which has significant advantages in anesthetic potency and safety.1 Easy control of the depth of anesthesia, rapid recovery of consciousness, and less postoperative nausea and vomiting are the characteristics that make propofol a standard agent for induction of anesthesia.2,3 Common adverse effects of propofol include hypotension, bradycardia, respiratory depression, myoclonus, and pain at the injection site.4 The most common among these adverse effects is propofol-induced hypotension; this decrease in blood pressure is due to the inhibition of myocardial contractility, a decrease in peripheral resistance, and sympathetic inhibition.5 Regarding the pharmacokinetics of propofol, its highly lipophilic feature leads to a high volume of distribution and a long elimination half-life, although it undergoes extensive metabolism in the liver. Propofol pharmacokinetics and pharmacodynamics are subjected to high interindividual variability, leading to the variabilities in the required induction dose and the adverse cardiovascular effects produced by the propofol. The factors that mainly contribute to this variability include age, sex, genetic variations, increase in adipose tissue, lean body weight, extracellular fluid, and cardiac output.6,7 The augmented central volume of distribution and variations in the clearance of drugs affect the plasma concentration of propofol in the obese population because obesity changes body composition and physiology.7 It was demonstrated in a previous study that about $40\%$ of the excess mass of obese individuals results from the increased fat-free mass.8 Keeping this in mind, it could be assumed that intravenous drug doses scaled according to the total body weight can result in overdosing in overweight and obese individuals and subsequent dose-related adverse effects. Available studies of weight-related propofol pharmacokinetics are scarce and derived from a small number of patients. We know that no public comparative research compares cardiovascular effects produced by standard propofol doses in patients with different body mass indexes in the Pakistani population. With this background, this study aimed to investigate the standard propofol dose-induced cardiovascular depression by measuring mean blood pressure in normal, overweight, and obese patients using a population-based approach to predict variabilities related to the same propofol dose. We can minimize the cardiovascular risks associated with propofol anesthesia by predicting these variabilities. ## METHODS This descriptive cross-sectional study was carried out in Holy Family hospital Rawalpindi’s anesthesia department for six months, from August 2021 to January 2022, collaborating with Islamic International Medical College Rawalpindi. The study procedure was accepted and approved by the institutional ethical review board (Riphah/IIMC/IRC/$\frac{20}{002}$) and the clinical trial number is NCT05383534. After taking the written and verbal informed consent, one hundred twenty Pakistani patients (both male and female) aged more than 18 and less than 60 were selected for this study. Another essential inclusion criterion was the BMI of patients; 40 patients with normal BMI (Group-N), forty were overweight (Group-OW), and 40 were obese (Group-O). Patients of extreme age and fitted in classes III, IV, V &, and VI of the ASA (American Society of Anesthesiology) scale were excluded from the study. The patient’s BMI was calculated using the formula BMI = kg/m2, where kg is a person’s weight in kilograms and m2 is their height in meters squared. The patients with a BMI of 18.5 to 24.9, 25 to 29.5, and 30 to 34.9 were considered normal, overweight, and obese, respectively.9 The sample size for this study was estimated by employing a previously published study.10 All these patients were in classes I and II of the American Society of Anesthesiologists (ASA)11 scale and received Propofol at a dose of 1.5 mg/kg for induction of anesthesia.12 The first reading of the mean blood pressure of patients under study was measured before the propofol induction and after 1, 3, and 10 minutes of propofol injection.11 Mean arterial pressure (MAP) was calculated using the formula MAP = DP + $\frac{1}{3}$(SP – DP). Here DP is the diastolic blood pressure, and SP is the systolic blood pressure.13 ## Statistical analysis: BMI and cardiovascular outcomes in the form of mean blood pressure were recorded for all the patients in the form of excel files. Excel files were brought into the Statistical Package for Social Sciences (SPSS) version 22 for analysis. The normal distribution of data was confirmed with the Kolmogorov–Smirnov test, and differences in the mean blood pressure between the three groups at 0, 3, and 10 minutes were analyzed using Student’s paired t-tests with $P \leq 0.05$ considered significant. All values are given as the mean (±SD). Comparison among the groups was made by Dunn’s All-Pairwise Comparisons Test. ## RESULTS Our study included 120 Pakistani patients, divided into three groups according to their BMI, N (standard), OW (overweight), and O (obese). Participants’ variables are shown in Table-I. Table-I \| Groups \| Male \| Female \| Age ± SD \| BMI ± SD \| \| --- \| --- \| --- \| --- \| --- \| \| N (n=40) \| 27 \| 13 \| 45 ± 5.61 \| 21 ± 2.34 \| \| OW(n=40) \| 19 \| 21 \| 49 ± 9.59 \| 27.5 ± 2.11 \| \| O (n=40) \| 11 \| 29 \| 43 ± 7.51 \| 32.5 ± 2.19 \| Patients of Group-N showed a significant decrease in the mean blood pressure after one and three minutes compared with the pre-propofol reading, with a p-value of <0.001, while the reduction in mean blood pressure was not significant after 10 minutes (p-value 0.061). Within ten minutes of propofol injection, blood pressure comes back to normal in patients with a BMI of 18.5 to 24.9 (Group-N). While in patients with BMI between 25 to 29.9 (Groups-OW & O), we observed that this decrease in blood pressure was significant at three-time intervals with p-values of <0.001, <0.001, and 0.005, respectively for Group-OW, and <0.001, <.001, & <0.001 for Group-O. All these results are summarized in Table-II. Table-II \| Groups \| BMI of patients \| Mean BP before propofol injection \| Mean BP before propofol injection.1 \| Mean BP after one minute of propofol injection \| Mean BP after one minute of propofol injection.1 \| Mean BP after one minute of propofol injection.2 \| Mean BP after 3 minutes of propofol injection \| Mean BP after 3 minutes of propofol injection.1 \| Mean BP after 3 minutes of propofol injection.2 \| Mean BP after 10 minutes of Propofol \| Mean BP after 10 minutes of Propofol.1 \| Mean BP after 10 minutes of Propofol.2 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Groups \| BMI of patients \| \| \| \| \| \| \| \| \| \| \| \| \| Groups \| BMI of patients \| Mean \| SD \| Mean \| SD \| p1 \| Mean \| SD \| p2 \| Mean \| SD \| p3 \| \| Normal (N) \| 18.5 to 24.9 kg/m2 \| 97.800 \| 6.365 \| 86.900 \| 6.644 \| <.001 \| 90.950 \| 6.304 \| <.001 \| 96.600 \| 6.067 \| 0.061 \| \| Overweight (OW) \| 25 to 29.9 kg/m2 \| 101.050 \| 7.099 \| 83.975 \| 9.960 \| <.001 \| 95.550 \| 10.869 \| <.001 \| 98.450 \| 8.819 \| 0.005 \| \| Obese (O) \| 30 to 34.9 kg/m2 \| 98.80 \| 8.919 \| 76.125 \| 9.648 \| <.001 \| 80.200 \| 12.999 \| <.001 \| 84.525 \| 17.885 \| <.001 \| An important finding of our study was the higher values of standard deviation (SD) among the individuals of the same group, as shown in Table-I. This shows the inter-individual variability or different responses of the individuals with the same BMI and the same dose of Propofol (Fig.1, 2, & 3). These differences are more marked in Group-O (with a BMI of 30 to 34.9), as shown in Fig.3. Fig.1: Comparison of mean blood pressures of patients with normal BMI. Fig.2: Comparison of mean blood pressures of overweight patients. Fig.3: Comparison of mean blood pressures of obese patients. ## DISCUSSION In our study, all the participants were Pakistani with no discrimination of sex. The mean dosage of propofol used to induce the anesthesia was 1.5 mg/Kg, the standard dose usually used for induction of anesthesia associated with minimum adverse effects.12,14 The fluctuations of mean blood pressure in our study were more marked in Group-O, less in Group-OW, and lesser in Group-N. The cardiovascular effects of propofol have been studied by many researchers, but very few of them explained its relationship with BMI in this way. Our findings of group N are consistent with a recent study from the United States of America using a propofol dose range of 1.5-2.5 mg/kg, which reported hypotensive effects of propofol at 45 seconds and three minute intervals in patients with normal BMI (p-value less than 0.05).15 Another study that Kawasaki and his colleagues conducted showed the association of age with hemodynamic fluctuations in patients with normal BMI and observed significant cardiovascular depression.16 Recently, in 2022, Oda reported the hypotensive effects of propofol in patients with intellectual disabilities.17 According to our knowledge, only one study was conducted in South Africa comparing normal weight, overweight, and obese individuals to observe propofol’s induced cardiovascular effects. But they used an adjusted body mass scaler and didn’t mention the time intervals at which different observations were made.18 In our study, the most marked mean blood pressure decrease was observed in Group-O at three-time intervals with p-values of <0.001. These findings are consistent with the results of Ding and Hu, who studied the efficacy and safety of propofol in obese patients and observed a significant decrease in mean arterial pressure (p-value was 0.03).19 A comprehensive study on the effects of BMI on propofol pharmacokinetics and dynamics was done by Dong and his colleagues in 2016, but they didn’t compare different BMIs and meant blood pressure monitoring for 10 minutes.5 Alam et al. and Khattak et al. from Pakistan used midazolam along with propofol to decrease its adverse effects in cirrhotic and non-cirrhotic patients.20 In 2019 safety of propofol was studied in Pakistan and the results showed a less marked decrease in blood pressure after induction with propofol 1 mg/kg.21 Hamid and his colleagues tried to use the priming principle to decrease the induction dose of propofol because propofol dose increment is directly related to cardiovascular depression. 22 The incidence of post-induction hypotension was also studied recently in 2022 and it was concluded that induction with propofol & thiopental and orthopedics surgery are the independent risk factors for this hypotension.23 The most remarkable finding of our study was the higher values of standard deviations in each comparison. These deviations indicate the individual differences in propofol handling, which affect its serum levels and adverse effects. The most common cause of these variations is the difference in the alleles of the enzymes responsible for the metabolism of propofol.24,25 To our knowledge, there has been no study from the Pakistan population to evaluate propofol’s effectiveness and adverse effect profile. ## Strengths of the study: The potential strengths of our study were the adjustment of a single per kg dose for all the patients, dividing them into three groups according to their BMI, and close monitoring of cardiovascular effects for ten minutes. Our findings will help to minimize the cardiovascular adverse effects of propofol. ## Limitations of the study: We also acknowledge the potential limitations of our study, including the comparison with other doses of Propofol, evaluation of the population’s genetic makeup, and its correlation with the propofol serum levels. ## CONCLUSION Propofol administration based on body weight in overweight and obese patients is associated with a high risk of overdosing and dose-related adverse effects. Several mass scalars, including the ideal body weight and lean body weight, have been introduced to calculate drug dosages more correctly for overweight and obese patients; however, it is essential to carefully monitor the effects and side effects of the drugs during administration. At an induction dose of 1.5 mg/kg when given to Pakistani patients (with different body weights), Propofol produces cardiovascular effects with marked standard deviations in each group, which indicate different personal responses apart from the body weight effects. ## Future recommendations: a comprehensive study on the correlation between the propofol serum levels and different alleles of the enzymes responsible for the metabolism of Propofol should be performed. ## Authors Contribution UN conceived, designed, and did statistical analysis & manuscript writing, and is responsible for research integrity and she is the corresponding author. YA did data collection NA did a review and final approval of the manuscript. ## Abbreviation: BMI: Body Mass Index OW: Overweight N: Normal ICU: Intensive Care Unit O: Obese MAP: Mean Arterial Pressure ASA: American Society of Anesthesiologists DP: Diastolic Pressure SP: Systolic Pressure SPSS: Statistical Package for Social Sciences SD: Standard Deviation ## References 1. Yang L, Chen Z, Xiang D. Effects of intravenous anesthesia with sevoflurane combined with propofol on intraoperative hemodynamics, postoperative stress disorder and cognitive function in elderly patients undergoing laparoscopic surgery. Pak J Med Sci (2022) 38 1938-1944. PMID: 36246684 2. Ricardo JDO. Metabolic Profiles of Propofol and Fospropofol: Clinical and Forensic Interpretative Aspects. Biomed Res Int (2018) 4 1-16 3. Muta K, Miyabe-Nishiwaki T, Masui K, Yajima I, Iizuka T, Kaneko A. Pharmacokinetics, and effects on clinical and physiological parameters following a single bolus dose of Propofol in common marmosets (Callithrix jacchus). J Vet Pharmacol Ther (2021) 44 18-27. PMID: 32880998 4. Doğanay F, Ak R, Alışkan H, Abut S, Sümer E, Onur Ö. The effects of intravenous lipid emulsion therapy in the prevention of depressive effects of Propofol on cardiovascular and respiratory systems: An experimental animal study. Medicina (2019) 55 1 5. Saeed R, Jahan SS, Bangash T, Sarmad N, Kazmi S, ALI I. Comparison of Haemodynamic Stability at Co-Induction of Propofol-Pethidine and Propofol-Nalbuphine. Pak J Med Health Sci (2021) 15 756-759 6. Dong D, Peng X, Liu J, Qian H, Li J, Wu B. Morbid Obesity Alters Both Pharmacokinetics and Pharmacodynamics of Propofol: Dosing Recommendation for Anesthesia Induction. Drug Metab Dispos (2016) 44 1579-1583. PMID: 27481855 7. Kim TK. Obesity and anesthetic pharmacology: simulation of target-controlled infusion models of Propofol and remifentanil. Korean J Anesthesiol (2021) 74 478-487. PMID: 34407372 8. Smith FJ, Coetzee JF, Jurgens FX, Becker PJ. An observational study is the induction of anesthesia with Propofol according to the adjusted ideal body mass in obese and non-obese patients. South Afr J Anaesth Analg (2019) 25 6-11 9. Suvarna TP, Raj JO, Prakash N. Correlation between Balance, and BMI in Collegiate students: A cross-sectional study. Int J Physiother Res (2021) 9 3759-3764 10. Kokong DD, Pam IC, Zoakah AI, Danbauchi SS, Mador ES, Mandong BM. Estimation of weight in adults from height: a novel option for a quick bedside technique. Int J Emerg Med (2018) 11 1-9. PMID: 29299773 11. Mikstacki A, Zakerska-Banaszak O, Skrzypczak-Zielinska M, Tamowicz B, Prendecki M, Dorszewska J. The effect of UGT1A9, CYP2B6 and CYP2C9 genes polymorphism on individual differences in propofol pharmacokinetics among Polish patients undergoing general anesthesia. J Appl Genet (2017) 58 213-220. PMID: 27826892 12. Ababneh OA, Suleiman AM, Bsisu IK. A Co-Induction Technique Utilizing 4% Sevoflurane Followed by 0.75 mg/kg Propofol in Elderly Patients Undergoing Minimally Invasive Procedures: A Prospective Randomized Control Study. Medicina (Kaunas) (2020) 56 682. PMID: 33321778 13. DeMers D, Wachs D. Physiology, Mean Arterial Pressure. StatPearls internet (2022) 14. Xiaoqian Z, Tao Z, Bingsong L, Jing L, Yu D, Weilan Z. Clinical comparative study on Nitrous Oxide inhalation versus intravenous propofol and Midazolam sedation in Transnasal Gastroscopy. Pak J Med Sci (2017) 33 891-894. PMID: 29067060 15. Saugel B, Bebert EJ, Briesenick L, Hoppe P, Greiwe G, Yang D. Mechanisms contributing to hypotension after anesthetic induction with sufentanil, Propofol, and rocuronium: a prospective observational study. J Clin Monit Comput (2021) 1 1-7 16. Kawasaki S, Kiyohara C, Tokunaga S, Hoka S. Prediction of hemodynamic fluctuations after induction of general anesthesia using Propofol in non-cardiac surgery: a retrospective cohort study. BMC Anesthesiol (2018) 18 1-10. PMID: 29298664 17. Oda Y, Yoshida K, Kawano R, Yoshinaka T, Oda A, Takahashi T. Effects of antipsychotics on intravenous sedation with midazolam and propofol during dental treatment for patients with intellectual disabilities. J Intellect Disabil Res (2022) 66 323-331. PMID: 35040230 18. Smith FJ, Coetzee JF, Jurgens FX, Becker PJ. An observational study is the induction of anesthesia with Propofol according to the adjusted ideal body mass in obese and non-obese patients. South Afr J Anaesth Analg (2019) 25 6-11 19. Ding T, Hu YL. Comparison of etomidate and propofol-mediated anesthesia induction followed by intubation and sevoflurane maintenance during ERCP in obese patients. Am J Transl Res (2021) 13 9853-9859. PMID: 34540121 20. Alam L, Khattak MA, Alam M. Safety of balanced propofol and midazolam in upper gastrointestinal endoscopy for sedation in cirrhotic patients. J Pak Med Assoc (2021) 71 64-68. PMID: 33484521 21. Kumar D, Afshan G, Zubair M, Hamid M. Isoflurane alone versus small dose propofol with isoflurane for removal of laryngeal mask airway in children-a randomized controlled trial. J Pak Med Assoc (2019) 69 1596-1600. PMID: 31740862 22. Hamid HM, Masud S, Waseem A, Bashir A, Samreen A, Ashraf S. Effect of Priming Principle on Propofol Dose Required to Induce General Anesthesia. J Bah Uni Med Dent Coll (2020) 10 40-43 23. Ahmed SA, Nega MH, Tawuye HY, Mustofa SY. Incidence and factors associated with post-induction hypotension among adult surgical patients: A prospective follow-up study. Int J Surg. Open (2022) 100565 24. Budic I, Jevtovic Stoimenov T, Pavlovic D, Marjanovic V, Djordjevic I, Stevic M. Clinical Importance of Potential Genetic Determinants Affecting Propofol Pharmacokinetics and Pharmacodynamics. Front Med (Lausanne) (2022) 28 809393 25. Poma MAM, Ribeiro Junior HL, Costa EA, Paier CRK, Brasil LL, Lima LB. Effect of CYB2B6 (c.516G>T), CYP2C9(c.1075A>C), and UGT1A9 (c.98T>C) polymorphisms on propofol pharmacokinetics in patients submitted to colonoscopy: a cohort study. Postgrad Med J (2022) 15
--- title: Clinical effects of Chemotherapy combined with Immunotherapy in patients with advanced NSCLC and the effect on their nutritional status and immune function authors: - Jin Jiao - Wen-wen Li - Yan-hong Shang - Xiao-fang Li - Meng Jiao journal: Pakistan Journal of Medical Sciences year: 2023 pmcid: PMC10025708 doi: 10.12669/pjms.39.2.6365 license: CC BY 3.0 --- # Clinical effects of Chemotherapy combined with Immunotherapy in patients with advanced NSCLC and the effect on their nutritional status and immune function ## Abstract ### Objectives: To evaluate the clinical effects of chemotherapy combined with immunotherapy in patients with advanced non-small-cell lung cancer (NSCLC) and the effect on their nutritional status and immune function. ### Methods: Total 120 patients with advanced NSCLC admitted to Affiliated Hospital of Hebei University from May 2019 to October 2021 were randomly divided into two groups ($$n = 60$$, respectively). Patients in the control group were treated by chemotherapy with cisplatin-paclitaxel (TP) alone: 120 mg/m2 paclitaxel was used on d1; and 25mg/m2 cisplatin (CDDP) was used for more than two hour, once every 14 days, for three consecutive three cycles. Patients in the study group were additionally given 200 mg sindilizumab by intravenous drip, once every three weeks. The contrastive analysis of clinical effects, the incidence of adverse reactions, improvement of the nutrient index and the changes in levels of CD3+, CD4+, CD8+, and CD4+/CD8+ in T-lymphocyte subsets was performed between the two groups. ### Result: The overall response rate (ORR) was $80\%$ and $61\%$ in the study group and the control group, respectively; and the difference was statistically significant ($$p \leq 0.03$$); the contrast analysis of the incidence of post-treatment adverse drug reactions (ADRs) in patients in the two groups suggested that the incidence of adverse reactions was $33.3\%$ and $45\%$ in the study group and the control group, respectively; and the difference was not statistically significant ($$p \leq 0.19$$). After the treatment, the improvement of hemoglobin, albumin, serum iron and ferritin levels in the study group was more significant than that in the control group; and the difference was statistically significant ($p \leq 0.05$). After the treatment, the levels of CD3+, CD4+ and CD4+/CD8+ in the study group were much higher than those in the control group; and the difference was statistically significant ($p \leq 0.05$). ### Conclusion: Chemotherapy combined with immunotherapy is effective in treating patients with advanced NSCLC without increasing the incidence of adverse reactions, and can significantly improve their nutritional status and T-lymphocyte function. This therapeutic regimen is of much higher clinical value than the chemotherapy-only regimen. ## INTRODUCTION The data from the epidemiological survey showed that1 lung cancer is still the malignant tumor with the highest morbidity and mortality worldwide. Non-small cell lung cancer (NSCLC) accounts for more than $85\%$.2 What’s worse, more than $70\%$ of patients with NSCLC are in an advanced stage when diagnosed and cannot receive operative treatment.3 The risk factors of NSCLC include smoking, environmental factors and genetic factors. NSCLC has no specific clinical manifestations in the early stage. Therefore, most of the confirmed cases found in clinical work are in an advanced stage, which imposes great impacts on patients’ health and life.4 As a common palliative therapy for patients with advanced NSCLC, chemotherapy realizes clinical treatment by killing or inhibiting tumor cells with chemotherapy drugs. Clinically, patients with NSCLC are often treated by platinum-based chemotherapy regimens. However, there is a high risk of recurrence after the treatment.5 Moreover, patients receive chemotherapy repeatedly, which reduces the immune function of the body and increases relevant adverse reactions. Meanwhile, the nutrient depletion status of tumors also leads to the abnormal nutritional status of patients and lower treatment tolerance, which ultimately results in the reduction of the efficacy of chemotherapy.6 The emergence of immunotherapy has completely changed the situation of advanced NSCLC. Multiple clinical trials have proved the safety and feasibility of the combination of chemotherapy and immunotherapy.7 The programmed cell death receptor-1 (PD-1) and its ligand of immune checkpoint proteins can interact with anti-PD-1 antibodies, improving the objective response rate of cancer patients. Sindilizumab is a humanized monoclonal antibody against PD-1 and can be used for the treatment of recurrent or refractory tumors after second-line systemic chemotherapy.8 The combination of TP chemotherapy with sindilizumab immunotherapy had good clinical effects in patients with advanced NSCLC. Our objective was to evaluate the clinical effects of chemotherapy combined with immunotherapy in patients with advanced non-small-cell lung cancer (NSCLC) ## METHODS One hundred twenty patients with advanced NSCLC admitted to our hospital from May, 2019 to October, 2021 were selected and randomly divided into two groups ($$n = 60$$, respectively). There were 38 male and 22 female patients aged 46~77 (average 62.47±11.92 years) in the study group. There were 35 male and 25 female patients aged 45~77 (average 62.08±10.97 years) in the control group. There was no significant difference in the general data of patients between the two groups. However, there still was comparability between the two groups (Table-I). The study was approved by the Institutional Ethics Committee of Affiliated Hospital of Hebei University (No.:2019Q054; dated: 1st March, 2019), and written informed consent was obtained from all participants. Table-I \| Index \| Study group \| Control group \| t/χ2 \| P \| \| --- \| --- \| --- \| --- \| --- \| \| Age (y) \| 62.47±11.92 \| 62.08±10.97 \| 0.19 \| 0.85 \| \| Male (%) \| 38 (65%) \| 35 (62.5%) \| 0.31 \| 0.57 \| \| Clinical stage \| \| \| 0.21 \| 0.65 \| \| III \| 47 (70%) \| 49 (75%) \| \| \| \| IV \| 13 (30%) \| 11 (25%) \| \| \| \| Tumor location \| \| \| 0.14 \| 0.71 \| \| Peripheral \| 38 (67.5%) \| 36 (62.5%) \| \| \| \| Central \| 22 (32.5%) \| 24 (37.5%) \| \| \| \| Pathological type \| \| \| \| \| \| Adenocarcinoma \| 32 (57.5%) \| 35 (55%) \| 0.3 \| 0.58 \| \| Squamous cell carcinoma \| 16 (35%) \| 14 (40%) \| 0.18 \| 0.67 \| \| Miscellaneous \| 12 (7.5%) \| 11 (5%) \| 0.05 \| 0.82 \| ## Inclusion criteria: Patients who met the diagnostic criteria for advanced NSCLC;9Patients with chest imaging (CT or MRI) showing the presence of lesions that can be accurately measured;10Patients who were in a good physical condition instead of a dependent state (KPS>80);11Patients between 40 and 77 years of age;Patients and their families who had good compliance with treatment and were willing and able to cooperate with the completion of this study;Patients who had no contraindications for drugs used in this study;Patients who had signed the informed consent. ## Exclusion Criteria: Patients with the poor general condition and unstable vital signs;Patient complicated with other systemic malignancy;Patients complicated with severe organic disease;Patients who were allergic or intolerant to any drug involved in this study;Patients who were unable to cooperate to complete this study;Patients who had taken any drug that affects the study, such as immunosuppressor and hormone. ## Therapies After admission, patients in both groups completed relevant laboratory examinations, including blood cell analysis, coagulation function, liver function and kidney function. Patients with abnormal indexes were treated accordingly. Hydration was performed one day before chemotherapy, patients in the control group were treated with TP regimen alone, specifically as follows: 120 mg/m2 paclitaxel on d1; 25 mg/m2 CDDP, for more than two hours, tested during chemotherapy, with antiemetic, liver and kidney function protection, rehydration and other therapies applied, once every 14 days, for three consecutive cycles.12 Patients in the study group were additionally given 200 mg sindilizumab by intravenous drip, once every three weeks.13 ## Observation Indicators: Efficacy evaluation: Tumor efficacy evaluation was performed once every two cycles after treatment; and patients were observed for three consecutive months. Methods of clinical efficacy determination14: complete remission (CR): The lesions disappear completely and the tumor markers return to normal for more than four weeks, which including carcinoembryonic antigen (CEA), neuron specific enolase (NSE) and cytokeratin 19 serum fragment 21-1 (CYFRA21-1); partial remission (PR): The volume of lesions decreases by more than $30\%$ for more than four weeks; stable (SD): The volume of lesions decreases by < $30\%$ or increases by < $30\%$; progress (PD): The volume of lesions increases by more than $30\%$ or new lesions appear; overall response rate (RR) = CR+PR%; Evaluation of ADRs: ADRs occurred in both groups within one month after medication, including fever, bone marrow suppression, gastrointestinal reactions, liver and kidney dysfunction and other adverse reactions, were recorded; Improvement of nutritional status: The fasting blood in the morning was sampled before and after the treatment, respectively; and the changes in such nutritional indexes as hemoglobin, albumin, serum iron and ferritin before and after the treatment were compared and analyzed. 4) Analysis of immune status: The fasting blood in the morning was sampled before and after the treatment respectively to detect the levels of CD3+, CD4+, CD8+ and CD4+/CD8+ in T-lymphocyte subsets; and the contrastive analysis of the differences before and after the treatment between the two groups was performed. ## Statistical Analysis: The software SPSS 20.0 was used for the statistical analysis of all data. The measurement data were expressed as (±S). Independent samples t-test was used for the data analysis between the two groups. Paired t-test was applied to intra-group data analysis. χ2 test was used for rate comparison. A p-value of <0.05 was considered statistically significant. ## RESULTS The contrast analysis of the effects between the two groups is shown in Table-II. It suggested that the ORR was $80\%$ in the study group and $61\%$ in the control group. The effects in the study group were evidently better than those in the control group, and the difference was statistically significant ($$p \leq 0.03$$). Table-II \| Group \| CR \| PR \| SD \| PD \| ORR \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| Study group \| 23.0 \| 25.0 \| 7.0 \| 5.0 \| 48 (80%) \| \| Control group \| 21.0 \| 16.0 \| 14.0 \| 9.0 \| 37 (61%) \| \| c2 \| \| \| \| \| 4.88 \| \| P \| \| \| \| \| 0.03 \| The contras analysis of the incidence of ADRs after the treatment between the two groups suggested that the incidence of adverse reactions was $33.3\%$ in the study group and $45\%$ in the control group; the difference was not statistically significant ($$p \leq 0.19$$). ( Table-III) Table-III \| Group \| Fever \| Bone marrow suppression \| Liver dysfunction \| Kidney dysfunction \| Gastrointestinal reactions \| Incidence \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Study group \| 5.0 \| 3.0 \| 4.0 \| 5.0 \| 3.0 \| 20 (33.3%) \| \| Control group \| 6.0 \| 6.0 \| 4.0 \| 7.0 \| 4.0 \| 27 (45%) \| \| c2 \| \| \| \| \| \| 1.71 \| \| P \| \| \| \| \| \| 0.19 \| After the treatment, the levels of nutritional indexes including hemoglobin, albumin, serum iron and ferritin in the study group and the control group were higher than those before the treatment, which indicated that patients’ nutritional status was improved after chemotherapy; the improvement in the study group was greater than that in the control group; and the difference was statistically significant ($$p \leq 0.00$$) (Table-IV). Table-IV \| Group \| Hemoglobin (g/L)* \| Albumin (g/L)* \| Serum iron (mmol/L)* \| Ferritin (ug/L)* \| \| --- \| --- \| --- \| --- \| --- \| \| Study group \| 6.68±2.17 \| 5.03±1.75 \| 7.43±2.15 \| 5.42±2.30 \| \| Control group \| 4.36±1.49 \| 3.57±1.38 \| 5.32±2.04 \| 3.19±1.74 \| \| t \| 6.23 \| 5.07 \| 5.51 \| 5.98 \| \| p \| 0.00 \| 0.00 \| 0.00 \| 0.00 \| There was no significant difference in the pre-treatment levels of CD3+, CD4+, CD8+ and CD4+/CD8+ between the two groups ($P \leq 0.05$). The post-treatment levels of CD3+, CD4+, CD8+ and CD4+/CD8+ in the study group were significantly higher than those in the control group; and the difference was statistically significant ($p \leq 0.05$). However, the changes in the levels of CD8+ were not obvious ($$p \leq 0.96$$) (Table-V). Table-V \| Index \| Unnamed: 1 \| Study group \| Control group \| t \| p \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| CD3+ (%) \| Pre-treatment \| 44.73±8.75 \| 44.32±8.25 \| 0.26 \| 0.79 \| \| CD3+ (%) \| Post-treatment* \| 50.31±8.53 \| 46.24±8.07 \| 2.67 \| 0.01 \| \| CD4+ (%) \| Pre-treatment \| 25.40±4.51 \| 25.53±4.82 \| 0.15 \| 0.88 \| \| CD4+ (%) \| Post-treatment* \| 38.11±7.39 \| 34.27±7.84 \| 2.62 \| 0.01 \| \| CD8+ (%) \| Pre-treatment \| 21.85±4.13 \| 21.75±4.31 \| 0.13 \| 0.89 \| \| CD8+ (%) \| Post-treatment \| 22.62±5.14 \| 22.57±5.07 \| 0.05 \| 0.96 \| \| CD4+/CD8+ \| Pre-treatment \| 1.38±0.41 \| 1.35±0.52 \| 0.35 \| 0.73 \| \| CD4+/CD8+ \| Post-treatment* \| 1.96±0.51 \| 1.46±0.27 \| 6.71 \| 0.0 \| ## DISCUSSION Lung cancer is the malignant tumor with the highest morbidity and mortality in China. Approximately $85\%$ of patients with lung cancer were diagnosed with NSCLC. Since there are no specific clinical symptoms in the early stage, most patients with NSCLC are in an advanced stage when diagnosed. These patients are usually treated by chemotherapy. For some advanced-stage patients, although chemotherapy can improve prognosis and survival rate, adverse reactions are also obvious. The efficacy of chemotherapy in patients with NSCLC in the middle and advanced stages has been in a bottleneck period.15 In recent years, the effect of the immune mechanism in tumorigenesis and progression has been clarified. Accordingly, targeted therapy is also becoming clear. Targeted therapy can significantly improve the prognosis of some patients with NSCLC, but some patients treated by targeted therapy have drug resistance problems.16 Furthermore, there are no corresponding targeted drugs for patients with NSCLC with negative driver genes17, making new therapy a clinical problem to be solved urgently. Immunotherapy has greatly changed the therapy for newly diagnosed advanced NSCLC. At present, more and more lung cancer diagnosis and treatment guidelines have recommended immune drugs in the treatment of advanced NSCLC, so as to benefit patients.18 Immunotherapy combined with chemotherapy has achieved good results and has improved the progression-free survival of these patients.19 Proto et al.20 held that PD-1 inhibitors combined with platinum-based chemotherapy have become an effective first-line therapy. PD-1 is distributed on the surface of immune cells, while programmed death-1 ligand (PDL-1) is distributed on the surface of tumor cells. The combination of the two results in the immune escape of tumor cells by activating the signal pathway in immune cells.21 PD-/PDL-1 blockade can improve the tumor infiltrating lymphocytes (TIL) killing effect.22 The combination and improvement of immunotherapy and chemotherapy have significantly improved the prognosis of patients with NSCLC. The immune checkpoint blockade with PD-1 and PD-L1 antibodies can produce long-lasting reactions of clinical significance in patients with advanced NSCLC. The more extensive use of these drugs can improve the nutritional status and survival rate of patients with advanced lung cancer.23 Immunotherapy combined with platinum-based chemotherapy is showing increasing benefits in the treatment of patients with advanced NSCLC.24 The study of Zhang et al.25 proved that this regimen can improve the function of T lymphocytes and restore their anti-tumor effect. Jiang et al.26 believed that the combination shows controlled toxicity and encourages anti-tumor activity. The results of a multi-center study showed that27: When the combination of chemotherapy and immunotherapy was adopted, only about $20\%$ of patients withdrew from the study due to adverse reactions, while most patients could complete the whole course of treatment and achieve good results. Besides, the study of Leonetti28 also proved the significant clinical benefits of immune checkpoint inhibitors combined with chemotherapy in treating NSCLC. This study demonstrated that the ORR of chemotherapy combined with immunotherapy for advanced NSCLC was $80\%$, while the ORR in the control group was $61\%$ ($$p \leq 0.03$$); the incidence of adverse reactions was $33.3\%$ and $45\%$ in the study group and the control group, respectively, and the difference was not statistically significant ($$p \leq 0.19$$); the post-treatment levels of CD3+, CD4+ and CD4+/CD8+ in the study group increased significantly, and the difference was statistically significant ($p \leq 0.05$). The results showed that the cellular immune status of patients was significantly improved after combined immunotherapy and that the clinical effect was significant without significantly increasing adverse reactions. The common adverse reactions during chemotherapy include gastrointestinal reactions, nutritional deficiencies and impaired immune function29 Improving patients’ immune status is of great significance for the nutritional status of cancer patients.30 This study also proved that the improvement of the post-treatment levels of hemoglobin, albumin, serum iron and ferritin in the study group was more significant than that in the control group; and the difference was statistically significant. ## Limitations: It includes small sample size and short follow-up period. In the future clinical work, the sample size and follow-up period will be further increased and the effect of different therapies on the long-term effect and survival of patients further improved, so as to evaluate the benefits of this regimen to patients in a more comprehensive manner. ## CONCLUSION In conclusion, chemotherapy combined with immunotherapy is effective in treating patients with advanced NSCLC, and can significantly improve their nutritional status and T-lymphocyte function. There was no obvious increase in the incidence of adverse reactions. Therefore, this therapeutic regimen is of much higher clinical value than the chemotherapy-only regimen. ## Authors’ Contributions: JJ and WL designed this study, prepared this manuscript, are responsible and accountable for the accuracy and integrity of the work. MJ collected and analyzed clinical data. YS and XL participated in acquisition, analysis, or interpretation of data and draft the manuscript. ## References 1. Arbour KC, Riely GJ. Systemic Therapy for Locally Advanced and Metastatic Non-Small Cell Lung Cancer: A Review. JAMA (2019) 322 764-774. PMID: 31454018 2. Alexander M, Kim SY, Cheng H. Update 2020: Management of Non-Small Cell Lung Cancer. Lung (2020) 198 897-907. PMID: 33175991 3. Passiglia F, Bertolaccini L, Del Re M, Facchinetti F, Ferrara R, Franchina T. Diagnosis and treatment of early and locally advanced non-small-cell lung cancer: The 2019 AIOM (Italian Association of Medical Oncology) clinical practice guidelines. Crit Rev Oncol Hematol (2020) 148 102862. PMID: 32062311 4. Tagliamento M, Rijavec E, Barletta G, Biello F, Rossi G, Grossi F. CIMAvax-EGF, a therapeutic non-small cell lung cancer vaccine. Expert Opin Biol Ther (2018) 18 829-835. PMID: 29936901 5. Chen Q, Ji X, Zhou X, Shi Q, Yu H, Fu H. Clinical observation of docetaxel or gemcitabine combined with cisplatin in the chemotherapy after surgery for stage II-III non-small cell lung cancer. Pak J Med Sci (2015) 31 1095-1098. PMID: 26648993 6. Duma N, Santana-Davila R, Molina JR. Non-Small Cell Lung Cancer: Epidemiology, Screening, Diagnosis, and Treatment. Mayo Clin Proc (2019) 94 1623-1640. PMID: 31378236 7. Broderick SR. Adjuvant and Neoadjuvant Immunotherapy in Non-small Cell Lung Cancer. Thorac Surg Clin (2020) 30 215-220. PMID: 32327180 8. Shafique M, Tanvetyanon T. Immunotherapy alone or chemo-immunotherapy as front-line treatment for advanced non-small cell lung cancer. Expert Opin Biol Ther (2019) 19 225-232. PMID: 30657338 9. Duma N, Santana-Davila R, Molina JR. Non-Small Cell Lung Cancer: Epidemiology, Screening, Diagnosis, and Treatment. Mayo Clin Proc (2019) 94 1623-1640. PMID: 31378236 10. Balata H, Fong KM, Hendriks LE, Lam S, Ostroff JS, Peled N. Prevention and Early Detection for NSCLC: Advances in Thoracic Oncology 2018. J Thorac Oncol (2019) 14 1513-1527. PMID: 31228621 11. Barbetta C, Allgar V, Maddocks M, Ribeiro C, Wilcock A, Currow DC. Australia-modified Karnofsky Performance Scale and physical activity in COPD and lung cancer: an exploratory pooled data analysis [published online ahead of print 2019 Jul 11]. BMJ Support Palliat Care (2019) 12. Chisaki Y, Nakamura N, Yano Y. Time-Series Modeling and Simulation for Comparative Cost-Effective Analysis in Cancer Chemotherapy: An Application to Platinum-Based Regimens for Advanced Non-small Cell Lung Cancer. Biol Pharm Bull (2017) 40 73-81. PMID: 28049952 13. Tao X, Li N, Wu N, He J, Ying J, Gao S. The efficiency of 18F-FDG PET-CT for predicting the major pathologic response to the neoadjuvant PD-1 blockade in resectable non-small cell lung cancer. Eur J Nucl Med Mol Imaging (2020) 47 1209-1219. PMID: 32043180 14. Wang S, Zimmermann S, Parikh K, Mansfield AS, Adjei AA. Current Diagnosis and Management of Small-Cell Lung Cancer. Mayo Clin Proc (2019) 94 1599-1622. PMID: 31378235 15. Hao Z, Tian C, Yang F, Zhang J. Correlation between expression of epidermal growth factor receptor and adverse reactions after chemotherapy of advanced non-small-cell lung cancer. Pak J Med Sci (2015) 31 1115-1120. PMID: 26648997 16. Tabchi S, Kassouf E, Rassy EE, Kourie HR, Martin J, Campeau MP. Management of stage III non-small cell lung cancer. Semin Oncol (2017) 44 163-177. PMID: 29248128 17. Herbst RS, Morgensztern D, Boshoff C. The biology and management of non-small cell lung cancer. Nature (2018) 553 446-454. PMID: 29364287 18. Osmani L, Askin F, Gabrielson E, Li QK. Current WHO guidelines and the critical role of immunohistochemical markers in the subclassification of non-small cell lung carcinoma (NSCLC): Moving from targeted therapy to immunotherapy. Semin Cancer Biol (2018) 52 103-109. PMID: 29183778 19. Rocco D, Della Gravara L, Battiloro C, Gridelli C. The role of combination chemo-immunotherapy in advanced non-small cell lung cancer. Expert Rev Anticancer Ther (2019) 19 561-568. PMID: 31188040 20. Proto C, Ferrara R, Signorelli D, Lo Russo G, Galli G, Imbimbo M. Choosing wisely first line immunotherapy in non-small cell lung cancer (NSCLC): What to add and what to leave out. Cancer Treat Rev (2019) 75 39-51. PMID: 30954906 21. Reck M, Borghaei H, O'Byrne KJ. Nivolumab plus ipilimumab in non-small-cell lung cancer. Future Oncol (2019) 15 2287-2302. PMID: 31066582 22. He Y, Yu H, Rozeboom L, Rivard CJ, Ellison K, Dziadziuszko R. LAG-3 Protein Expression in Non-Small Cell Lung Cancer and Its Relationship with PD-1/PD-L1 and Tumor-Infiltrating Lymphocytes. J Thorac Oncol (2017) 12 814-823. PMID: 28132868 23. Patel SA, Weiss J. Advances in the Treatment of Non-Small Cell Lung Cancer: Immunotherapy. Clin Chest Med (2020) 41 237-247. PMID: 32402359 24. Kim SY, Halmos B. Choosing the best first-line therapy: NSCLC with no actionable oncogenic driver. Lung Cancer Manag (2020) 9 LMT36. PMID: 32774467 25. Zhang L, Mai W, Jiang W, Geng Q. Sintilimab: A Promising Anti-Tumor PD-1 Antibody. Front Oncol (2020) 10 594558. PMID: 33324564 26. Jiang HP, Zheng YL, Qian J, Mao CY, Xu X, Li L. Safety and efficacy of sintilimab combined with oxaliplatin/capecitabine as first-line treatment in patients with locally advanced or metastatic gastric/gastroesophageal junction adenocarcinoma in a phase Ib clinical trial. BMC cancer (2020) 20 760. PMID: 32795349 27. Provencio M, Nadal E, Cobo M, Insa A, Calvo V. Neoadjuvant chemo/immunotherapy for the treatment of stages IIIA resectable non-small cell lung cancer (NSCLC): A phase II multicenter exploratory study-NADIM study-SLCG. J Clin Oncol (2018) 36 8521-8521 28. Leonetti A, Wever B, Mazzaschi G, Assaraf YG, Rolfo C, Quaini F. Molecular basis and rationale for combining immune checkpoint inhibitors with chemotherapy in non-small cell lung cancer. Drug Resist Updat (2019) 46 100644. PMID: 31585395 29. Hebuterne X, Lemarie E, Michallet M, de Montreuil CB, Schneider SM, Goldwasser F. Prevalence of malnutrition and current use of nutrition support in patients with cancer. J Parenter Enteral Nutr (2014) 38 196-204 30. Sanchez-Lara K, Ugalde-Morales E, Motola-Kuba D, Green D. Gastrointestinal symptoms and weight loss in cancer patients receiving chemotherapy. Br J Nutr (2013) 109 894-897. PMID: 22691288
--- title: Comparison of residual silicone oil index after removal of silicone oil with fluid-air versus oil-fluid exchange authors: - Amna Rizwan - Rana Muhammad Mohsin Javaid - Sidrah Latif - Muhammad Suhail Sarwar - Asad Aslam Khan journal: Pakistan Journal of Medical Sciences year: 2023 pmcid: PMC10025723 doi: 10.12669/pjms.39.2.6243 license: CC BY 3.0 --- # Comparison of residual silicone oil index after removal of silicone oil with fluid-air versus oil-fluid exchange ## Abstract ### Objectives: To compare the effectiveness of fluid-air exchange with silicone oil-fluid exchange in reducing the residual silicone oil (SO) droplets after the removal of SO. ### Methods: This was a prospective, quasi-experimental study conducted from October 2021 to February 2022 at Eye Unit-III, COAVS, Mayo Hospital, Lahore. Sixty-one patients with siliconized eyes underwent removal of SO with two different techniques and were divided into fluid-air exchange and oil-fluid exchange groups. To quantify the residual silicone droplets objectively, B-scan echographic images were analyzed within seven days of surgery. Silicone oil index (SOI) which is the amount of residual SO droplets/vitreal area in the images was calculated with the help of imagej software. ### Results: The residual SOI of the fluid-air exchange group (0.99 ± $1.76\%$) was significantly lower than the oil-fluid exchange group (3.25 ± $3.85\%$). The SOI is positively correlated with the duration of tamponade, preoperative intraocular- pressure and axial length. Persistent IOP elevation post-operatively was seen in $16.67\%$ individuals in the fluid-air exchange group and $54.8\%$ individuals in the oil-fluid exchange group. ### Conclusion: Fluid-air exchange group was found to be superior in reducing residual SO droplets than the oil-fluid exchange group. ## INTRODUCTION Silicone oil (SO) was first introduced as an internal tamponade in vitreoretinal surgery in the 1960s.1 It has since then developed into a valued method used in retinal detachment (RD) surgery. SO is used in complex rhegmatogenous retinal detachment (RRD) especially with severe proliferative vitreoretinopathy (PVR), giant retinal tears, proliferative diabetic retinopathy (PDR), traumatic RD, endophthalmitis, macular hole surgery, RD associated with choroidal coloboma and complicated pediatric RDs.2-4 Complications of SO include glaucoma, chronic hypotony, cataract formation, recurrent RD, SO emulsification, keratopathy and migration of SO in the anterior chamber.5 In practice, it is difficult to make the judgment call to remove SO. As a reference, in the Silicone study, SO was removed after a minimum of eight weeks. In general, however, SO is typically removed within six months following surgery.6 Various active or passive methods may be employed for removing SO, with active ones generally preferred.7,8 Complete removal of silicone oil (ROSO) is seldom possible. Emulsified droplets adhere to the ciliary recess, zonules and posterior aspects of the iris. Many techniques are being used to remove emulsified silicone oil droplets. The suction method used can involve active aspiration or passive perfusion.9 One method is by silicone oil-fluid exchange (OFX).9 Other active method is based on repeated fluid-air exchange (FAX) cycles. In this method, the air replaces the SO in the vitreous cavity.9 In practice, however, several SO droplets are generally seen post-procedure. The relative efficacy of these methods is traditionally quantified either in terms of surgical time or via post-operative slit-lamp examination which is a rather crude indicator of the removal. B-scan ultrasonography, on the other hand, provides us with a reliable quantitative tool to measure the efficacy of various methods. B-scan ultrasonography employs Rayleigh scattering to exaggerate the SO residue and, thus, enables us to make a good estimate of the residual SO droplets.10 This study investigated the efficacy of FAX technique vs OFX technique in reducing residual SO droplets using B-scan ultrasonography. ## METHODS A quasi-experimental study was conducted at Eye Unit-III, COAVS, Mayo Hospital, Lahore from October 2021 to February 2022 after approval from the ethical review board (No. COAVS/$\frac{1106}{2021}$, Date: 12-10-2021). A total of 61 patients were included by non-probability convenient sampling by using the level of significance as $95\%$ and power of test as $80\%$.10 All patients who underwent surgery for ROSO were included except patients with corneal opacities, SO tamponade of more than four years, emulsification of silicone oil, any event of post-operative vitreous hemorrhage, or low-quality ultrasonographic images. A written informed consent with demographic information was collected from each patient. Included eyes underwent slit-lamp examination of the anterior segment, intraocular pressure (IOP) measurements with a Goldmann applanation tonometer, fundus evaluation, and pre-operative axial length (AL) measurements. History, diagnosis of the disease, age, gender, the status of lens and duration of the SO tamponade were noted. ## Surgical Technique: All surgeries were done by the same surgeon. Three port sclerotomy with 23-gauge trocars were made with infusion port placed inferotemporally and two superior ports for aspiration. B.E.S (Balanced Electrolyte Solution) was allowed to replace globe volume as SO was aspirated with a bottle height of 80 centimeters above the eye. The bulk of the SO was removed from the sclerotomy site with a 10 ml syringe by pulling the plunger of the syringe to the end to create maximum negative pressure. After the bulk removal, patients were divided into two groups; a FAX group in which two to three fluid-air exchange cycles were done, and an OFX group in which the posterior segment was washed continually with infusion fluid for at least three minutes. After both techniques, the fundus was examined. The anterior chamber was washed in both techniques if required. Sutures were placed to close the sclerotomies. B-scan ultrasonography was performed (by the same ophthalmologist) with a standard ultrasonographic device within seven days of surgery or when the air was absorbed. To remove observer bias, a total of three B-scan images were taken and the mean was used. Patients were further asked to look either towards right or left shoulder to avoid the lenticular and intraocular lens shadows. B-scan machine used was compact touch, with 90 gain, zoom 170 and time-gain compensation (TGC) zero. To quantify the residual SO droplets objectively, a binarization method was applied to the B-scan images using color threshold adjustment as shown in Fig.1. These images were assessed using ImageJ software (ImageJ version 1.47, National Institutes of Health, Bethesda, MD; available at: http://imagej.nih.gov/ij/). The ratio of the sum of the SO droplet areas to that of the vitreous cavity is defined as silicone oil index (SOI) and was calculated using the formula.10 Fig.1: Representation of image processing using imageJ software to quantify residual SO droplets. a. Ultrasound B-scan image of a patient. b. Binarization of the image to highlight signals from the residual SO droplets in “color threshold” mode. c. vitreous cavity area was demarcated. d. Image showing number and area of residual SO droplets. ## Statistical analysis: SPSS 26 was used to analyze the data. Mean and standard deviation were calculated for quantitative data while percentages and frequencies were calculated for qualitative data. The relationships between SOI and the different ocular parameters were determined by Pearson and spearman’s correlation tests. The SOIs between the OFX group and the FAX group were compared using the Mann–Whitney U test. A p-value ≤0.05 was considered significant. ## RESULTS A total of 61 individuals including 45 males ($73.8\%$) and 16 females ($26.2\%$) were included, with a mean age of 44.72 ± 17.06 years. Out of the 61 eyes, 45 ($73.8\%$) had been treated for an RRD, 13 eyes ($21.3\%$) for a PDR and three ($4.9\%$) for endophthalmitis. Twenty-four individuals ($39.3\%$) were diabetic, and 21 ($34.4\%$) were hypertensive. The mean AL was 24.27 ± 2.22 mm, and 16 eyes were phakic, 40 were pseudo-phakic and five were aphakic. Demographic findings of FAX and OFX groups were shown in Table-I. Table-I \| Variables \| Fluid-Air exchange (n = 30) \| Fluid-Air exchange (n = 30).1 \| Fluid-Air exchange (n = 30).2 \| Fluid-Air exchange (n = 30).3 \| Oil-Fluid exchange (n = 31) \| Oil-Fluid exchange (n = 31).1 \| Oil-Fluid exchange (n = 31).2 \| P-value \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Age \| 44.50 ± 15.06 \| 44.50 ± 15.06 \| 44.50 ± 15.06 \| 44.50 ± 15.06 \| 44.94 ± 19.04 \| 44.94 ± 19.04 \| 44.94 ± 19.04 \| 0.608a \| \| Gender (M/F) \| 21/9 \| 21/9 \| 21/9 \| 21/9 \| 24/7 \| 24/7 \| 24/7 \| 0.510b \| \| Preoperative IOP \| 24.00 ± 7.10 \| 24.00 ± 7.10 \| 24.00 ± 7.10 \| 24.00 ± 7.10 \| 28.39 ± 9.83 \| 28.39 ± 9.83 \| 28.39 ± 9.83 \| 0.109a \| \| AL \| 23.61 ± 1.72 \| 23.61 ± 1.72 \| 23.61 ± 1.72 \| 23.61 ± 1.72 \| 24.92 ± 2.49 \| 24.92 ± 2.49 \| 24.92 ± 2.49 \| 0.036a \| \| SOI \| 0.99 ± 1.76 \| 0.99 ± 1.76 \| 0.99 ± 1.76 \| 0.99 ± 1.76 \| 3.25 ± 3.85 \| 3.25 ± 3.85 \| 3.25 ± 3.85 \| 0.004a \| \| SOI Correlations \| Preoperative IOP \| Pearson’s r= 0.709 \| Pearson’s r= 0.709 \| P<0.001 \| Preoperative IOP \| Pearson’s r= 0.533 \| Pearson’s r= 0.533 \| P=0.002 \| \| SOI Correlations \| Duration of tamponade \| Spearman’s r = 0.496 \| Spearman’s r = 0.496 \| P=0.005 \| Duration of tamponade \| Spearman’s r = 0.385 \| Spearman’s r = 0.385 \| P=0.033 \| \| SOI Correlations \| Axial length \| Pearson’s r = 0.265 \| Pearson’s r = 0.265 \| P=0.157 \| Axial length \| Pearson’s r = 0.394 \| Pearson’s r = 0.394 \| P=0.028 \| \| SOI Correlations \| Postoperative IOP \| Pearson’s r = 0.787 \| Pearson’s r = 0.787 \| P<0.001 \| Postoperative IOP \| Pearson’s r = 0.533 \| Pearson’s r = 0.533 \| P=0.002 \| \| Visual Impairment \| Preoperative \| Preoperative \| Postoperative \| Postoperative \| Preoperative \| Preoperative \| Postoperative \| Postoperative \| \| Mild Moderate Blind \| 0 \| 0 \| 1 \| 1 \| 1 \| 1 \| 2 \| 2 \| \| Mild Moderate Blind \| 11 \| 11 \| 18 \| 18 \| 5 \| 5 \| 14 \| 14 \| \| Mild Moderate Blind \| 19 \| 19 \| 11 \| 11 \| 25 \| 25 \| 15 \| 15 \| Using the WHO criteria, preoperatively, one ($1.6\%$) individual had mild visual impairment (VA≤$\frac{6}{18}$), 16 ($26.2\%$) had moderate impairment (VA > $\frac{6}{18}$ to $\frac{6}{60}$), and 44 ($72.1\%$) individuals were blind (VA< $\frac{3}{60}$), as shown also in Table 1. The mean IOP was 26.23 ± 8.804 mmHg. SO tamponade duration was less than six months in nine individuals, up to one year in nine, and more than one year in 43 individuals. Postoperatively, three ($4.9\%$) individuals had mild impairment, 32 ($52.5\%$) had a moderate impairment, and 26 ($42.6\%$) were blind. The mean IOP was 22.05 ± 9.19 mm. Across the sample, the mean of SOI was 2.14 ± $3\%$. For the FAX group, the mean was 0.99 ± $1.76\%$. For the OFX group, the mean was 3.25 ± $3.85\%$. The Mann Whitney-U test was used to compare means of SOI in FAX and OFX groups and was found to be significant (mean rank 24.33 vs 37.45, respectively, $$p \leq 0.004$$). Pearson and spearman’s correlation was used for the SOI with various ocular parameters where appropriate, as shown in Table-I. It was moderately positively correlated with preoperative IOP (Pearson’s $r = 0.600$), AL (Pearson’s $r = 0.425$) and duration of SO tamponade (Spearman’s $r = 0.441$, $p \leq 0.001$). Neither gender nor age was associated with SOI. For the AFX group, SOI was strongly correlated with preoperative IOP (Pearson’s $r = 0.709$) and moderately correlated with duration of SO tamponade (Spearman’s $r = 0.496$, $$p \leq 0.005$$). It was not related to AL but was strongly correlated with postoperative IOP (Pearson’s $r = 0.787$). For the OFX group. SOI was moderately correlated with preoperative IOP (Pearson’s $r = 0.533$, AL (Pearson’s $r = 0.394$) and duration of tamponade (Spearman’s $r = 0.385$, $$p \leq 0.033$$). It was also related to post operative IOP (Pearson’s $r = 0.533$). Finally, multivariate analysis was used for those variables found to be significant in univariate analysis, as shown in Table-II. Table-II \| Variable \| Variable.1 \| Number of individuals \| Univariate analysis \| Univariate analysis.1 \| Univariate analysis.2 \| Multivariate analysis \| Multivariate analysis.1 \| Multivariate analysis.2 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Variable \| Variable \| Number of individuals \| \| \| \| \| \| \| \| Variable \| Variable \| Number of individuals \| Standardized Beta \| P-value \| 95% CI \| Standardized Beta \| P-value \| 95% CI \| \| 1 \| Duration \| \| \| \| \| \| \| \| \| 1 \| < 6 months* \| 9 \| Reference \| Reference \| Reference \| Reference \| Reference \| Reference \| \| 1 \| ≥ 6 months \| 52 \| 0.259 \| 0.044 \| 0.062 - 4.557 \| 0.085 \| 0.410 \| -1.067 - 2.577 \| \| 2 \| Axial length \| 61 \| 0.425 \| 0.001 \| 0.272 - 0.948 \| 0.241 \| 0.026 \| 0.044 - 0.647 \| \| 3 \| Group \| \| \| \| \| \| \| \| \| 3 \| FAX* \| 30 \| Reference \| Reference \| Reference \| Reference \| Reference \| Reference \| \| 3 \| OFX \| 31 \| 0.356 \| 0.005 \| 0.713 - 3.798 \| 0.161 \| 0.129 \| -0.304 - 2.339 \| \| 4 \| Preoperative IOP \| 61 \| 0.600 \| <0.001 \| 0.142 - 0.293 \| 0.476 \| <0.001 \| 0.096 - 0.249 \| In the FAX group, improvement in vision was noted in 14 ($46.67\%$) individuals and 16 ($53.3\%$) individuals showed no change in vision. In OFX, vision improvement was noted in 14 ($45.2\%$) individuals, no change in 16 ($51.6\%$) and one ($3.2\%$) showing decreased vision. Preoperative IOP was between 10-19 mmHg in 13 ($21.3\%$) individuals, 20 - 29 in 29 ($47.5\%$), and ≥30 in 19 ($31.1\%$) people. Post-op IOP was between 10 - 19 mmHg in 35 ($57.3\%$) individuals, 20 - 29 in 11 ($18\%$), and ≥ 30 in 15 ($24.5\%$) people. The mean IOP difference in the FAX group was -4.266 ± 2.362 mmHg, and for the OFX group, it was -4.09 ± 5. 461. Although they were not found to be significantly different. Persistent IOP elevation defined as at least 22 mm Hg post-operative was seen in five ($16.67\%$) individuals in the FAX group and 17 ($54.8\%$) individuals in the OFX group. ( Spearman’s $r = 0.397$, $$p \leq 0.002$$). ## DISCUSSION The principal result of our study is that that FAX technique is superior to OFX technique in reducing SO droplets as SOI (0.99 ± $1.76\%$) of FAX was less than OFX group (3.2 5 ± $3.85\%$). We have come to this conclusion after a careful analysis of residual silicone oil using B scan ultrasonography aided by ImajeJ software. In complicated RD surgery, the choice of using an SO tamponade has clear advantages which include a shorter recovery time and quicker visual rehabilitation, no restriction on air travel, and allowance of comfortable post-operative posture. The only drawback is the necessity of a follow-up procedure to remove the silicone oil.11 It is important to use the best available method to remove SO as incomplete removal of these small oil droplets can cause complications like secondary glaucoma, keratopathy, cataract, trabeculitis and chronic elevated IOP.12 Our principal result of the superiority of SO removal using FAX is consistent with the findings of Yu J et al in which another method (Coulter counter) to measure the number of droplets directly was used.13 Also consistent with our results, the superiority of FAX has been variously argued in other studies as well.9,14 In supine position, when air is injected, SO collects in the macula and forms a thin layer between the infusion fluid and air. A backflush cannula inserted at the level of this oil infusion fluid interface can easily extract all the SO. Another location for small SO residual particles is the retroiridial plane. Flow of air can dislodge these and make the removal possible via the backflush cannula. This mechanism, of course, is not available while using OFX. However, our findings are contrary to the results of Shiihara et al who concluded that the number of residual SO droplets increase after the FAX process.10 Shihaara et al mentioned that the primary difficulty with FAX is the removal of the thin layer of residual droplets formed at the macula. They suggest that this layer cannot be removed using a vitrectomy probe or a flute needle. However, we don’t see any reason for this difficulty if the backflush cannula is inserted at the appropriate level of the SO infusion fluid interface and, in fact, have found good results using FAX. One other advantage of FAX which we have not experienced in our study is that FAX cycles can allow an occult break to collect subretinal fluid, and this will reveal a subtle detachment that otherwise may have been recognized only postoperatively. In both of our groups, SOI is positively correlated with the duration of tamponade. While some studies have shown that a prolonged SO tamponade does not lead to ocular complications, the most common recommendation is to remove SO within three-six months. However, it is important to individually evaluate every patient before removing SO tamponade to ensure that the retina has properly attached.15,16 We did not find statistically significant correlation of SOI with AL in FAX group. On the other hand, in the OFX group, we found a statistically significant positive correlation of SOI with AL (Pearson’s $r = 0.394$). Shihara et al, who use OFX, have also reported a positive correlation of residual SOI with AL.17 These results seemingly suggest that FAX should be considered the preferred method in cases of eyes with a longer AL. In the OFX group, two patients ($6.45\%$) were excluded due to re-detachment. One of these patients had the SO tamponade for three months while the other for three years. FAX group did not resulted in any re-detachment whereas there was a small ($6.45\%$) rate of re-detachment in the OFX group. Re-detachment in cases of ROSO for various methods has been reported in the literature between $6\%$ to $34\%$.18 In the case of a FAX, Akkan et al reported a re-detachment rate of $5.5\%$.19 It is to be noted that 360° laser photocoagulation was applied in all cases either preoperatively or per-operatively during ROSO. In the FAX group, one patient was exluded from consideration who had developed endophthalmitis. This study did not find any correlation of SOI with the indication of SO tamponade, vision, diabetes, hypertension, or lens status. Some studies have shown that IOP returns to the normal range after ROSO. We, however, do see a persistent raised IOP (≥22 mm Hg) after ROSO in both groups. In the FAX group, this persistent raised IOP was seen in only $16.67\%$ of individuals whereas in the OFX group the same was seen in $54.8\%$ of individuals. This raised IOP may be caused by trabecular meshwork edema due to post-operative inflammation. Another reason may be the mechanical impact of infusion fluid during ROSO may split the SO droplets into much smaller drops, which are more likely to obstruct the trabecular meshwork.20 This study did not report any cases of post-op transient hypotony which is reported in the literature between $5\%$ to $40\%$ of the cases. This is likely because sutures were applied in all of our patients to close sclerostomies.21 Our study is the first in Pakistan to quantitatively measure the residual SO droplets while earlier studies gauged efficacy of SO removal techniques indirectly by looking at side effects resulting from residual SO. We have demonstrated that FAX is superior to OFX and would recommend it as the protocol of choice. ## Limitation The primary limitation of our study is a relatively small sample size. ## CONCLUSION Fluid-air exchange group was found to be superior in reducing residual SO droplets compared with the oil-fluid exchange group. Fluid-air exchange is the preferred method as it decreases residual SO droplets thereby decreasing SO related complications, resulting in less number of patients with a reported increased IOP and no case presenting with re-detachment. ## Author’s Contributions: AR: Conception and design of study, Data collection. RMJ: Manuscript writing. SL: Study design, Data collection. SS: Data analysis, Editing of manuscript. AAK: Revision of manuscript, responsible for integrity of study. ## References 1. Steel DHW, Wong D, Sakamoto T. Silicone oils compared and found wanting. Graefes Arch Clin Exp Ophthalmol (2021) 259 11-12. PMID: 32572605 2. Sultan ZN, Agorogiannis EI, Iannetta D, Steel D, Sandinha T. Rhegmatogenous retinal detachment: a review of current practice in diagnosis and management [published correction appears in BMJ Open Ophthalmol. 2021;6(1): e000474corr1]. BMJ Open Ophthalmol (2020) 5 e000474 3. Ramezani A, Ahmadieh H, Rozegar A, Soheilian M, Entezari M, Moradian S. Predictors and Outcomes of Vitrectomy and Silicone Oil Injection in Advanced Diabetic Retinopathy. Korean J Ophthalmol (2017) 31 217-229. PMID: 28534343 4. Ghoraba HH, Leila M, Shebl M, Abdelhafez MA, Abdelfattah HM. Long-Term Outcome After Silicone Oil Removal in Eyes with Myopic Retinal Detachment Associated with Macular Hole. Clin Ophthalmol (2021) 15 1003-1011. PMID: 33727783 5. Wang E, Chen Y, Li N, Min H. Effect of silicone oil on peripapillary capillary density in patients with rhegmatogenous retinal detachment. BMC Ophthalmol (2020) 20 268. PMID: 32635899 6. Ata-ur-Rasool Chaudhry N, Khan AA, Mahjoo T, Manan K. Retinal Re-Detachment after Silicone Oil Removal. Pak J Ophthalmol (2017) 33 253-258 7. Siyal NA, Hargun LD, Wahab S. Passive removal of silicone oil through 23gauge transconjunctival sutureless vitrectomy system. Pak J Med Sci (2016) 32 652-656. PMID: 27375708 8. Kaya M, Özyurt A, Öztürk AT, Er D, Kaynak S, Koçak N. Active Silicone Oil Removal with a Transconjunctival Sutureless System: Is the 23-Gauge System Safe and Effective. Turk J Ophthalmol (2016) 46 11-15. PMID: 27800251 9. Gujral GS. Basics of fluid-air exchange in vitreoretinal surgery. Kerala J Ophthalmol (2020) 32 284-286 10. Shiihara H, Sakamoto T, Terasaki H, Yamashita T, Yoshihara N, Okamoto F. Effect of fluid-air exchange on reducing residual silicone oil after silicone oil removal. Graefes Arch Clin Exp Ophthalmol (2017) 255 1697-1704. PMID: 28616714 11. Funatsu R, Terasaki H, Koriyama C, Yamashita T, Shiihara H, Sakamoto T. Silicone oil versus gas tamponade for primary rhegmatogenous retinal detachment treated successfully with a propensity score analysis. Japan Retinal Detachment Registry. Br J Ophthalmol (2022) 106 1044-1050. PMID: 34373251 12. Abu-Yaghi NE, Abu Gharbieh YA, Al-Amer AM, AlRyalat SAS, Nawaiseh MB, Darweesh MJ. Characteristics, fates and complications of long-term silicone oil tamponade after pars plana vitrectomy. BMC Ophthalmol (2020) 20 336. PMID: 32807120 13. Yu J, Zong Y, Tan Y, Jiang C, Xu G. Comparison of Repeated Fluid-Air Exchange and Passive Drainage for Removing Residual Emulsified Silicone Oil Droplets. J Ophthalmol 2020 (2020) 8184607 14. Soliman W, Mohamed TA, Abdelazeem K, Sharaf M. Trans-scleral posterior capsulorhexis in combined lens extractionand silicone oil removal. Eur J Ophthalmol (2020) 30 224-228. PMID: 30871372 15. Ismail NS, Phang LK, Min TW, Halim WH, Ali HM. Intraocular silicone oil removal: timing, outcome, and silicone oil complications encountered. Malaysian J Ophthalmol (2019) 1 37-49 16. Adhi MI, Siyal N. Retinal re-detachments after removal of silicone oil: Frequency and timings in a retrospective clinical study. J Pak Med Assoc (2019) 69 1822-1826. PMID: 31853111 17. Shiihara H, Terasaki H, Yoshihara N, Shirasawa M, Otsuka H, Yamashita T. Amount of residual silicone oil in vitreous cavity is significantly correlated with axial length. Retina (2016) 36 181-187. PMID: 26049621 18. Rhatigan M, McElnea E, Murtagh P, Stephenson K, Harris E, Connell P. Final anatomic and visual outcomes appear independent of duration of silicone oil intraocular tamponade in complex retinal detachment surgery. Int J Ophthalmol (2018) 11 83-88. PMID: 29375996 19. Akkan F, Dinc D. A novel design silicone oil removal cannula. Asian J Med Sci (2021) 12 39-43 20. Bhoot M, Agarwal A, Dubey S, Pegu J, Gandhi M. Silicone Oil Induced Glaucoma. Delhi J f Ophthalmol (2018) 29 9-13 21. Branisteanu DC, Moraru AD, Maranduca MA, Branisteanu DE, Stoleriu G, Branisteanu CI. Intraocular pressure changes during and after silicone oil endotamponade (Review). Exp Ther Med (2020) 20 204. PMID: 33123233
--- title: Zinc-alpha 2 glycoprotein a diagnostic Biomarker for early stage oral Squamous Cell Carcinoma authors: - Mehwish Feroz Ali - Mervyn Hosein - Saima Butt - Rehan Siddiqui journal: Pakistan Journal of Medical Sciences year: 2023 pmcid: PMC10025728 doi: 10.12669/pjms.39.2.6488 license: CC BY 3.0 --- # Zinc-alpha 2 glycoprotein a diagnostic Biomarker for early stage oral Squamous Cell Carcinoma ## Abstract ### Objectives: In this study, we investigated the expression of zinc alpha-2 glycoprotein in oral squamous cell carcinoma tissue samples. Additionally, ascertained its association to the oral cancer stage and subscale parameters (TNM). ### Methods: This observational study was conducted at Ziauddin University from January to December 2020. Using the Open-Epi software, the sample size of 120 oral squamous cell carcinomas was calculated at $95\%$ confidence interval and a $5\%$ margin of error. Ethical approval was taken from the Institutional Ethical Review Committee. Histologically diagnosed cases of oral squamous cell carcinoma were obtained from the Histopathology Department of Ziauddin University, Karachi. Study data was analyzed through SPSS version-20 and p-value ≤0.05 considered as significant. One-way ANOVA and Multiple linear regression were applied for analysis of data. ### Result: In the study, none of the oral squamous cell carcinoma tissue samples from the later stages were stained for ZAG. However $71\%$ ($\frac{35}{49}$) of the early stage OSCC samples showed positive IHC results for ZAG expression in the cytoplasm. One-way ANOVA indicates that high ZAG expression was significantly associated with smaller tumor size ($p \leq 0.001$), lymph node involvement ($$p \leq 0.002$$), early stages of OSCC ($p \leq 0.001$) and less differentiated tumor ($$p \leq 0.001$$). The site of the tumor was also significantly associated with ZAG staining ($p \leq 0.001$). ### Conclusion: Zinc alpha-2 glycoprotein expressed in the early stages of oral cancer development so that effective treatment modalities can be planned as per the patient’s status. This may also assist a clinician to achieve tumor-free surgical margins and monitor the post treatment outcomes. ## INTRODUCTION Oral squamous cell carcinoma (OSCC) is the most common type of head and neck cancer, accounting for $90\%$ of all oral cancers.1 According to the literature, the global occurrence of oral cancer is 2-4 percent, 10.9 percent in Pakistan, and 40 percent in India.2,3 This could be attributed to increased consumption of both smoked and smokeless tobacco in South-East Asia.4 Despite advanced treatment options, treatment outcomes are unsatisfactory due to tumor invasion, metastasis and recurrence.5 There is a scope to identify a potential biomarker that have tumor suppressor activity, inhibits degree of differentiation and can improve patients’ prognosis. The biomarker can assist a clinician to plan treatment, use it as adjuvant and monitor the post-treatment outcomes.6 In this study, we looked at the expression of Zinc alpha 2-glycoprotein (ZAG) in oral squamous cell carcinoma tissue samples and how it correlated with clinical and histological parameters. ZAG is a novel adipokine which is secreted by adipose tissue, liver, epithelial ductal cells and the tumor itself.7 ZAG induces lipid catabolic activity directly in adipocytes by a cyclic AMP-mediated process and this is initiated through binding to a β3-adrenoceptor.8 This is due to its high structural and functional resemblance to lipid mobilizing factor, with the exception of a minor difference in post-translational modification.7 Vidoto et al. have found increased expression of ZAG in head and neck cancer, suggesting that the high levels of ZAG inhibit tumor cells proliferation via immune activity against tumor antigen.9 According to Hasan et al, ZAG inhibits enzyme-mediated tumor invasion and proliferation due to its high structural and functional similarity to class I MHC molecules and its complex with enzymes (macroglobulin/hydrolases).10 According to studies, high expression of ZAG is associated with elimination of mutated RNAs and their by-products, as well as the down-regulation of cyclin-dependent kinase 2 (rate limiting enzymes) in the cell cycle.11 The enzyme is required to regulate the G2-M phase of the cell cycle, and by doing so, it may contribute to the inhibition of tumor cell growth. ZAG overexpression promotes epithelial-mesenchymal transition (EMT), tumor invasion, and apoptosis via the TGF1-ERK2 signaling pathway.12 This could be accomplished by down-regulating epithelial markers (E-cadherin) and increasing mesenchymal markers (N-Cadherin). According to the literature, ZAG has been found in a variety of tumors including oropharyngeal, esophageal, gastric, breast, prostate, pancreas, and liver tumours.9-11 The up-regulation of ZAG in various cancers was observed in early tumor stages, which was associated with a longer disease free survival and overall survival rate.11-13 There is a discrepancy in the literature about ZAG expression in tumors and its role as an early diagnostic marker. There have been very few studies on its expression in oral squamous cell carcinoma.13-15 In this study, we assessed ZAG expression in the tissue samples of oral squamous cell carcinoma cases. And also determine its correlation with staging and subscale parameters (TNM) of oral cancer. ## METHODS This observational study was conducted at Ziauddin University from January to December 2020. The sample size of 120 cases of oral squamous cell carcinoma was calculated using Open-Epi software with a $95\%$ confidence interval and a $5\%$ margin of error. The Ziauddin University Ethical Review Committee provided ethical approval (Reference code: 1531010MFOM). The sampling method used was consecutive. The Histopathology Department of Ziauddin University, North Campus, Karachi, provided histologically diagnosed cases of oral squamous carcinoma. By obtaining informed consent from each department, all samples were included in the study. Cases of OSCC that had been surgically removed and histologically determined match the inclusion criteria. Patients aged ≥18 were included in the study. Recurrent OSCC, malignancy other than OSCC, improperly stained histological slides, and a paucity of tissue in paraffin blocks are all exclusion criteria. The paraffin-embedded tissue blocks were sectioned into 3-5µm wide slices. The tissue sections were placed on the glass slide. Two glass slides were prepared, one for IHC and another slide for hematoxylin & eosin staining. The expression of zinc α-2 glycoprotein was investigated by Immunohistochemical staining using biotinylated antibodies against ZAG antigen on the cancer tissue sample. The cases were graded by using WHO/Broder classification and staging was evaluated by using the AJCC (TNM classification). The AZGP1 polyclonal antibody kit was purchased from the Thermo-scientific company. The standard protocol of IHC was applied. The histopathologist found 20 areas which were distributed among the marked hot spots with high brown staining. Each area $80\%$ covered by tumor cells without any artifact. The IHC slides were interpreted as ZAG positive cells scores as 0; <$5\%$, 1; $5\%$-$25\%$, 2; $25\%$-$50\%$, 3; $50\%$-$75\%$, and 4; >$75\%$. The ZAG positive staining intensity is scored as 0[-] Negative, 1(+) Mild, 2(++) Moderate, and 3(+++) Strong. Slides were examined under Ts2R-FL inverted research microscope (Nikon). Images were captured using NIS element D software. Images were processed using Photoshop. The following variables were recorded: gender, age, site of the tumor, size and thickness of the tumor, nodal involvement, distant metastasis, clinical stages, histological grades, lymphovascular invasion, perineural invasion, and extracapsular spread. Statistical analysis was performed by SPSS version 20.0 (SPSS Inc., Chicago, USA). The frequency and percentage were calculated for the categorical data. Mean and SD for the quantitative data. One-way ANOVA and Multi-regression analysis were used to find correlation of ZAG staining intensity with clinical and histological features. P-value ≤0.05 was considered statistically significant. ## RESULTS In the study, there were 100 ($83.3\%$) males and 20 ($16.7\%$) females. The maximum OSCC cases fell within the third (36; $30\%$) and fourth (33; $27.5\%$) decades of life. The most common site was buccal mucosa (80; $66.7\%$) followed by the lateral border of the tongue (21; $17.5\%$) and alveolar mucosa (9; $7.5\%$). The mean age was 47.2±11.1 years, the mean size of the tumor was 3.7 ± 1.9 mm, and the mean thickness was 1.5 ± 1.3 mm. In the reported cases of OSCC, there were 90 ($75\%$) cases of moderately differentiated OSCC, 18 ($15\%$) well differentiated and 12($10\%$) of poorly differentiated. Majority of patients presented at the late stages of OSCC around [71] $59\%$. None of the advanced stage OSCC tissue samples observed to be positive for ZAG staining on IHC but $71\%$ ($\frac{35}{49}$) early stage OSCC samples showed positive staining of ZAG. The positive ZAG staining in the tissue samples of OSCC through Immunohistochemistry (IHC) is represented in Fig-I. The ZAG protein was expressed in the cytoplasm of the tumor cells. The image A shows ZAG staining of mild intensity (1+) in the cytoplasm of the oral cancer cells around 25-$50\%$ tumor cells that were positively stained. The image-C shows staining of moderate intensity (2++) about 50-$75\%$ oral cancer cells that were positively stained. The image-E shows staining of strong intensity (3+++) >$75\%$ oral cancer cells that were positively stained. The images B, D & F were the hematoxylin and eosin counterparts of the IHC slides. Fig.1: Images A, C & E represents the Immunohistochemical staining (IHC) of ZAG in oral squamous cell carcinoma tissue samples with hematoxylin and eosin counterstain (images B, D & F). The image A shows ZAG staining of mild intensity (1+) in the cytoplasm of the oral cancer cells around 25-50% tumor cells that were positively stained for ZAG. The image C shows staining of moderate intensity (2++) about 50-75% oral cancer cells that were positively stained for ZAG. The image E shows staining of strong intensity (3+++) >75% oral cancer cells that were positively stained for ZAG. The association of staining intensity with clinicopathologic parameters of OSCC patients by using One-way ANOVA is represented in Table-I. The statistics indicate that high ZAG expression was significantly associated with smaller tumor size ($p \leq 0.001$), lymph node involvement ($$p \leq 0.002$$), early stages of OSCC ($p \leq 0.001$) and less differentiated tumor ($$p \leq 0.001$$). The site of the tumor were also significantly associated with ZAG staining ($p \leq 0.001$). Moreover, ZAG didn’t show any association with gender ($$p \leq 0.097$$), and age ($$p \leq 0.889$$). To confirm the correlation of ZAG staining with staging and grading of OSCC, multiple linear regression method was applied which showed statistically significant results ($p \leq 0.001$, $$p \leq 0.001$$) respectively. ## DISCUSSION In the study, the zinc alpha-2 glycoprotein was positively expressed in the early stages of OSCC tissue samples. Similarly, studies have found that high levels of zinc alpha-2 glycoprotein expression are associated with a lower degree of differentiation in many cancers.9-15 Though the precise mechanism is unknown, ZAG expression has been regulated by acetylation of histone which controls genes through chromatin conformation.16 Another research found that histone deacetylation leads ZAG expression to be decreased in pancreatic cancer.16 Zinc alpha-2 glycoprotein is a clinically significant protein that plays a role in tumor formation and proliferation.9-16 It was first identified as a lipid mobilizing factor, subsequently as a tumor marker that increases in disease-specific cachexia in humans.17,18 The increased level of ZAG mRNA and protein in cancer cachexia occurred by peroxisome-activated receptor gamma (PPARγ) and uncoupling protein (UCP-2).7 The down-regulation of ZAG usually associated with aggressive patterns of tumor proliferation, poor prognosis and worse treatment outcomes. In the study, a multiple linear regression analysis was used to assess the association between ZAG and many clinical and pathologic parameters. The majority of ZAG positive stained samples were from men, owing to the fact that men in our demographic consume the most smokeless tobacco.2 The buccal mucosa was shown to be the most common positive site for ZAG staining in the study. The explanation for this is the use of powdered tobacco in one or more forms. It is typical practice in our population to place powered tobacco in the buccal sulcus, which releases its content locally.5 This results in continual mucosal irritation, chronic inflammation, and premalignant and cancerous alterations.3 In the study, we investigated how ZAG correlated with histological components such histological grades, lymphovascular invasion, perineural invasion, and extracapsular spread to predict patient prognosis. ZAG was not expressed in any of the OSCC cases that had lymphovascular invasion, perineural invasion, or extracapsular spread. Poropatich et al also found an up-regulation of ZAG in the low grade oropharyngeal squamous cell carcinoma with significantly longer recurrence free survival.17 According to Vidotto et al, increased ZAG expression induces an immune response against tumor antigens and mucosal breakdown by proteolytic enzymes in Head and Neck Squamous Cell Carcinoma.9 This could imply that ZAG inhibits tumor proliferation and differentiation. This could also suggest that reduction of ZAG expression is linked to metastasis, poor prognosis, local recurrence, and unsuccessful treatment outcomes. On multivariate analysis, greater ZAG expression was associated with smaller tumor size T1 and T2, as well as no lymph node metastases in the study. According to a study conducted by Haung et al, ZAG protein downregulation was correlated with larger tumor size but not with nodal involvement or distant metastasis.19 This indicates that the more oral tissues involved in the malignant process, the lower the expression of ZAG. According to Xu et al study, low ZAG levels in hepatocellular carcinoma cell lines enhance epithelial to mesenchymal transition (EMT) via the TGF1-ERK2 signaling pathway.20 This could occur by down-regulating epithelial markers (E-cadherin) and up-regulating mesenchymal markers (N-Cadherin).20 This may be linked to the fact that less differentiated tumors in the study did not express positive ZAG staining. Tang and colleagues have reported down-regulation of ZAG in the aggressive stages of esophageal squamous cell carcinoma, which could indicate increased rate of tumor progression.16 ZAG overexpression activates the rapamycin signaling pathway and inhibits the cyclin dependent 2 kinase enzyme, which controls the cell cycle’s G2-M transition.22 It inhibits tumor cell growth, proliferation, and migration in this way.22,23 In contrast, Dengbo et al. study reported high levels of ZAG in the advanced stages of colorectal carcinoma and associated with hepatic metastasis, shorter disease free survival and overall survival.24,25 All of the aforementioned studies suggested that ZAG may play a role in the initiation and progression of carcinomas.16-25 In future, ZAG could be used as a reliable biomarker to maintain epithelial phenotype, predict prognosis, and post-treatment outcomes. To the best of our knowledge, we have investigated ZAG expression for the first time in oral squamous cell carcinoma tissue samples. The findings demonstrated a significant expression of ZAG in the early stages of OSCC, suggesting that it may be a useful marker for predicting the prognosis and effectiveness of treatment in such patients. It could serve as a therapeutic agent or assist a clinician plan a course of treatment according to the ZAG expression. However, further thorough investigations are warranted to find the precise role of the ZAG in oral squamous cell carcinoma. ## Limitations: It includes that only one method was used to assess the ZAG expression in the tissue samples due to financial restraints. Because of the same reason we were also unable to perform protein validation in our study. ## CONCLUSION The zinc alpha-2 glycoprotein positive stained samples in this study belonged to the early stages of oral squamous cell carcinoma. Positively stained samples are less differentiated, showing that ZAG expression decreased with increased tumor cell proliferation and differentiation. The zinc alpha-2 glycoprotein may be used to diagnose OSCC in its early stages and to plan treatment modalities based on the needs of the patient. This may also aid a surgeon in achieving tumor-free surgical margins. According to the study’s findings, ZAG could be used as a potential biomarker for prognosis and post-therapy outcomes in oral cancer. ## Authors’ contributions: MF: Conceived and designed the study, collected the data, performed the analysis, and writing of the manuscript. MH: Conceived and designed the study, clinical integrity of the study, interpretation of data, and proof-reading. SB & RS: Conceived and designed the study, interpretation of data, and proof-reading. All authors are responsible and accountable for the accuracy and integrity of the work. ## References 1. Mahmood N, Hanif M, Ahmed A, Jamal Q, Mushtaq S, Khan A, Saqib M. Circulating miR-21 as a prognostic and predictive biomarker in oral squamous cell carcinoma. Pak J Med Sci (2019) 35 1408-1412. PMID: 31489016 2. Basit A, Younus BB, Waris N, Fawwad A, Members* N. Prevalence of tobacco use in urban and rural areas of Pakistan;a sub-study from second National Diabetes Survey of Pakistan (NDSP) 2016 - 2017: Prevalence of tobacco use in Pakistan. Pak J Med Sci (2020) 36 808-815. PMID: 32494279 3. Asthana S, Vohra P, Labani S. Association of smokeless tobacco with Oral cancer: A Review of systematic Reviews. Tob Prev Cessat (2019) 5 34. PMID: 32411897 4. Sahaf R, Naseem N, Rehman AU, Anjum R, Nagi AH. A Study of 89 Cases of Oral Squamous Cell Carcinoma Presenting at Teaching Hospitals of Lahore, Pakistan. J Pak Dent Assoc (2017) 26 26-31 5. Ahmed R, Malik S, Khan MF, Khattak MR. Epidemiological and clinical correlates of oral squamous cell carcinoma in patients from north-west Pakistan. J Pak Med Assoc (2019) 69 1074-1078. PMID: 31431755 6. Rasool M, Khan SR, Malik A, Khan KM, Zahid S, Manan A, Qazi MH, Naseer MI. Comparative Studies of Salivary and Blood Sialic Acid, Lipid Peroxidation and Antioxidative Status in Oral Squamous Cell Carcinoma (OSCC). Pak J Med Sci (2014) 30 466-471. PMID: 24948960 7. Escote X, Aranda GB, Mora M, Casals G, Ensenat J. Zinc alpha-2 glycoprotein is overproduced in Cushing's syndrome. Endocrinology Diabetes Nutr (2017) 64 26-33 8. Cabassi A, Tedeschi S. Zinc-α2-glycoprotein as a marker of fat catabolism in humans. Curr Opin Clin Nutr Metab Care (2013) 16 267-271. PMID: 23448999 9. Vidotto A, Henrique T, Raposo LS, Maniglia JV, Tajara EH. Salivary and serum proteomics in head and neck carcinomas: Before and after surgery and radiotherapy. Cancer Biomarker (2010) 8 95-107 10. Hassan MI, Waheed A, Yadav S, Singh TP, Ahmad F. Zinc α2-glycoprotein: a multidisciplinary protein. Mol Cancer Res (2008) 6 892-906. PMID: 18567794 11. Ebadi M, Mazurak VC. Review Article Potential Biomarkers of Fat Loss as a Feature of Cancer Cachexia. Mediators Inflamm 2015 (2015) 820934 12. Xu MY, Chen R, Yu JX, Liu T, Qu Y. AZGP1 suppresses epithelial-mesenchymal transition and hepatic carcinogenesis by blocking TGFβ-ERK2 pathways. Cancer Lett (2016) 374 241-249. PMID: 26902423 13. Feroz M, Hosein M, Butt S, Siddiqui RA. Zinc α-2 Glycoprotein (ZAG) a Novel Biomarker for the Detection of Oral Squamous Cell Carcinoma. Pak J Med Dent (2020) 9 77-83 14. Heawchaiyaphuma C, Pientong C, Phusingha P, Vatanasapt P, Promthet S. Peroxiredoxin-2 and zinc-alpha-2-glycoprotein as potentially combined novel salivary biomarkers for early detection of oral squamous cell carcinoma using proteomic approaches. J Proteomics (2017) 173 52-61. PMID: 29199150 15. Chen YT, Chen HW, Wu CF, Chu LJ, Chiang WF. Development of a multiplexed liquid chromatography multiple-reaction-monitoring mass spectrometry (LC-MRM/MS) method for evaluation of salivary proteins as oral cancer biomarkers. Mol Cell Proteomics (2017) 16 799-811. PMID: 28235782 16. Tang H, Wu Y, Qin Y, Wang H, Wang L, Guan X. Reduction of AZGP1 predicts poor prognosis in esophageal squamous cell carcinoma patients in Northern China. Onco-Targets Ther (2016) 10 85-94. PMID: 28053542 17. Poropatich K, Paunesku T, Zander A, Wray B, Schipma M. Elemental Zn and its Binding Protein Zinc-α2-Glycoprotein are elevated in HPV-Positive Oropharyngeal Squamous Cell Carcinoma. Sci Rep (2019) 9 16965. PMID: 31740720 18. Ryd´en M, Agustsson T, Andersson J, Bolinder J, Toft E. Adipose zinc-??2-glycoprotein is a catabolic marker in cancer and noncancerous states. J Intern Med (2012) 271 414-420. PMID: 21883534 19. Huang CY, Zhao JJ, Lv L, Chen YB, Li YF, Jiang SS. Decreased expression of AZGP1 is associated with poor prognosis in primary gastric cancer PLoS One;. (2013) 8 e69155/1-7 20. Xue Y, Yu F, Yan D, Cui F, Tang H. Zinc-α-2-Glycoprotein: A Candidate Biomarker for Colon Cancer Diagnosis in Chinese Population. Int J Mol Sci (2015) 16 691-703 21. Kong B, Michalski CW, Hong X, Valkovskaya N, Rieder S. AZGP1 is a tumor suppressor in pancreatic cancer inducing mesenchymal-to-epithelial transdifferentiation by inhibiting TGF-beta-mediated ERK signaling. Oncogene (2010) 29 5146-5158. PMID: 20581862 22. He N, Brysk H, Tyring SK, Ohkubo I, Brysk MM. Zinc-alpha (2)-glycoprotein hinders cell proliferation and reduces cdc2 expression. J Cell Biochem (Suppl) (2001) 36 162-169 23. Xu MY, Chen R, Yu JX, Liu T, Qu Y. AZGP1 suppresses epithelial-mesenchymal transition and hepatic carcinogenesis by blocking TGFβ-ERK2 pathways. Cancer Lett (2016) 374 241-249. PMID: 26902423 24. Dengbo Ji, Ming Li, Tiancheng Zhan, Yunfeng Yao, Jing Shen. Prognostic role of serum AZGP1, PEDF and PRDX2 in colorectal cancer patient. Carcinogenesis (2013) 34 1265-1272. PMID: 23393224 25. Ji M, Li W, He G, Zhu D, Lv S, Tang W. Zinc-α2-glycoprotein 1 promotes EMT in colorectal cancer by filamin A mediated focal adhesion pathway. J Cancer (2019) 10 5557-5566. PMID: 31632499
--- title: Maternal and perinatal outcome of Ramadan fasting in women with gestational diabetes authors: - Saba Abdullah - Shumaila - Saba Mughal - Mahwish Samuel - Nazli Hossain journal: Pakistan Journal of Medical Sciences year: 2023 pmcid: PMC10025732 doi: 10.12669/pjms.39.2.7332 license: CC BY 3.0 --- # Maternal and perinatal outcome of Ramadan fasting in women with gestational diabetes ## Abstract ### Objective: To compare maternal and perinatal outcome of Ramadan fasting during pregnancy in women with/without gestational diabetes. ### Methods: This prospective case-control study was conducted at Department of Obstetrics & Gynecology Unit 1 Ruth PKM Civil Hospital & Dow Medical College and Holy Family Hospital, Karachi during 1st April to 31st July, 2022. In this study normoglycemic pregnant women and those identified as gestational diabetes ($$n = 52$$) on oral glucose tolerance test, who fasted during Ramadan were included. Women, on diet control or diet plus metformin were included in the study. Study questionnaire included demographic details, days of fasting, self-reported hypoglycemic episodes. Maternal outcomes included preterm birth, pregnancy induced hypertension. Perinatal outcome included hyperbilirubinemia, hypoglycemia, weight of placenta, and apgar score. ### Result: Eighty two women were included in the study, gestational diabetes ($$n = 57$$) and normoglycemic ($$n = 25$$). Average days of fasting were 16 ±9.0 days (range 5-30). Women with GDM were older (28.6 vs. 26.0 years, p-value=0.034), had raised levels of HbA1c (5.5 vs. 5.1, p-value=0.004), mean FBS (102.8 vs. 84.6 mg/dl, p-value <0.001), mean RBS (135.3 vs. 106.4 mg/dl, p-value <0.001) and had higher BMI at delivery (31.0 vs. 26.6 kg/m2, p-value=0.004). HbA1c (p-value=0.016) and head circumference of baby (p-value=0.038) were found lower in the group who fasted for more than 20 days among normoglycemic pregnant women. No other maternal and neonatal outcomes were found to be significantly affected by Ramadan fasting among pregnant women with/without GDM. ### Conclusion: Gestational diabetes do not affect maternal and perinatal outcome among pregnant women. ## INTRODUCTION Ramadan fasting is an important pillar of Islam. Fasting during pregnancy is not obligatory and Muslim women are required to complete the count, at a later stage. Majority of the women prefer to complete their count during the holy month of Ramadan. Not only they find it easier but socially more convenient.1 Surveys have found that more than $50\%$ of pregnant women prefer to fast during the month of Ramadan, despite religious exemption.2 There are around two billion Muslims in the world, with majority living in the Asia-Pacific region.3 Though there are guidelines and scientific opinions for Ramadan fasting and pregnancy,4 there is limited data available for maternal fasting and gestational diabetes. This is due to lack of literature and scientific data. In a small study, pregnant diabetic women, who fasted during Ramadan were compared with normoglycemic women. The investigators did not find any adverse effect on fetal development, or any significant effect on metabolic profile of diabetic women who fasted.5 In another study diabetic women on insulin, and gestational diabetes were compared for any adverse effect due to fasting. There was improvement in levels of glycosylated hemoglobin levels(HbA1C) and serum fructosamine in both groups of women, without any adverse fetal effects.6 The changes in glycosylated hemoglobin levels persisted after Ramadan. The authors suggested instead of complete ban on fasting among women on Insulin therapy, better collaboration with physician and blood glucose monitoring may be more helpful. There is increased prevalence of gestational diabetes and diabetes mellitus among people of South-Asian origin. The prevalence of gestational diabetes for Pakistan is estimated to be around $14\%$.7 This translates into the fact that in considerable number of women Ramadan intersects pregnancy. During the month of Ramadan, depending on the region, the duration of fasting is between 14-16 hours. This year, the month of Ramadan was observed in April-May, with 14-15 hours of continuous fasting. Both Suhoor, (pre sun-rise meal, when fast starts) and Iftar (post sun-set meal, when fast is broken) have been found to be associated with periods of hyperglycemia and hypoglycemia. Fasting in women with gestational diabetes is not recommended, for adverse maternal and fetal outcome. But majority of women do not pay attention to the medical advice and continue to fast in the month of Ramadan. The aim of the study was to see the maternal and fetal outcome of Ramadan fasting in women with gestational diabetes. ## METHODS This prospective study was conducted at the Dept. of Obstetrics & Gynecology Unit-I, Ruth Pfau KM Civil Hospital & Dow Medical College, and Holy Family Hospital, Karachi. ## Inclusion criteria: Singleton pregnancy,Diagnosis of gestational diabetes mellitus (either made for first time or already diagnosed on routine screening as GDM in current pregnancy),Pregnant gestational diabetic women on diet control or diet and metformin therapy,Pregnant women with normal glucose levels who fasted. ## Exclusion criteria: Multiple pregnancy,Insulin dependent diabetes mellitus,Anomalous baby and diagnosed intrauterine demise. Pregnant women, diagnosed with gestational diabetes mellitus, registered with the antenatal clinic of the department or delivering with the unit were included in the study, after informed verbal consent. The study period was from April to July 2022. The diagnosis of gestational diabetes mellitus (GDM) is made on oral glucose tolerance test, carried out with 75gm glucose, between 24-28 weeks of gestation. Two abnormal glucose values indicated the presence of gestational diabetes mellitus. Patients diagnosed with GDM, on diet control only or those with diet control plus metformin were included in the study. Days of fasting were stratified as 1-10 days, 11-20 days and more than 20 days of fasting. This group was compared with women who were found normoglycemic on oral glucose tolerance test, and fasted during the month of Ramadan. Demographic variables included maternal age, parity, family history of diabetes mellitus, body mass index at the start of pregnancy, weight gain during pregnancy. Also noted were history of continuous or intermittent fasting, and days of fasting. Maternal outcomes included pregnancy induced hypertension, preterm delivery, and mode of delivery. Hypertension was defined as new onset systolic blood pressure ≥140 mm hg, and diastolic blood pressure of ≥ 90 mm hg, with or without proteinuria. Preterm delivery was defined as delivery before 37 completed weeks of gestation. Perinatal outcome included fetal weight, apgar score at one and five minute, and weight of placenta at delivery. Around 3cc of blood was taken from neonate for serum blood sugar levels and serum bilirubin levels, within 24 hours of delivery. Hypoglycemia in newborn was defined as blood sugar levels ≤ 40mg/dl. Raised serum bilirubin levels was defined as baby requiring phototherapy or exchange transfusion. Also included were macrosomia (> 4.5kg) or small for gestational age (< 10th percentile). The details were collected on a predesigned questionnaire, and data was transferred on excel sheet for further analysis. Neonatal blood 3cc was collected for serum bilirubin and glucose levels within 24 hours of delivery. Weight of the placenta was done according to the standard technique, as elaborated in the previous study.8 Women were asked to maintain blood glucose levels, at least post Iftar, pre Iftar. Other biochemical markers included complete blood picture, glycosylated hemoglobin (HBA1c) levels before and after Ramadan. Women were also advised for self-monitoring of hypoglycemic events and to report in emergency in event of hypoglycemic events. The study group was compared with fasting pregnant women, without the diagnosis of gestational diabetes mellitus. The study was approved by the Institutional Review Board of University. ( IRB 2488/DUHS/Approval/$\frac{2022}{786}$). ## Sample size calculation and statistical analysis: Sample size was calculated using OpenEpi calculator using $5\%$ margin of error, $80\%$ power, $95\%$ confidence level, proportion of non-fasting group of pregnant women ($13.9\%$) and odds ratio (6.6, as pregnant women with exposure of > 10 hours of Ramadan fasting more likely to develop hyperbilirubinemia in neonates).9 The total sample size came out to be 56 pregnant women with GDM. Statistical Package for Social Sciences (SPSS) version 22.0 was used for data analysis. Descriptive statistics were reported as frequency (percentage) for categorical variables and mean (standard deviation) for numerical variables. Normality of continuous variables was checked by Shapiro-Wilk test. Independent t test/ Mann-Whitney U test/ Kruskal Wallis test were applied to check mean/ median differences of maternal and neonatal outcomes between the GDM and non-GDM groups and days of fasting. Chi-square test/ Fisher Exact test were used to check association between categorical outcome and independent variables. P-value <0.05 were considered to be statistically significant. ## RESULTS A total of 82 pregnant women were included in this study who fasted in Ramadan during pregnancy. Average age of pregnant women was 27.8 ±5.4 years which ranged from 18 to 40 years whereas average BMI at start of pregnancy was 28.4 ±7.1 and at delivery was 29.7 ±6.7. Women fasted in Ramadan on average 16 ±9.0 days which ranged between five and 30 days. Among all, 57 ($69.5\%$) women were diagnosed with GDM and 25 ($30.5\%$) without GDM. Sixteen percent of the women experienced preterm delivery and sixty one percent underwent caesarean section. Neonatal blood glucose level was 67.2 ±14.1and serum bilirubin was 3.3 ±1.7 mg/dL whereas mean birth weight was 2.9 ±0.5 kg (Table-I). Table-I \| Characteristics \| Characteristics.1 \| Unnamed: 2 \| Range \| \| --- \| --- \| --- \| --- \| \| Demographics- pregnant women \| Demographics- pregnant women \| \| \| \| Maternal age (years) \| Maternal age (years) \| 27.8 ± 5.4 \| (18 - 40) \| \| BMI at start (kg/m2) \| BMI at start (kg/m2) \| 28.4 ± 7.1 \| (18 - 45) \| \| Days of fasting \| Days of fasting \| 15.8 ± 9.0 \| (5 - 30) \| \| HbA1c (%) \| HbA1c (%) \| 5.3 ± 0.6 \| (4.0 - 7.3) \| \| Mean FBS (mg/dL) \| Mean FBS (mg/dL) \| 96.9 ± 14.7 \| (72 - 134) \| \| Mean RBS (mg/dL) \| Mean RBS (mg/dL) \| 125.8 ± 31.9 \| (78 - 232) \| \| Consanguineous marriage, yes \| Consanguineous marriage, yes \| 33 (40.2) \| \| \| Family history of DM, yes \| Family history of DM, yes \| 23 (28.0) \| \| \| Maternal characteristics \| Maternal characteristics \| \| \| \| Gestational age (weeks) \| Gestational age (weeks) \| 37.7 ± 1.4 \| (35 - 41) \| \| BMI at delivery (kg/m2) \| BMI at delivery (kg/m2) \| 29.7 ± 6.7 \| (20 - 47) \| \| GDM, yes \| GDM, yes \| 57 (69.5) \| \| \| Treatment of GDM (n=57) \| Treatment of GDM (n=57) \| \| \| \| Diet alone \| Diet alone \| 26 (45.6) \| \| \| Diet & Metformin \| Diet & Metformin \| 31 (54.4) \| \| \| Mode of delivery \| Mode of delivery \| \| \| \| SVD/ Instruments \| SVD/ Instruments \| 32 (39.0) \| \| \| LSCS \| LSCS \| 50 (61.0) \| \| \| Preterm delivery, yes \| Preterm delivery, yes \| 13 (15.9) \| \| \| Neonatal characteristics \| Neonatal characteristics \| \| \| \| Weight of baby (kg) \| Weight of baby (kg) \| 2.9 ± 0.5 \| (1.9 - 4.0) \| \| Length of baby (cm) \| Length of baby (cm) \| 48.8 ± 3.1 \| (40 - 57) \| \| Head circumference (cm) \| Head circumference (cm) \| 33.4 ± 2.6 \| (29 - 53) \| \| Mid upper arm circumference (cm) \| Mid upper arm circumference (cm) \| 9.8 ± 1.0 \| (8 - 13) \| \| Weight of placenta (kg) \| Weight of placenta (kg) \| 0.5 ± 0.1 \| (0.2 - 0.9) \| \| Apgar score at 1-min \| Apgar score at 1-min \| 6.7 ± 0.8 \| (5 - 8) \| \| Apgar score at 5-min \| Apgar score at 5-min \| 8.7 ± 0.5 \| (7 - 10) \| \| Neonatal blood glucose level (mg/dL) \| Neonatal blood glucose level (mg/dL) \| 67.2 ± 14.1 \| (41 - 104) \| \| Neonatal serum bilirubin (mg/dL) \| Neonatal serum bilirubin (mg/dL) \| 3.3 ± 1.7 \| (1.3 - 9.9) \| \| Gender of baby \| Gender of baby \| \| \| \| Male \| Male \| 39 (47.6) \| \| \| Female \| Female \| 43 (52.4) \| \| Comparisons of maternal and neonatal outcomes were made between groups of GDM and non-GDM. Those women who were diagnosed with GDM were older (28.6 vs. 26.0 years, p-value=0.034), had raised levels of HbA1c (5.5 vs. 5.1, p-value=0.004), mean FBS (102.8 vs. 84.6 mg/dl, p-value <0.001), mean RBS (135.3 vs. 106.4 mg/dl, p-value <0.001) and higher BMI at delivery (31.0 vs. 26.6 kg/m2, p-value=0.004) as compared to non-GDM pregnant women. Other maternal and neonatal outcomes did not show significant differences between the groups of GDM (Table-II). Table-II \| Variables \| GDM \| Non-GDM \| p-value* \| \| --- \| --- \| --- \| --- \| \| Variables \| \| \| p-value* \| \| Variables \| (n = 57) \| (n = 25) \| p-value* \| \| Demographics- pregnant women \| \| \| \| \| Maternal age (years) \| 28.6 ± 4.9 \| 26.0 ± 5.9 \| 0.034 \| \| BMI at start (kg/m2) \| 29.1 ± 7.2 \| 25.2 ± 6.1 \| 0.099 \| \| Days of fasting \| 15.8 ± 8.8 \| 15.7 ± 9.5 \| 0.820 \| \| HbA1c (%) \| 5.5 ± 0.6 \| 5.1 ± 0.5 \| 0.004 \| \| Mean FBS (mg/dL) \| 102.8 ± 13.1 \| 84.6 ± 9.5 \| < 0.001 \| \| Mean RBS (mg/dL) \| 135.3 ± 33.2 \| 106.4 ± 17.3 \| < 0.001 \| \| Family history of DM \| \| \| \| \| Yes \| 19 (33.3) \| 4 (16.0) \| 0.108 \| \| No \| 38 (66.7) \| 21 (84.0) \| \| \| Maternal outcomes \| \| \| \| \| Gestational age (weeks) \| 37.6 ± 1.4 \| 38.0 ± 1.3 \| 0.178 \| \| BMI at delivery (kg/m2) \| 31.0 ± 6.8 \| 26.6 ± 5.2 \| 0.004 \| \| Mode of delivery \| \| \| \| \| SVD/ Instruments \| 19 (33.3) \| 13 (52.0) \| 0.111 \| \| LSCS \| 38 (66.7) \| 12 (48.0) \| \| \| Preterm delivery \| \| \| \| \| Yes \| 10 (17.5) \| 3 (12.0) \| 0.745 \| \| No \| 47 (82.5) \| 22 (88.0) \| \| \| Neonatal outcomes \| \| \| \| \| Weight of baby (kg) \| 2.9 ± 0.5 \| 2.8 ± 0.4 \| 0.635 \| \| Length of baby (cm) \| 48.8 ± 3.3 \| 48.8 ± 2.5 \| 0.714 \| \| Head circumference (cm) \| 33.1 ± 1.5 \| 34.0 ± 4.1 \| 0.385 \| \| Mid upper arm circumference (cm) \| 9.9 ± 1.0 \| 9.6 ± 0.8 \| 0.443 \| \| Weight of placenta (kg) \| 0.5 ± 0.1 \| 0.5 ± 0.1 \| 0.897 \| \| Apgar score at 1-min \| 6.8 ± 0.8 \| 6.6 ± 0.7 \| 0.547 \| \| Apgar score at 5-min \| 8.7 ± 0.5 \| 8.6 ± 0.6 \| 0.885 \| \| Neonatal blood glucose (mg/dL) \| 67.5 ± 13.8 \| 66.3 ± 15.1 \| 0.870 \| \| Neonatal serum bilirubin (mg/dL) \| 3.6 ± 1.9 \| 2.7 ± 0.9 \| 0.057 \| \| Gender of baby \| \| \| \| \| Male \| 26 (45.6) \| 13 (52.0) \| 0.594 \| \| Female \| 31 (54.4) \| 12 (48.0) \| \| Effect of Ramadan fasting on maternal and neonatal outcomes was examined separately among women with GDM and without GDM by further categorizing the groups into days of fasting (<11, 11-20, >20). It was noted that fasting was not significantly associated with any maternal and neonatal outcomes among pregnant women with GDM. However significant lower levels of HbA1c (p-value=0.016) and head circumference of baby (p-value=0.038) were found in the group who fasted for more than 20 days among non-GDM pregnant women. No other maternal and neonatal outcomes were found to be significantly affected by Ramadan fasting among pregnant women without GDM (Table-III). Table-III \| Variables \| Women with GDM (n=57) \| Women with GDM (n=57).1 \| Women with GDM (n=57).2 \| p-value* \| Women without GDM (n=25) \| Women without GDM (n=25).1 \| Women without GDM (n=25).2 \| p-value.1 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Variables \| \| \| \| p-value \| \| \| \| p-value* \| \| Variables \| < 11 \| 11 - 20 \| > 20 \| p-value* \| < 11 \| 11 - 20 \| > 20 \| p-value* \| \| Variables \| \| \| \| p-value* \| \| \| \| p-value* \| \| Variables \| (n = 24) \| (n = 17) \| (n = 16) \| p-value* \| (n = 11) \| (n = 6) \| (n = 8) \| p-value* \| \| Demographics- pregnant women \| \| \| \| \| \| \| \| \| \| Maternal age (years) \| 28.6 ± 5.4 \| 29.2 ± 4.8 \| 28.1 ± 4.6 \| 0.872 \| 24.5 ± 6.0 \| 28.3 ± 6.6 \| 26.4 ± 5.3 \| 0.322 \| \| BMI at start (kg/m2) \| 30.3 ± 7.4 \| 27.2 ± 7.2 \| 29.4 ± 7.2 \| 0.475 \| 28.2 ± 7.8 \| 25.2 ± 3.8 \| 23.3 ± 6.6 \| 0.372 \| \| HbA1c (%) \| 5.7 ± 0.6 \| 5.5 ± 0.7 \| 5.3 ± 0.4 \| 0.120 \| 5.2 ± 0.5 \| 5.4 ± 0.2 \| 4.6 ± 0.4 \| 0.016 \| \| Mean FBS (mg/dL) \| 104.2 ± 13.5 \| 101.7 ± 14.9 \| 101.8 ± 10.8 \| 0.655 \| 82.5 ± 7.2 \| 87.2 ± 14.1 \| 85.5 ± 9.0 \| 0.755 \| \| Mean RBS (mg/dL) \| 141.1 ± 34.4 \| 117.0 ± 16.6 \| 143.6 ± 37.7 \| 0.082 \| 110.1 ± 18.4 \| 101.0 ± 13.7 \| 105.2 ± 18.7 \| 0.573 \| \| Maternal outcomes \| \| \| \| \| \| \| \| \| \| Gestational age (weeks) \| 37.3 ± 1.5 \| 37.7 ± 1.4 \| 38.0 ± 1.3 \| 0.273 \| 37.7 ± 1.3 \| 38.2 ± 1.7 \| 38.5 ± 1.2 \| 0.501 \| \| BMI at delivery (kg/m2) \| 31.8 ± 7.0 \| 30.4 ± 6.4 \| 30.4 ± 7.4 \| 0.763 \| 27.0 ± 5.4 \| 25.9 ± 4.7 \| 26.6 ± 5.7 \| 0.859 \| \| Treatment of GDM (n=57) \| \| \| \| \| \| \| \| \| \| Diet alone \| 10 (41.7) \| 9 (52.9) \| 7 (43.8) \| 0.763 \| \| - \| \| \| \| Diet & Metformin \| 14 (58.3) \| 8 (47.1) \| 9 (56.2) \| \| \| - \| \| \| \| Mode of delivery \| \| \| \| \| \| \| \| \| \| SVD/ Instruments \| 7 (29.2) \| 6 (35.3) \| 6 (37.5) \| 0.843 \| 5 (45.5) \| 3 (50.0) \| 5 (62.5) \| 0.876 \| \| LSCS \| 17 (70.8) \| 11 (64.7) \| 10 (62.5) \| \| 6 (54.5) \| 3 (50.0) \| 3 (37.5) \| \| \| Preterm delivery \| \| \| \| \| \| \| \| \| \| Yes \| 5 (20.8) \| 4 (23.5) \| 1 (6.2) \| 0.448 \| 1 (9.1) \| 1 (16.7) \| 1 (12.5) \| 0.999 \| \| No \| 19 (79.2) \| 13 (76.5) \| 15 (93.8) \| \| 10 (90.9) \| 5 (83.3) \| 7 (87.5) \| \| \| Neonatal outcomes \| \| \| \| \| \| \| \| \| \| Weight of baby (kg) \| 2.9 ± 0.5 \| 2.9 ± 0.6 \| 2.8 ± 0.4 \| 0.923 \| 2.8 ± 0.3 \| 2.8 ± 0.6 \| 3.0 ± 0.3 \| 0.338 \| \| Length of baby (cm) \| 47.7 ± 3.7 \| 49.9 ± 3.1 \| 49.2 ± 2.5 \| 0.298 \| 49.3 ± 1.9 \| 47.0 ± 2.3 \| 49.3 ± 2.8 \| 0.090 \| \| Head circumference (cm) \| 32.6 ± 0.9 \| 33.5 ± 2.1 \| 33.4 ± 1.3 \| 0.055 \| 33.3 ± 0.6 \| 37.2 ± 7.8 \| 32.5 ± 1.4 \| 0.038 \| \| MUAC (cm) \| 9.8 ± 1.0 \| 9.9 ± 1.1 \| 9.9 ± 1.0 \| 0.762 \| 9.8 ± 0.6 \| 9.5 ± 1.3 \| 9.4 ± 0.9 \| 0.625 \| \| Weight of placenta (kg) \| 0.5 ± 0.1 \| 0.6 ± 0.1 \| 0.5 ± 0.1 \| 0.546 \| 0.6 ± 0.1 \| 0.5 ± 0.1 \| 0.5 ± 0.1 \| 0.070 \| \| Apgar score at 1-min \| 6.8 ± 0.8 \| 6.6 ± 0.9 \| 6.8 ± 0.8 \| 0.399 \| 6.6 ± 0.9 \| 6.7 ± 0.5 \| 6.6 ± 0.7 \| 0.965 \| \| Apgar score at 5-min \| 8.7 ± 0.5 \| 8.7 ± 0.4 \| 8.6 ± 0.5 \| 0.650 \| 8.6 ± 0.8 \| 8.6 ± 0.5 \| 8.7 ± 0.4 \| 0.942 \| \| Neonatal blood glucose (mg/dL) \| 65.0 ± 13.4 \| 67.7 ± 12.7 \| 71.0 ± 15.5 \| 0.417 \| 68.6 ± 18.1 \| 67.0 ± 18.4 \| 62.6 ± 6.7 \| 0.984 \| \| Neonatal serum bilirubin (mg/dL) \| 3.6 ± 2.1 \| 3.5 ± 1.6 \| 3.6 ± 2.1 \| 0.992 \| 2.9 ± 1.1 \| 2.6 ± 1.1 \| 2.4 ± 0.7 \| 0.642 \| ## DISCUSSION This study aimed to find out maternal and perinatal outcome among normoglycemic and in women with gestational diabetes mellitus. Fasting itself is associated with episodes of metabolic disturbances. These are aggravated in women with diabetes. Both hypoglycemia and hyperglycemia are detrimental for maternal and perinatal health. The average duration of fast varied between 15-16 hours. There is evidence of more hypoglycemia in women on diet±metformin, who fast during pregnancy. In our study, there was a significant difference between mean fasting and random blood glucose level among normoglycemic and gestational diabetes women. Studies from other parts of the world have also shown a decrease in mean blood glucose levels and glycosylated hemoglobin levels. Azlin et al., studied pregnant diabetic women with Type-1 diabetes mellitus and observed a decrease in mean blood glucose levels post Ramadan.10 There were no self- reported episodes of hypoglycemia requiring medical attention. In a study from Malaysia, including both Insulin dependent diabetes and GDM, no episodes of hypoglycemia were reported, among the study population.6 Women with gestational diabetes had a higher body mass index, and positive family history of diabetes mellitus. Though we did not have facility for continuous glucose monitoring, it has been suggested that without such monitoring of blood glucose levels, serious episodes of hypoglycemia can be missed. It has also been observed that women tend not to report about the hypoglycemic events which they encounter during fasting.11 Afandi et al conducted a study on pregnant gestational diabetes women in their second and third trimesters, either on diet or diet and metformin.12 These women were monitored with continuous blood glucose monitoring or self-blood glucose monitoring. The investigators found more incidence of hypoglycemia, in women with continuous glucose monitoring. Though, no adverse event or hospitalization were reported in both groups for hypoglycemia. In another study of 25 pregnant women with gestational diabetes, either on diet alone, diet plus metformin or on insulin therapy, significant reduction in blood glucose levels and HbA1c was observed during Ramadan, when compared to Pre-Ramadan.13 We did not measure HbA1c levels post Ramadan in our study group. But there was significant difference in the levels of HbA1c among gestational diabetic and normoglycemic women. Also there was increased prevalence of family history of diabetes mellitus in women with gestational diabetes. Though the studies on pregnancy with gestational diabetes are very few, there are no studies on diabetic ketoacidosis during Ramadan fasting among pregnant women. In a systemic review and meta-analysis of more than 18,000 women exposed to Ramadan fasting, there was no evidence of low birth weight or preterm deliveries.14 Preterm delivery was also not seen among our study participants. Fetal effects of maternal fasting can be anticipated, as fetus receives nutrition from mother directly through facilitated diffusion. In a retrospective cohort study from Turkyie, pregnant women who were found hypoglycemic (blood glucose level < 70mg/dl) on oral glucose tolerance test with 75-gm glucose, gave birth to lower birth weight babies with decreased length and head circumference.15 We also found decreased head circumference in our study participants, who were normoglycemic and fasted. Apart from this we did not find any evidence of hypoglycemia or raised bilirubin levels in our study group. Maternal fasting may induce a condition of stress in-utero for fetus, and has been found to be associated with adult hood diseases like hypertension.16 Large scale population-based studies are needed to study the long-term effects, of decreased head circumference, in later part of life. Neonatal apgar scores at one and five minutes were better in gestational diabetes women, but did not reach statistical significance. ## Limitations: We did not take into account the caloric intake of carbohydrate and fat rich diet in both groups of women. Since there was self-monitoring of blood glucose levels, hypoglycemia may have been missed in the study group. ## CONCLUSION Fasting is an important pillar of Islam. Gestational diabetes affects metabolic profile of pregnant women. Fasting in this group of women may be associated with maternal and fetal risks. Though we did not find any adverse maternal and fetal effects of maternal fasting among women with gestational diabetes, larger scale studies with blinded continuous glucose monitoring is much needed to provide robust guidelines to pregnant diabetic women and physicians. ## Authors Contribution: SA, S and MS: helped in data collection and synopsis writing SM: Did data analysis and contributed in write up NH: Conceived idea, IRB approval and manuscript writing and is responsible for integrity of the study. ## References 1. Hossain N, Samuel M, Mughal S, Shafique K. Ramadan Fasting: Perception and maternal outcomes during Pregnancy. Pak J Med Sci (2021) 37 1262-1267. PMID: 34475896 2. Joosoph J, Abu J, Yu SL. A survey of fasting during pregnancy. Singapore Med J (2004) 45 583-586. PMID: 15568120 3. 3Muslims and Islam: Key findings in the U.S. and around the world [Internet]Pew Research Center2017cited 24-7-2022Available from: https://www.pewresearch.org/fact-tank/2017/08/09/muslims-and-islam-key-findings-in-the-u-s-and-around-the-world/. Pew Research Center (2017) 4. Bajaj S, Khan A, Fathima FN, Jaleel MA, Sheikh A, Azad K. South Asian consensus statement on women's health and Ramadan. Indian J Endocrinol Metab (2012) 16 508-511. PMID: 22837905 5. Dikensoy E, Balat O, Cebesoy B, Ozkur A, Cicek H, Can G. The effect of Ramadan fasting on maternal serum lipids, cortisol levels and fetal development. Arch Gynecol Obstet (2009) 279 119-123. PMID: 18488237 6. Ismail NA, Olaide Raji H, Abd Wahab N, Mustafa N, Kamaruddin NA, Abdul Jamil M. Glycemic Control among Pregnant Diabetic Women on Insulin Who Fasted During Ramadan. Iran J Med Sci (2011) 36 254-259. PMID: 23115409 7. Jawad F, Ejaz K. Gestational diabetes mellitus in South Asia: Epidemiology. J Pak Med Assoc (2016) 66 S5-S7. PMID: 27582153 8. Gul Z, Rajar S, Shaikh ZF, Shafique K, Hossain N. Perinatal outcome among fasting and non fasting mothers during the month of Ramadan. Pak J Med Sci (2018) 34 989-993. PMID: 30190767 9. AlMogbel TA, Ross G, Wu T, Molyneaux L, Constantino MI, McGill M. Ramadan and gestational diabetes: maternal and neonatal outcomes. Acta Diabetol (2022) 59 21-30. PMID: 34427780 10. Nor Azlin MI, Adam R, Sufian SS, Wahab NA, Mustafa N, Kamaruddin NA. Safety and tolerability of once or twice daily neutral protamine hagedorn insulin in fasting pregnant women with diabetes during Ramadan. J Obstet Gynaecol Res (2011) 37 132-137. PMID: 21159037 11. Cosson E, Baz B, Gary F, Pharisien I, Nguyen MT, Sandre-Banon D. Poor Reliability and Poor Adherence to Self-Monitoring of Blood Glucose Are Common in Women With Gestational Diabetes Mellitus and May Be Associated With Poor Pregnancy Outcomes. Diabetes Care (2017) 40 1181-1186. PMID: 28724718 12. Afandi B, Hassanein M, Roubi S, Nagelkerke N. The value of Continuous Glucose Monitoring and Self-Monitoring of Blood Glucose in patients with Gestational Diabetes Mellitus during Ramadan fasting. Diabetes Res Clin Pract (2019) 151 260-264. PMID: 30822494 13. Hassanein M, Abuelkheir S, Alsayyah F, Twair M, Abdelgadir E, Basheir A. Evaluation of optimum diabetes care on glycemic control of patients with gestational diabetes during Ramadan fasting. Diabetes Res Clin Pract (2021) 173 108669. PMID: 33460717 14. Glazier JD, Hayes DJL, Hussain S, D'Souza SW, Whitcombe J, Heazell AEP. The effect of Ramadan fasting during pregnancy on perinatal outcomes: a systematic review and meta-analysis. BMC Pregnancy Childbirth (2018) 18 421. PMID: 30359228 15. Bayraktar B, Balıkoğlu M, Kanmaz AG. Pregnancy outcomes of women with hypoglycemia in the oral glucose tolerance test. J Gynecol Obstet Hum Reprod (2020) 49 101703. PMID: 32018048 16. Oosterwijk VNL, Molenaar JM, van Bilsen LA, Kiefte-de Jong JC. Ramadan Fasting during Pregnancy and Health Outcomes in Offspring: A Systematic Review. Nutrients (2021) 13 10
--- title: Risk factors of previously undiagnosed and known untreated hypertension among patients with Type-2 diabetes mellitus authors: - Muhammad Adnan - Wasif Noor - Mirza Muhammad Ayub Baig journal: Pakistan Journal of Medical Sciences year: 2023 pmcid: PMC10025734 doi: 10.12669/pjms.39.2.6329 license: CC BY 3.0 --- # Risk factors of previously undiagnosed and known untreated hypertension among patients with Type-2 diabetes mellitus ## Abstract ### Objectives: To find the risk factors of previously undiagnosed and known untreated hypertension among patients with Type- 2 diabetes mellitus. ### Methods: The cross-sectional analytical study was conducted at Diabetes Clinic of Sir Ganga Ram Hospital Lahore during Oct–Dec 2021. Total 153 known diabetics were enrolled using convenience sampling. Patients ($$n = 24$$) with ischemic heart disease, hepatitis or missing information excluded. Data from 129 cases of Type-2 diabetes presenting with and without hypertension analyzed using SPSS. Binary logistic regression analyses were performed to calculate the adjusted odds ratios. ### Result: Mean age of all diabetics ($$n = 129$$) was 49.0±10.7 years. The participation of females was higher than males ($65.1\%$ vs. $34.9\%$). The frequency of hypertension, previously undiagnosed hypertension and known untreated hypertension was $58.1\%$, $25.3\%$ and $19.6\%$, respectively. Among risk factors, frequency of high intake of salt was $67.4\%$, sedentary lifestyle was $65.1\%$, obesity was $37.2\%$, and poor glycemic control was $58.9\%$. Young age [aOR=2.01, $95.0\%$ CI 0.53–7.61], low family income <20000 PKR/month [aOR=2.70, $95.0\%$ CI 0.92–7.96], high intake of salt [aOR=3.22, $95.0\%$ CI 0.98–10.61], elevated total cholesterol [aOR=3.68, $95.0\%$ CI 0.85–15.85], poor glycemic control [aOR=3.28, $95.0\%$ CI 0.51–21.13], and overweight/ obesity [aOR=9.07, $95.0\%$ CI 1.6–51.39] had higher risk of previously undiagnosed or known untreated HTN. ### Conclusions: Prevalence of previously undiagnosed and known untreated hypertension is high among Type-2 diabetics. Strict compliance to diabetes care guidelines is much needed to minimize the risk of undiagnosed and untreated hypertension. ## INTRODUCTION Hypertension (HTN) is a well-known risk factor of heart and kidney diseases and its coexistence with diabetes mellitus (DM), particularly if the blood pressure (BP) levels are not controlled, can further add to this risk.1 Similarly if the HTN remains undiagnosed or untreated, it can lead to uncontrolled BP and may result in poor health outcomes.2 For these reasons, the American diabetes association (ADA) recommends regular monitoring of BP levels among diabetics and treating them to the targets of <$\frac{140}{90}$ mmHg.3 The prevalence of HTN is higher in diabetic patients than of non-diabetic individuals.4 All-cause mortality and cardiovascular disease related mortality rates are higher in hypertensive diabetics than of non-hypertensive diabetics.5 Unfortunately, the prevalence of HTN $26.34\%$6 and DM $13.7\%$7 are on rise among adult population of Pakistan. However, the studies reporting the undiagnosed and/or untreated HTN and their risk factors in diabetic patients are still lacking. Therefore, the present study aimed to assess the risk factors of previously undiagnosed and known untreated HTN among patients with T2DM. ## METHODS The single-center cross-sectional analytical study was conducted at Diabetes Clinic of Sir Ganga Ram Hospital Lahore during Oct–Dec 2021. Sample size was calculated using 7.0 % previously undiagnosed HTN in Type- 2 diabetics 8, $95.0\%$ confidence level, $5.0\%$ absolute precision and $80.0\%$ anticipated response rate. Total 180 patients were asked to participate in the study. Inclusion criteria were known T2DM patients, with and without HTN, age ≥18 years and any gender. With a response rate of $85.0\%$, 153 cases enrolled using convenience sampling. Patients ($$n = 24$$) with ischemic heart disease (IHD), hepatitis C and missing information were excluded. Consequently, data from 129 diabetics with and without HTN were analyzed. Upon enrollment, an interviewer-administered proforma used to collect the data including age, gender, education, income, history of smoking, physical activity and salt intake. Patient’s file used to record antihypertensive medications, HbA1c & BP levels. Body weight, height and waist circumference were measured. ## Operational Definitions: BP level ≥$\frac{140}{90}$ mmHg with history of previous HTN or taking anti-HTN medication defined as known HTN; BP ≥$\frac{140}{90}$ mmHg without history of previous HTN defined as previously undiagnosed HTN; and known HTN not taking anti-HTN medication defined as known untreated HTN. WC ≥90.0 cm in men and ≥80.0 cm in women defined as central obesity; BMI 25.0–29.9 Kg/m2 defined as overweight and BMI ≥30.0 Kg/m2 as obesity; and HbA1c level ≥$8.0\%$ defined as poor glycemic control. ## Ethical Approval: Institutional Review Board (IRB) of the Fatima Jinnah Medical University Lahore Pakistan approved the study vide letter No.53-Res-Proposal-PHRC/FJ/IRB dated 27th November 2021. Written informed consent was obtained from all patients. ## Statistical Analysis: The IBM® SPSS® Statistics version 26 was used for data analysis. The mean±SD calculated for continuous variables; and number (percent) for categorical variables. Crosstabs analyses were performed to calculate odds ratios and binary logistic regression analyses to calculate adjusted odds ratios with $95.0\%$ confidence intervals. Microsoft Excel used to construct the doughnut chart presenting prescription patterns of antihypertensive drugs. The value of p ≤ 0.05 was considered as significant. ## RESULTS The sociodemographic and clinical characteristics of study participants are shown in Table-I. Upon enrollment ($$n = 129$$), $43.4\%$ diabetics reported with known HTN; while HTN status of $56.6\%$ diabetics was not known. Among them, 19 new cases of HTN resulted in $25.3\%$ previously undiagnosed HTN. Hence, the frequency of overall HTN was $58.1\%$. Among hypertensive diabetics ($$n = 75$$), $74.7\%$ diabetics had known HTN and $19.6\%$ of them were non-adherent to anti-HTN medication. The frequency of patients with elevated DBP was higher than of elevated SBP in adherent ($15.6\%$ vs. $4.4\%$), non-adherent ($36.4\%$ vs. $0.0\%$) and previously undiagnosed HTN groups ($36.8\%$ vs. $0.0\%$), Table-II. Crosstabs analyses showed that females had higher risk of previously undiagnosed HTN [OR=1.34, $95.0\%$ CI 0.57–3.13], whereas males had higher risk of untreated HTN [OR=1.31, $95.0\%$ CI 0.44–3.89]. Being illiterate had higher risk of previously undiagnosed HTN [OR=1.44, $95.0\%$ CI 0.67–3.11], whereas being literate had higher risk of untreated HTN [OR=1.80, $95.0\%$ CI 0.43–7.43]. Sedentary lifestyle had higher risk of previously undiagnosed HTN [OR=1.64, $95.0\%$ CI 0.66–4.05], whereas active lifestyle had higher risk of untreated HTN [OR=2.53, $95.0\%$ CI 0.89–7.21]. Binary logistic regression analyses showed that age ≤55 years [aOR=2.01, $95.0\%$ CI 0.53–7.61], family income <20000 PKR/month [aOR=2.70, $95.0\%$ CI 0.92–7.96], salt intake med/high [aOR=3.22, $95.0\%$ CI 0.98–10.61], total cholesterol ≥200 mg/dl [aOR=3.68, $95.0\%$ CI 0.85–15.85], and HbA1c ≥7.0 % [aOR=3.28, $95.0\%$ CI 0.51–21.13] had 2–3 time higher risk; and BMI ≥25.0 Kg/m2 [aOR=9.07, $95.0\%$ CI 1.6–51.39] showed the highest risk of previously undiagnosed or known untreated HTN, Table-III. Overall $80.6\%$ diabetics with and without undiagnosed & known untreated HTN were predicted correctly at Step-1 and the prediction rate increased to $82.2\%$ at Step-5, Table-IV. Angiotensin-converting enzyme inhibitors (ACEIs) $32.14\%$ was the most frequently prescribed drug as monotherapy, followed by $19.64\%$ calcium channel blockers (CCBs), $10.71\%$ β blockers (BBs), and $7.14\%$ Angiotensin II receptor blockers (ARBs). As combined therapy, CCBs with ARBs were prescribed to $10.71\%$ patients. ## DISCUSSION HTN, if remains undiagnosed or untreated, can lead to uncontrolled BP and results in poor health outcomes.2 Therefore, the study aimed to assess the factors associated with previously undiagnosed and known untreated HTN in patients with T2DM. In the present study, overall rate of HTN $58.1\%$ exhibit that HTN was a common comorbid condition of T2DM in the settings. Although, it was equivalent to $59.5\%$ HTN observed in Ethiopian diabetics9, but was markedly lower than of $70.5\%$10 and $74.0\%$11 in Pakistani diabetics, $72.4\%$ in Jordanian diabetics8, $79.4\%$ in Spanish diabetics12, and $83.4\%$ in Emirati diabetics.13 The previously undiagnosed HTN $25.3\%$ suggesting that every 4th diabetic remain with undiagnosed HTN in the settings was lower than $37.4\%$ undiagnosed HTN in Spanish diabetics12, but three times higher than $7.0\%$ in Jordanian diabetics.8 The known untreated HTN $19.6\%$ suggesting that every 5th diabetic was non-compliant to antidiabetic treatment in the settings was higher than $11.7\%$ untreated HTN in Spanish diabetics.12 In addition, the present study found that young age, low family income, high salt intake, elevated total cholesterol, poor glycemic control, and overweight/ obesity had higher risk of previously undiagnosed or known untreated HTN. Similarly, younger age14,15, gender female16, low or no education17,18, low income15,19,20, and being overweight and obese19,21 had been reported as risk factors of undiagnosed HTN. Whereas, old age19-21, gender male12,15,19, and being underweight 15 had been reported as risk factors of undiagnosed HTN. In the present study, $73.3\%$ diabetics could not achieve their BP levels within target limits, which was notably higher than $50.4\%$ uncontrolled HTN in Jordanian diabetics.8 Isolated diastolic hypertension (IDH) is a less common type of HTN and accounts for <$20.0\%$ of HTN cases.22 It is an independent risk factor for stroke and heart disease.23 Surprisingly, a higher rate of IDH $24.0\%$ observed in the study. Adherence rate to anti-HTN medications was $80.4\%$; and ACEI ($32.14\%$) was the most frequently prescribed monotherapy, followed by CCB ($19.64\%$) and BB ($10.71\%$). Differently, Menendez et al. reported a little higher adherence rate $88.3\%$; and ACEI ($39.0\%$) as the most frequently prescribed monotherapy, followed by ARB ($19.9\%$) and diuretics ($19.5\%$).12 Kanj et al. reported that ACEI+ARB ($26.0\%$) was the most frequently prescribed drug, followed by BB ($15.0\%$) and diuretics ($10.0\%$).14 ## Limitations: The single-center observational study included small sample size, convenience enrollment of cases and higher participation rate of poor class with poorly controlled diabetes. ## CONCLUSIONS Large numbers of T2DM patients remain with previously undiagnosed and known untreated HTN in our population. The modifiable factors such as no education, sedentary lifestyle and unhealthy diet are also contributing to the risk of undiagnosed and untreated HTN. Thus, strict compliance to diabetes care guidelines by both the physicians and the patients is much needed to minimize the risk of undiagnosed and untreated HTN. ## Author’s Contribution: MA: Conceived, designed the study; collection, entry, analysis and interpretation of data, and wrote original draft. WN & MMAB: Collection and interpretation of data. All authors critically reviewed and revised the manuscript, approved the final version to be published and take responsibility for the content and similarity index of the manuscript. ## References 1. Raza SA, Hassan M, Badar F, Rasheed F, Meerza F, Azam S. Cardiovascular disease risk factors in Pakistani population with newly diagnosed Type 2 diabetes mellitus: A cross-sectional study of selected family practitioner clinics in four provinces of Pakistan (CardiP Study). J Pak Med Assoc (2019) 69 306-312. PMID: 30890819 2. Ali A, Taj A, Amin MJ, Iqbal F, Iqbal Z. Correlation between microalbuminuria and hypertension in Type 2 diabetic patients. Pak J Med Sci (2014) 30 511-514. PMID: 24948969 3. 3American Diabetes Association. 10. Cardiovascular disease and risk management: standards of medical care in diabetes-2020Diabetes Care2020v43Suppl-1S111S134doi: 10.2337/dc20-S010. Diabetes Care (2020) v43 S111-S134 4. Lal B, Chnadio MA, Sattar N, Taqi T, Memon S. Prevalence of hypertension in diabetes. J Peoples Uni Med Health Sci (2018) 8 180-184 5. Mughal SA, Aziz H. Risk factors, mortality and recovery of Stroke: A prospective study on 1000 Patients. Pak J Med Sci (2010) 26 520-525 6. Shah N, Shah Q, Shah AJ. The burden and high prevalence of hypertension in Pakistani adolescents: a meta-analysis of the published studies. Arch Public Health (2018) 76 20. PMID: 29619218 7. Adnan M, Aasim M. Prevalence of type 2 diabetes mellitus in adult population of Pakistan: a meta-analysis of prospective cross-sectional surveys. Ann Glob Health (2020) 86 7. PMID: 32025503 8. Mubarak FM, Froelicher ES, Jaddou HY, Ajlouni KM. Hypertension among 1000 patients with type 2 diabetes attending a national diabetes center in Jordan. Ann Saudi Med (2008) 28 346-351. PMID: 18779643 9. Akalu Y, Belsti Y. Hypertension and its associated factors among type 2 diabetes mellitus patients at Debre Tabor General Hospital, Northwest Ethiopia. Diabetes Metab Syndr Obes (2020) 13 1621-1631. PMID: 32494180 10. Adnan M, Rahat T, Hashmat N, Ali Z. Metabolic syndrome;agreement between diagnostic criteria among type 2 diabetes mellitus patients. Prof Med J (2017) 24 539-544 11. Chaudhary GMD, Chaudhary FMD, Tanveer A, Tameez Ud Din A, Chaudhary SMD, Tameez Ud Din A. Demographic and clinical characteristics of 4556 type 2 diabetes mellitus patients at a tertiary care hospital in southern Punjab. Cureus (2019) 11 e4592. PMID: 31309017 12. Menendez E, Delgado E, Fernandez-Vega F, Prieto MA, Bordiu E, Calle A. Prevalence, diagnosis, treatment, and control of hypertension in Spain. Results of the di@bet.es study. Rev Esp Cardiol (2016) 69 572-578. PMID: 26979767 13. Jelinek HF, Osman WM, Khandoker AH, Khalaf K, Lee S, Almahmeed W. Clinical profiles, comorbidities and complications of type 2 diabetes mellitus in patients from United Arab Emirates. BMJ Open Diabetes Res Care (2017) 5 e000427 14. Kanj H, Khalil A, Kossaify M, Kossaify A. Predictors of undiagnosed and uncontrolled hypertension in the local community of Byblos, Lebanon. Health Serv Insights (2018) 11 1178632918791576. PMID: 30127615 15. Ahmed S, Tariqujjaman M, Rahman MA, Hasan MZ, Hasan MM. Inequalities in the prevalence of undiagnosed hypertension among Bangladeshi adults: Evidence from a nationwide survey. Int J Equity Health (2019) 18 33. PMID: 30770739 16. Cuschieri S, Vassallo J, Calleja N, Pace N, Mamo J. The effects of socioeconomic determinants on hypertension in a cardiometabolic at-risk European country. Int J Hypertens (2017) 2017 7107385. PMID: 28932598 17. Zoellner J, Thomson JL, Landry AS, Anderson-Lewis C, Connell C, Molaison EF. Improvements in blood pressure among undiagnosed hypertensive participants in a community-based lifestyle intervention, Mississippi 2010. Prev Chronic Dis (2014) 11 E53. PMID: 24698531 18. Osman el FM, Suleiman I, Alzubair AG. Clinico-epidemiological features of hypertensive subjects in kassala town, eastern Sudan. J Family Community Med (2007) 14 77-80. PMID: 23012150 19. Lim OW, Yong CC. The risk factors for undiagnosed and known hypertension among Malaysians. Malays J Med Sci (2019) 26 98-112. PMID: 31728122 20. Zhang H, Deng M, Xu H, Wang H, Song F, Bao C. Pre- and undiagnosed-hypertension in urban Chinese adults: A population-based cross-sectional study. J Hum Hypertens (2017) 31 263-269. PMID: 27654328 21. Bushara SO, Noor SK, Elmadhoun WM, Sulaiman AA, Ahmed MH. Undiagnosed hypertension in a rural community in Sudan and association with some features of the metabolic syndrome: how serious is the situation?. Ren Fail (2015) 37 1022-1026. PMID: 26042342 22. Opadijo OG, Salami TA, Sanya EO, Omotoso AB. Systolic hypertension in adult Nigerians with hypertension. J Coll Physicians Surg Pak (2007) 17 8-11. PMID: 17204211 23. Lotfaliany M, Akbarpour S, Mozafary A, Boloukat RR, Azizi F, Hadaegh F. Hypertension phenotypes and incident cardiovascular disease and mortality events in a decade follow-up of a Middle East cohort. J Hypertens (2015) 33 1153-1161. PMID: 25699976
--- title: 'Epigenome-wide association study of plasma lipids in West Africans: the RODAM study' authors: - Eva L. van der Linden - Karlijn A.C. Meeks - Felix Chilunga - Charles Hayfron-Benjamin - Silver Bahendeka - Kerstin Klipstein-Grobusch - Andrea Venema - Bert-Jan van den Born - Charles Agyemang - Peter Henneman - Adebowale Adeyemo journal: eBioMedicine year: 2023 pmcid: PMC10025759 doi: 10.1016/j.ebiom.2023.104469 license: CC BY 4.0 --- # Epigenome-wide association study of plasma lipids in West Africans: the RODAM study ## Body Research in contextEvidence before this studyWest African populations have a more favourable lipid profile than European populations, with lower levels of plasma triglycerides and higher levels of high-density lipoprotein (HDL) cholesterol, without an accompanying lower prevalence of cardiovascular disease outcomes. These differences might be influenced by DNA methylation. We searched PubMed in January and September of 2021 for articles describing DNA methylation associated with plasma lipids using a combination of Mesh terms “dyslipidaemia”, “cholesterol”, “DNA methylation” and “epigenomics”. Additionally, we searched the EWAS Atlas, a curated database of epigenome-wide association studies, for studies reporting on plasma lipids. We found several publications, including a meta-analysis of cohorts, reporting on methylation loci associated with plasma lipid concentrations. However, few studies were conducted in populations of African ancestry (African Americans), and only one study was conducted in a population-based study in sub-Sahara Africa itself (South African Batswana). None of the studies was conducted in a West African population. Added value of this studyIn this epigenome-wide analysis we reported on the association between DNA methylation and plasma lipids conducted in a West African population, i.e. in Ghanaians, a population for whom epigenetic data are scarce. We identified several methylation loci that have previously been linked to lipid metabolism and contribute substantially to the variance in plasma lipid concentration in this Ghanaian population. Several of identified loci replicated in populations of other ethnicities, suggesting that these loci may play a role in lipid metabolism across populations, including West Africans. Additionally, we identified other loci that are potentially relevant in lipid metabolism in Ghanaians specifically. These might contribute to the favourable lipid profile of West African populations, and these may be potentially relevant biomarkers in the pathogenesis of dyslipidaemia. Implications of all the available evidenceThe results of this study can serve as a reference for future replication studies and contribute to elucidating mechanisms underlying lipid metabolism in diverse populations. Increasing ethnic diversity in epigenetic research is critical to prevent exacerbation of existing health disparities. Future studies should include larger sample size and a longitudinal study design to increase our pathophysiological understanding of dyslipidaemia among West African populations, thereby informing targeted strategies to curb the rising prevalence of cardiometabolic disorders in sub-Saharan African populations. ## Summary ### Background DNA-methylation has been associated with plasma lipid concentration in populations of diverse ethnic backgrounds, but epigenome-wide association studies (EWAS) in West-Africans are lacking. The aim of this study was to identify DNA-methylation loci associated with plasma lipids in Ghanaians. ### Methods We conducted an EWAS using Illumina 450k DNA-methylation array profiles of extracted DNA from 663 Ghanaian participants. Differentially methylated positions (DMPs) were examined for association with plasma total cholesterol (TC), LDL-cholesterol, HDL-cholesterol, and triglycerides concentrations using linear regression models adjusted for age, sex, body mass index, diabetes mellitus, and technical covariates. Findings were replicated in independent cohorts of different ethnicities. ### Findings We identified one significantly associated DMP with triglycerides (cg19693031 annotated to TXNIP, regression coefficient beta −0.26, false discovery rate adjusted p-value 0.001), which replicated in-silico in South African Batswana, African American, and European populations. From the top five DMPs with the lowest nominal p-values, two additional DMPs for triglycerides (CPT1A, ABCG1), two DMPs for LDL-cholesterol (EPSTI1, cg13781819), and one for TC (TXNIP) replicated. With the exception of EPSTI1, these loci are involved in lipid transport/metabolism or are known GWAS-associated loci. The top 5 DMPs per lipid trait explained $9.5\%$ in the variance of TC, $8.3\%$ in LDL-cholesterol, $6.1\%$ in HDL-cholesterol, and $11.0\%$ in triglycerides. ### Interpretation The top DMPs identified in this study are in loci that play a role in lipid metabolism across populations, including West-Africans. Future studies including larger sample size, longitudinal study design and translational research is needed to increase our understanding on the epigenetic regulation of lipid metabolism among West-African populations. ### Funding $\frac{10.13039}{501100000780}$European Commission under the Framework Programme (grant number: 278901). ## Evidence before this study West African populations have a more favourable lipid profile than European populations, with lower levels of plasma triglycerides and higher levels of high-density lipoprotein (HDL) cholesterol, without an accompanying lower prevalence of cardiovascular disease outcomes. These differences might be influenced by DNA methylation. We searched PubMed in January and September of 2021 for articles describing DNA methylation associated with plasma lipids using a combination of Mesh terms “dyslipidaemia”, “cholesterol”, “DNA methylation” and “epigenomics”. Additionally, we searched the EWAS Atlas, a curated database of epigenome-wide association studies, for studies reporting on plasma lipids. We found several publications, including a meta-analysis of cohorts, reporting on methylation loci associated with plasma lipid concentrations. However, few studies were conducted in populations of African ancestry (African Americans), and only one study was conducted in a population-based study in sub-Sahara Africa itself (South African Batswana). None of the studies was conducted in a West African population. ## Added value of this study In this epigenome-wide analysis we reported on the association between DNA methylation and plasma lipids conducted in a West African population, i.e. in Ghanaians, a population for whom epigenetic data are scarce. We identified several methylation loci that have previously been linked to lipid metabolism and contribute substantially to the variance in plasma lipid concentration in this Ghanaian population. Several of identified loci replicated in populations of other ethnicities, suggesting that these loci may play a role in lipid metabolism across populations, including West Africans. Additionally, we identified other loci that are potentially relevant in lipid metabolism in Ghanaians specifically. These might contribute to the favourable lipid profile of West African populations, and these may be potentially relevant biomarkers in the pathogenesis of dyslipidaemia. ## Implications of all the available evidence The results of this study can serve as a reference for future replication studies and contribute to elucidating mechanisms underlying lipid metabolism in diverse populations. Increasing ethnic diversity in epigenetic research is critical to prevent exacerbation of existing health disparities. Future studies should include larger sample size and a longitudinal study design to increase our pathophysiological understanding of dyslipidaemia among West African populations, thereby informing targeted strategies to curb the rising prevalence of cardiometabolic disorders in sub-Saharan African populations. ## Introduction Dyslipidaemia is a major risk factor for cardiovascular diseases (CVDs) in general, and ischaemic heart disease in particular.1 West African origin populations are considered to have a more favourable lipid profile than other ethnic groups, with lower plasma levels of triglycerides and low-density lipoprotein cholesterol (LDL-C), and higher plasma levels of high-density lipoprotein cholesterol (HDL-C).2 However, whereas in high-income western regions mean non-HDL-C levels have decreased over the past few decades, an opposite trend can be observed in most parts of sub-Saharan Africa (SSA),3 with a potentially important impact on the CVD burden in this region. Lipid metabolism is determined by both genetic and environmental factors. Plasma lipid concentrations are 40–$60\%$ heritable, but common variants explain only 10–$25\%$ of the variance in lipid levels.4 Additionally, genome-wide association studies (GWAS) show different loci associated with plasma lipid concentration between African and European origin populations.5 Environmental factors such as urbanisation and “westernisation” are shifting patterns in behavioural factors towards less physical activity and more consumption of (fast) food high in salt, sugar, and saturated fat, impacting lipid metabolism.6 However, neither genetic variants nor environmental factors alone can completely explain the variation in plasma lipid phenotypes. Gene-environment interaction, mediated by epigenetic modifications, potentially accounts for a proportion of this unexplained variation.7 Epigenetic studies facilitate understanding of the regulation of gene expression that occur without changes in the DNA sequence itself.8 Several studies have reported on epigenetic processes associated with lipid profiles,9 with DNA methylation (DNAm) being studied most widely. While there are few epigenetic studies in African-ancestry populations in general, epigenetics studies in SSA populations are particularly scarce. Only one epigenome-wide association study (EWAS) assessing DNAm in lipid traits has been conducted in an SSA population.10 Additionally, as genetic heterogeneity and environmental diversity are large in SSA, epigenetic analyses in other SSA populations can contribute to the discovery of new epigenetic loci associated with lipids. This can improve our understanding of this complex trait in SSA populations, which is highly relevant in the context of CVD prevention. In this study, we aim to identify DNAm loci associated with plasma lipid concentrations in Ghanaians. ## Study population and study design This study used baseline data from the prospective, multicentre Research on Obesity and Diabetes among African Migrants (RODAM) study. Details on this study have been published before11 and are summarised here. Between 2012 and 2015, 6385 Ghanaian men and women were recruited in rural Ghana (Ashanti region), urban Ghana (Kumasi and Obuasi), and the European cities of London, Amsterdam, and Berlin. Most participants were of Akan ethnicity, and Ghanaians residing in Europe were first-generation migrants originating from the villages and towns in the Ashanti region. Of those participants aged 25 years and over, with complete data on physical examination and blood sample profile ($$n = 5659$$), 736 participants were selected for DNAm profiling (Supplementary Fig. S1). The selection process was based on a case–control design, including about 300 non-drug treated diabetic cases, 300 non-diabetic controls, and 135 non-diabetic, non-obese controls. This sample size was originally chosen to have $80\%$ power to detect a $5\%$ methylation difference between diabetic cases and controls. After exclusion of sex discordances ($$n = 11$$), duplicates ($$n = 8$$), and those not meeting the quality control thresholds ($$n = 12$$), 713 eligible participants remained. Participants with missing data on lipid profile ($$n = 6$$), or those using lipid-lowering medication ($$n = 38$$) were excluded from the analysis. Additionally, six participants were excluded from the analysis because of outliers in lipid concentrations, resulting in 663 participants included in the current analyses. ## Ethics Before the start of data collection, ethical approval was obtained from the respective ethics committees of the involved institutions in Ghana (School of Medical Sciences/Komfo Anokye Teaching Hospital Committee on Human Research, Publication & Ethical Review Board, ref. CHRPE/AP/200-12), UK (London School of Hygiene and Tropical Medicine Research Ethics Committee, ref. 6208), the Netherlands (Institutional Review Board of the Academic Medical Center, University of Amsterdam, ref. W12_062#12.17.0086) and Germany (Ethics Committee of Charité-Universitätsmedizin Berlin, ref. EA$\frac{1}{307}$/12). All participants provided written informed consent before enrolment in the study. ## Phenotypic measurements Data collection procedures for questionnaire and physical examination were highly standardised across the different study locations, to allow for comparison between the sites. Data on sex, age, and length of stay in Europe were obtained using questionnaires. The use of lipid-lowering medication was based on the Anatomical Therapeutic Chemical classification of medication that participants brought with them to the research location. Physical examination was performed using validated devices. Weight was measured in light clothing without shoes with a SECA 877 scale (Seca GmbH & Co. KG, Hamburg, Germany) to the nearest 0.1 kg. Height was measured without shoes using a SECA 2017 portable stadiometer to the nearest 0.1 cm (Seca GmbH & Co. KG, Hamburg, Germany). Anthropometric measures were taken twice and the mean was used in analyses. Body mass index (BMI) was calculated by dividing the weight in kilograms by the square of the height in meters. Venous blood samples were collected after an overnight fast of at least 10 h. All biochemical analyses were performed in Berlin to avoid inter-laboratory bias. Fasting plasma glucose concentration was measured using the hexokinase method by colorimetry. Diabetes mellitus was defined according to self-reported diabetes and/or fasting glucose ≥7.0 mmol/L. Participants using glucose-lowering medication were excluded from DNAm analysis, because of the potential confounding effect of medication use on methylation profile. A ready-to-use reagent for colorimetry was used to obtain concentrations of total cholesterol (TC), HDL-C, and triglycerides. All analyses were performed using and ABX Pentra 400 chemistry analyser (Horiba ABX SAS, Oberursel, Germany). LDL-C concentration was calculated using the Friedewald equation for individuals with triglyceride levels <4.5 mmol/L. The distribution of the lipid concentration was assessed using histograms and the Shapiro–Wilk test. To ensure normal distribution of the lipid traits, rank-based inverse normal transformation was performed for TC, LDL-C, and HDL-C. Triglyceride concentration was natural log-transformed because of its skewed distribution. ## DNA methylation profiling, processing, and quality control Source BioScience, Nottingham, UK, conducted the DNA extraction and methylation profiling on participant’s whole blood samples. The process of DNAm profiling, processing, and quality control on RODAM whole blood samples has been described previously.12,13 In short, the Zymo EZ DNAm™ kit (Zymo Research Corp., Irvine, CA, USA) was used for bisulphite conversion of DNA. Using the Infinium® HumanMethylation450 BeadChip (Illumina, San Diego, CA, USA), the converted DNA was amplified and hybridised, thereby quantifying DNAm levels of approximately 485,000 CpG sites. Methylation levels were measured based on the intensities of the methylated and unmethylated probes for each CpG site on the array. These intensities were expressed as methylation Beta-value, which is a value between zero (unmethylated) and one (methylated). A log2 ratio of the intensities of methylated versus unmethylated probes was calculated, which is referred to as M-values. Quality control was performed using the MethylAid package (version 1.28.0) in R statistical software (version 4.1.2). The minfi package (version 1.40.0) was used for functional normalisation of the raw 450K data. A total set of 429,459 CpG sites remained after removal of probes annotated to the X and Y chromosomes, known to involve cross-hybridisation or to involve common SNPs with a minor allele frequency of ≥$5\%$.14 Blood cell mixture estimation was based on the method described by Houseman et al.15 ## Association between lipids and DNA methylation To identify differentially methylated positions (DMPs), the association between lipid concentration (independent variable) and DNAm M-values (dependent variable), were examined using multivariate linear regression analysis using the lmFit function of the Limma package (version 3.50.1). M-values were used for DMP analyses because of the non-normal distribution of Beta-values. Beta-values were reported for visualisation and to help interpretation of the results.16 Because of correlation with DNAm, sex, age, geographical location, estimated cell count (CD8+, CD4+, natural killer cells, B cells, monocytes, and granulocytes), hybridisation batch and array position were included as covariates in the models, based on principal components analysis (Supplementary Fig. S2). Additionally, BMI and diabetes were included in the model, because of an overrepresentation of participants with diabetes and high BMI in the sample. QQ-plots were used to assess model fit (Supplementary Fig. S3). The DMP analysis was run stratified by geographical location, because of the previously observed large difference in plasma lipid profile between the sites (rural and urban Ghana, London, Amsterdam, and Berlin), and to reduce the impact of unobserved confounding factors that differ between the geographical locations. The results for the EWAS per site were then meta-analysed using METAL statistical software (version 2011-03-25). A fixed-effect model, based on effect size and accompanying standard errors was applied. Direction of effect per site was summarised as ‘+’ for positive effect size, or ‘–‘ for negative effect size. Heterogeneity between the sites was considered significant if the p-value for Chi-squared test for heterogeneity was <0.05. To correct for multiple testing, false discovery rate (FDR) adjusted p-values were calculated using the Benjamini-Hochberg method. FDR-adjusted p-values of <0.05 were considered epigenome-wide significant. To examine the association between DNA methylation and lipid concentration, as well as the explained variance, the raw Beta-values of the top DMPs for each lipid trait were extracted and used as independent variable in models with untransformed lipid concentration as the dependent variable. Methylation Beta-values were used for this analysis to facilitate interpretation as the increase in plasma lipid concentration in mmol/L per percent increase in methylation Beta-value. The models included the same covariates as the DMP analysis. The multiple R squared statistic of the regression models with and without covariates was used to calculate the variance explained by the DMP. As this analysis was run in the total study population, the analysis was additionally adjusted for geographical location. ## Replication and transferability To determine whether the top DMPs with the lowest FDR-adjusted p-values in our study replicated in independent cohorts from different ethnic backgrounds, we performed a look-up using summary statistics from EWAS analyses among Batswana in South Africa,10 African Americans in the USA, and European ancestry populations in the USA and Europe.9 The criteria for replication were a nominal p-value of <0.05 in the replication cohort and a consistent direction of effect. Supplementary Table S1 provides detailed information on the population and design of the replication studies. We also evaluated whether findings from populations of different ethnic backgrounds, i.e. South African Batswana, African Americans, and Europeans, were transferable to our Ghanaian study population. Cohort-specific thresholds for epigenome-wide significance were used to determine which CpG sites to extract. For the African Americans and European ancestry populations, these were CpG sites with a Bonferroni adjusted p-value <1∗10ˆ–9 in the meta-analysis by Jhun et al.9 For the South African Batswana population, this was a nominal p-value of <1∗10ˆ–5 in the study by Cronjé et al.10 The association between lipid concentration and these candidate CpGs was assessed in the Ghanaian study population using linear regression models following the same strategy and covariates as in the DMP analysis. Bonferroni adjusted p-values were calculated for each trait and per ethnic group. Results were considered statistically significant if the p-value was <0.05/nCpGs. For TC, DMPs were only reported in the study by Cronjé et al. and showed a significant association of DNAm of the TXNIP gene in South African Batswana (Table 4). LDL-C was significantly associated with DNAm of EPSTI1 in African American and European populations, as was cg13781819 in African Americans. For HDL-C, none of the five top DMPs could be replicated in the independent cohorts including participants from South Africa Batswana, African Americans or Europeans (Table 4). For triglycerides, TXNIP was replicated in all three ethnic groups. Additionally, CPT1A and ABCG1 were replicated in African American and European descent populations. Table 4Replication of the top differentially methylated positions per lipid trait in South African, African American and European descent populations. TCCpGchrPosGeneSouth AfricanAfrican AmericanEuropeanRegress. Coeff.p-valueRegress. Coeff.p-valueRegress. Coeff.p-valuecg19693031chr11.45E+08TXNIP−3.49E–040.013NANANANAcg03753191chr1343566902EPSTI1−1.45E–040.253NANANANAcg26816907chr11.98E+08LHX9−2.90E–040.116NANANANAcg03167407chr22.41E+08Intergenic1.37E–040.697NANANANAcg11066601chr11.85E+08IntergenicNANANANANANALDL-CCpGchrposGeneRegress. Coeff.p-valueRegress. Coeff.p-valueRegress. Coeff.p-valuecg03753191chr1343566902EPSTI1−1.57E–040.3045.70E–050.0031.80E–050.025cg26816907chr11.98E+08LHX9−2.32E–040.3001.87E–050.3551.31E–050.332cg13781819chr147469065Intergenic−5.65E–050.427−2.30E–050.034−1.48E–050.119cg20294940chr141.06E+08Intergenic9.48E–060.894−3.39E–060.435−4.47E–060.279cg23970275chr22.08E+08KLF71.00E–040.396−2.59E–050.087−2.13E–050.074HDL-CCpGchrposGeneRegress. Coeff.p-valueRegress. Coeff.p-valueRegress. Coeff.p-valuecg05091570chr12.02E+08NAV1−4.35E–050.5497.92E–040.3−7.90E–050.803cg07622193chr1942701920Intergenic−1.38E–040.515−0.0020.2−0.0010.059cg00091964chr280530891CTNNA2−9.74E–050.536−2.91E–040.77.53E–040.086cg13767294chr1741856619DUSP37.68E–050.1250.0010.19.98E–040.008cg08926253chr11614761IRF7−6.00E–050.9080.0030.23.93E–040.819TriglyceridesCpGchrposGeneRegress. Coeff.p-valueRegress. Coeff.p-valueRegress. Coeff.p-valuecg19693031chr11.45E+08TXNIP−0.0483.94E–05−0.033.19E–24−0.021.38E–18cg17058475chr1168607737CPT1A−1.70E–040.979−0.017.19E–06−0.011.49E–13cg06500161chr2143656587ABCG10.0130.3050.021.55E–130.021.01E–24cg05697101chr238829104HNRPLL0.0020.539−7.96E–040.3484.24E–040.195cg11066601chr11.85E+08IntergenicNANA7.37E–040.863−0.0020.436In bold, DMPs replicated in independent cohort at a nominal p-value <0.05. TC, total cholesterol; LDL-C, low-density lipoprotein cholesterol; HDL-C, high-density lipoprotein cholesterol. The transferability of lipid DMPs identified in previous EWAS to our study population of Ghanaians was generally low (Supplementary Table S3). Overall, transferability was higher for HDL-C than for LDL-C and triglycerides. Transferability from African Americans to Ghanaians was $14\%$ ($\frac{1}{7}$ CpGs) for HDL-C, $60\%$ ($\frac{3}{5}$) for LDL-C, and $9\%$ ($\frac{4}{43}$ CpGs) for triglycerides. Transferability was even lower from Europeans to our Ghanaian sample with $1\%$ ($\frac{1}{69}$ CpGs) for HDL-C, $0\%$ ($\frac{0}{15}$) for LDL-C, and $5\%$ ($\frac{4}{86}$ CpGs) for triglycerides. CpGs reported in South African Batswana did not transfer to our study population. ## Location of residence As previous RODAM results have shown distinct differences in lipid profiles between Ghanaians in Europe (migrants) and their non-migrating counterparts in Ghana (rural and urban),17 a sensitivity analysis was performed to evaluate the effect of the location of residence on our findings. The median DNAm Beta-values for each of the top 5 CpGs as identified in the DMP analysis were compared between the geographical locations, using the Kruskal–Wallis test because of the non-normal distribution of Beta-values. For most DMPs, we observed a significant trend in mean methylation level from rural Ghana to urban Ghana to Europe. Across all lipid traits, most of the DMPs were highest methylated in rural Ghana, followed by urban Ghana and Europe, whereas a few showed an opposite trend (Fig. 1a-d). The largest difference was seen for the TXNIP and KLF7 genes, with around $5\%$ lower methylation levels in Europe than in rural Ghana (Fig. 1a and b).Fig. 1Methylation levels for top differentially methylated positions stratified associated lipids, stratified by location of residence, for TC (a), LDL-C (b), HDL-C (c), and triglycerides (d). Median methylation level with interquartile range (IQR) in percentage, calculated by Beta-value∗100. ## Excluding participants with diabetes mellitus To examine the impact of diabetes status on our findings, we re-fitted the DMP regression model in a subsample of participants without diabetes ($$n = 432$$ participants). The summary statistics of the top CpGs from the EWAS in the total population were then compared to the summary statistics in the subsample of participants without diabetes. Effect sizes for the top DMPs per lipid trait remained generally the same after excluding participants with diabetes mellitus (Supplementary Table S4). ## Biological relevance The function gaphunter within the minfi package was used to examine whether the top DMPs were potentially under the influence of a genetic variation. The function was run with a threshold of 0.05, reflecting a gap of $5\%$ in Beta-value, suggestive of genetic influence. Identified CpGs with gaps were next searched in the GoDMC database,18 to see whether they have previously been correlated to genetic variation. To assess whether genes annotated to our top DMPs have previously been linked to lipid traits, GeneCards,19 the GWAS catalog,20 and the EWAS atlas21 were examined. To evaluate the levels of gene expression of the top DMPs per lipid trait as identified in our EWAS, the iMETHYL database was consulted.22 This database includes whole-DNA-methylation, whole-genome, and whole-transcriptome data for CD4+ T-lymphocytes, monocytes, and neutrophils collected from about 100 subjects. Gene expression is expressed in log-transformed fragments per kilobase of transcripts per million mapped reads (FPKM). A negative value of FPKM suggests low gene expression, whereas a positive value suggests high expression. Additionally, a search in the EWAS toolkit23 was performed, to assess DNA methylation level in subcutaneous and visceral adipose tissue, and in liver tissue for our top DMPs. Pathway enrichment analysis was performed using canonical pathway analysis QIAGEN Ingenuity Pathway Analysis application,24 including CpGs that were associated with lipid concentration at a significance level of nominal p-value <1∗10ˆ–4. Pathways with a nominal p-value <0.01, as calculated by the right-tailed Fisher’s Exact Test, were considered to be significantly associated. Gaphunter identified one DMP with a gap in Beta-value distribution of the intergenic CpG cg03167407 associated with TC concentration. This DMP did not show any association with SNPs in the GoDMC database. The mean methylation levels of the top DMPs for lipid traits from the RODAM study were in line with the methylation levels as reported in the iMETHYL database (Supplementary Table S5). Generally, for those loci that expression data were available for in iMETHYL, low methylation levels of CpGs annotated to the gene body were associated with low gene expression, whereas high methylation in the gene body was associated with high gene expression. The DNA methylation levels in blood, however, did differ from levels reported in subcutaneous and visceral adipose tissue, and in liver tissue as reported in the EWAS toolkit. The pathway enrichment analysis for TC showed enrichment for glutamine biosynthesis, Rho GDP dissociation inhibitor, and actin cytoskeleton signalling pathways. For LDL-C, pathways involved in calcium pathway signalling, and nitric oxide synthase signalling were enriched. Pathway enrichment analysis of HDL-C showed enrichment for the nicotinamide adenine dinucleotide (NAD) biosynthesis pathway. For triglycerides, pathways involved in the lipopolysaccharide/interleukine-1 (LPD/IL-1) inhibition of the retinoid X receptor (RXR), RXR activation, mitochondrial l-carnitine shuttle, and pyroptosis signalling were enriched. ## Role of funders The study funders had no role in the study design, data collection, data analysis, data interpretation or writing of the report. The corresponding author had full access to all the data and the final responsibility to submit for publication. ## Characteristics of the study population Population characteristics are described in Table 1. Of the 663 Ghanaian participants, most participants lived in urban Ghana, followed by Amsterdam, London, rural Ghana, and Berlin. More than half of the participants were female and the mean age was 50.7 years. BMI was lowest in participants in rural Ghana and highest in Ghanaians living in London. About one-third of the participants had diabetes mellitus. Regardless of the location of residence, only a small proportion of the participants smoked or drank alcohol. Levels of TC and LDL-C were highest in participants residing in urban Ghana and Europe. In contrast, HDL-C levels were lower, and triglyceride levels were higher in those living in rural Ghana than in the other geographical locations. Population characteristics stratified by sex can be found in Supplementary Table S2.Table 1Population characteristics. TotalRural GhanaUrban GhanaAmsterdamBerlinLondonp-valuean (% of total)663101 (15.2)239 (36.0)139 (21.0)75 (11.3)109 (16.4)Sex, male (%)281 (42.4)32 (31.7)71 (29.7)83 (59.7)52 (69.3)43 (39.4)<0.001Age (mean (SD))50.67 (9.96)56.21 (8.86)50.57 (9.77)48.81 (8.00)46.68 (10.78)50.84 (10.94)<0.001BMI (mean (SD))26.73 (5.49)22.81 (4.35)26.17 (5.69)28.21 (4.73)27.30 (4.34)29.34 (5.41)<0.001Diabetes mellitus (%)231 (34.8)40 (39.6)85 (35.6)45 (32.4)28 (37.3)33 (30.3)0.619Alcohol intake (units/day) (median [IQR])0.00 [0.00, 0.07]0.00 [0.00, 0.07]0.00 [0.00, 0.03]0.00 [0.00, 0.13]0.13 [0.00, 0.71]0.00 [0.00, 0.00]<0.001Smoking (%)<0.001 No, but I used to smoke61 (9.5)11 (11.3)22 (9.4)12 (8.9)11 (14.9)5 (4.8) No, I have never smoked569 (88.4)86 (88.7)211 (90.2)120 (88.9)55 (74.3)97 (93.3) Yes14 (2.2)0 (0.0)1 (0.4)3 (2.2)8 (10.8)2 (1.9)Length of Stay in Europe (years) (mean (SD))18.55 (9.70)NANA19.00 (7.55)19.05 (10.38)17.53 (11.62)0.464Blood cell distribution (%) (mean (SD)) CD8+ T lymphocytes0.11 (0.05)0.12 (0.05)0.12 (0.04)0.10 (0.05)0.10 (0.05)0.10 (0.04)<0.001 CD4+T0.18 (0.06)0.18 (0.06)0.18 (0.06)0.19 (0.05)0.18 (0.06)0.18 (0.06)0.822 NK cells0.11 (0.06)0.13 (0.06)0.11 (0.06)0.09 (0.05)0.11 (0.05)0.10 (0.05)<0.001 B cells0.11 (0.03)0.11 (0.04)0.11 (0.03)0.10 (0.03)0.10 (0.03)0.10 (0.03)0.003 Monocytes0.08 (0.02)0.08 (0.02)0.08 (0.03)0.08 (0.02)0.08 (0.03)0.08 (0.02)0.082 Granulocytes0.45 (0.09)0.42 (0.10)0.44 (0.09)0.48 (0.09)0.47 (0.09)0.47 (0.09)<0.001Lipid profile (mmol/L) (median [IQR]) TC5.18 [4.43, 5.93]4.57 [3.91, 5.56]5.43 [4.58, 6.22]5.10 [4.39, 5.79]4.99 [4.54, 6.02]4.98 [4.53, 5.70]<0.001 LDL-C3.30 [2.69, 3.94]2.79 [2.35, 3.65]3.56 [2.91, 4.16]3.24 [2.66, 3.92]3.13 [2.60, 3.83]3.18 [2.82, 3.90]<0.001 HDL-C1.29 [1.10, 1.51]1.18 [1.01, 1.36]1.27 [1.08, 1.50]1.33 [1.10, 1.60]1.42 [1.22, 1.66]1.35 [1.15, 1.55]<0.001 Triglycerides0.97 [0.72, 1.38]1.09 [0.81, 1.47]1.10 [0.83, 1.54]0.84 [0.62, 1.16]0.92 [0.68, 1.35]0.87 [0.61, 1.12]<0.001HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; TC, total cholesterol; SD, standard deviation; IQR, interquartile range.ap-values represent the comparison between the geographical locations, using one-way ANOVA to compare normally distributed continuous variables, Kruskal–Wallis test for non-normally distributed continuous variables, and Chi-square test for categorical variables. ## Total cholesterol None of the CpGs associated with TC concentration was epigenome-wide significant at $5\%$ FDR (Supplementary Fig. S4a). The five CpGs with the smallest nominal p-values (all p-value ≤6∗10ˆ–6), were annotated to the TXNIP, the EPSTI1, the LHX9 genes, and to two intergenic CpGs cg11066601 and cg03167407 (Table 2). The associations had generally the same direction of effect in all five geographical locations. An increase in DNAm level of TXNIP, was associated with a 4.01 mmol/L decrease in TC level. DNAm levels of cg03167407, the LHX9, and the EPSTI1 DMPs were associated with an increase in TC level ranging from 1.32 to 8.33 mmol/L $9.5\%$ of the variance in TC concentration was attributable to the top CpGs (Table 3).Table 2Top differentially methylated positions associated with lipids. TCRegression CoeffaDirection of effectbp-valueFDR adj.pvalchrPosGene symbolcGene groupMethylation level, % (sd)dcg19693031−0.0957−−−−−8.43E–070.3622chr1145441552TXNIP3′UTR78.53 (6.36)cg037531910.0997+++++3.54E–060.6184chr1343566902EPSTI1TSS15008.65 (3.11)cg268169070.0718+++++5.49E–060.6184chr1197890812LHX9Body29.67 (6.25)cg11066601−0.2233−+−−−6.53E–060.6184chr1185373486Intergenic78.69 (11.31)cg031674070.1782+++++9.30E–060.6184chr2241261657Intergenic77.63 (12.99)LDL-CRegression CoeffDirection of effectp-valueFDR adj.pvalchrposGene symbolGene groupMethylation level, % (sd)cg037531910.0976+++++4.14E–060.9999chr1343566902EPSTI1TSS15008.65 (3.11)cg268169070.0679+++++1.36E–050.9999chr1197890812LHX9Body29.67 (6.25)cg13781819−0.053−−−−−3.40E–050.9999chr147469065Intergenic88.94 (2.14)cg20294940−0.049−−−−−5.30E–050.9999chr14105866596Intergenic92.38 (1.61)cg23970275–0.0674−−−−−5.75E–050.9999chr2208008052KLF7Body16.88 (6.27)HDL-CRegression Coeff. Direction of effectp-valueFDR adj.pvalchrposGene symbolGene groupMethylation level, % (sd)cg05091570−0.0746−−−−−9.09E–070.314chr1201709336NAV1Body2.83 (0.82)cg07622193−0.0624−−−−−1.68E–060.314chr1942701920Intergenic11.83 (3.33)cg00091964−0.0888−−−−−2.19E–060.314chr280530891CTNNA2Body3.91 (1.46)cg13767294−0.072−−−−−5.23E–060.517chr1741856619DUSP3TSS15004.47 (1.13)cg089262530.0481+++++6.03E–060.517chr11614761IRF7Body56.44 (4.35)TriglyceridesRegression Coeff. Direction of effectp-valueFDR adj.pvalchrposGene symbolGene groupMethylation level, % (sd)cg19693031−0.2637−−−−−1.67E–090.001chr1145441552TXNIP3′UTR78.53 (6.36)cg17058475−0.225−−−−−2.09E–060.448chr1168607737CPT1A5′UTR13.91 (5.10)cg065001610.1001+++++1.17E–050.9999chr2143656587ABCG1Body61.22 (3.95)cg05697101−0.3446−−−−−2.81E–050.9999chr238829104HNRPLLBody8.24 (3.75)cg11066601−0.4686−−−−−3.34E–050.9999chr1185373486Intergenic78.69 (11.31)HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; TC, total cholesterol; SD, standard deviation; UTR, untranslated region; TSS, transcription start site.aFor M-values, adjusted for covariates age, sex, BMI, diabetes mellitus, estimated cell count, batch and array position.bDirection of effect in each of the five sites, represented in order Amsterdam-Berlin-London-Rural Ghana-Urban Ghana; negative sign means negative direction of effect, positive sign means positive direction of effect.cAnnotated using UCSC catalogue.dMethylation level calculated as: methylation Beta-value∗100.Table 3Association between DNA methylation of top differentially methylated positions (independent variable) and lipid concentration (dependent variable).TCRegress. Coeff. ( $95\%$ CI)p-valueGene symbolTrait variance (%)cg19693031−4.01 (−5.54;–2.48)7.43E–07TXNIP3.5cg037531918.34 (4.75; 11.93)1.08E–06EPSTI12.7cg031674071.32 (0.68; 1.96)5.83E–05Intergenic2.2cg11066601−1.30 (−2.04;–0.57)1.84E–04Intergenic1.6cg268169072.93 (1.23; 4.63)3.42E–04LHX91.5LDL-CRegress. Coeff.p-valueGene symbolTrait variance (%)cg037531916.96 (3.9; 10.03)1.36E–06EPSTI12.6cg20294940−12.12 (−17.83;–6.40)2.88E–05Intergenic2.3cg268169072.70 (1.25; 4.15)1.23E–04LHX91.8cg23970275−3.97 (−5.95;–1.99)1.36E–04LKLF72.0cg13781819−6.78 (−10.29;–3.27)4.92E–04Intergenic1.9HDL -CRegress. Coeff.p-valueGene symbolTrait variance (%)cg089262531.42 (0.75; 2.1)3.73E–05IRF72.2cg07622193−2.53 (−3.83;–1.24)8.07E–05Intergenic0.9cg05091570−8.32 (−12.02;–4.61)2.26E–04NAV12.5cg00091964−3.75 (−5.87;–1.63)2.99E–04CTNNA21.6cg13767294−3.78 (−6.53;–1.03)9.14E–03DUSP31.0TriglyceridesRegress. Coeff.p-valueGene symbolTrait variance (%)cg19693031−3.06 (−3.91;–2.22)1.04E–11TXNIP6.2cg065001613.20 (1.92; 4.48)1.38E–06ABCG13.1cg17058475−2.90 (−4.17;–1.62)1.45E–06CPT1A2.5cg11066601−0.62 (−1.04;–0.21)1.21E–03Intergenic1.1cg05697101−2.04 (−3.63;–0.44)2.35E–03HNRPLL0.8Model = [lipid] (untransformed) ∼ Beta-value + sex + age + blood cell estimate + technical variables + BMI + diabetes + site. TC, total cholesterol; LDL-C, low-density lipoprotein cholesterol; HDL-C, high-density lipoprotein cholesterol. ## Low-density lipoprotein cholesterol None of the CpGs associated with LDL-C concentration was epigenome-wide significant at $5\%$ FDR (Supplementary Fig. S4b). The five CpGs with the smallest nominal p-values (all p-value <6∗10ˆ–5) were annotated to the EPSTI1, the LHX9, and to the KLF7 genes, and the intergenic CpGs cg13781819 and cg20294940 (Table 2). The association had the same direction of effect in all five geographical locations. A one percent ($1\%$) increase in DNAm level was associated with around 6.96 mmol/L increase in LDL-C levels for EPSTI1 and LHX9. For the other three CpGs, an increase in DNAm was associated with a 3.97–12.12 mmol/L decrease in LDL-C level. The top 5 CpGs contributed $8.3\%$ to the variance in LDL-C (Table 3). ## High-density lipoprotein cholesterol None of the CpGs was significantly associated with HDL-C concentration at <$5\%$ FDR (Supplementary Fig. S4c). The five CpGs with the smallest nominal p-values (all p-value ≤6∗10ˆ–6), were annotated to the NAV1, the CTNNA2, the CFAP97, and the IRF7 genes, and the intergenic CpG cg07622193 (Table 2). The associations had the same direction of effect in all five geographical locations. An increase in methylation level of the top DMPs was generally associated with a decrease in HDL-C level (Table 3), with a $1\%$ increase in DNAm being associated with a decrease in HDL-C levels up to 8.3 mmol/L. In contrast, cg08926253 showed a positive association between DNAm and HDL-C (regression coefficient beta 1.42). Overall, $6.1\%$ of the variance in HDL-C concentration was attributable to the five CpGs with the smallest nominal p-value. ## Triglycerides DNAm levels of cg19693031, were significantly associated with triglyceride concentrations at an epigenome-wide level (Supplementary Fig. S4d). This CpG is located in the 3’ UTR of the TXNIP gene. The other four CpGs with the smallest p-values were not epigenome-wide significantly associated, but all had a nominal p-value of <4∗10ˆ–5. These CpGs were annotated the CPT1A, the ABCG1, and the HNRPLL genes, and intergenic CpG cg11066601 (Table 2). The associations had the same direction of effect across all five geographical locations. $1\%$ higher DNAm levels of the ABCG1 DMP was associated with a 3.20 mmol/L increase in triglyceride levels. The DNAm levels of the other CpGs were associated with lower levels of triglycerides, ranging from −0.62 mmol/L for cg11066601, to −3.06 mmol/L for the TXNIP DMP. The combined effect of the top five CpGs explained $11.0\%$ of the variance in triglyceride concentration (Table 3). ## Discussion In this EWAS on lipid components in West African populations, we identified one epigenome-wide significant DMP associated with triglycerides (cg19693031 annotated to the TXNIP gene). We found that DNAm levels of DMPs annotated to theTXNIP, NAV1, CPT1A, and ABCG1 genes did contribute substantially to the variance in plasma TC, LDL-C, HDL-C, triglyceride concentrations. We were able to replicate our findings in independent cohorts of South African Batswana, African American, and European descent. Additionally, candidate DMPs identified in African American and European populations were transferable to Ghanaians in our study, but not from South African Batswana to Ghanaians. Mean DNAm levels for the top DMPs were generally lower in Ghanaians residing in Europe than in urban or rural Ghana. Our findings suggest that TXNIP methylation is associated with plasma lipids across populations,9,10 including West Africans who have a generally more favourable lipid profile than other populations. DMP cg19693031, located in the 3′UTR of the TXNIP gene, was epigenome-wide significantly associated with fasting triglyceride concentration in Ghanaians, and explained $6.2\%$ of the variation in triglyceride levels. In African American and European ancestry populations, this DMP has previously been linked to triglyceride and lipid metabolism,25,26 as well as to other cardiometabolic traits such as weight,27 blood glucose,13 blood pressure,28 and to BMI in a previous RODAM EWAS study.12 The TXNIP gene encodes for the thioredoxin interacting protein, which is primarily involved in inflammatory, metabolic and apoptotic processes,29 and plays an important role in the development of diabetes, by influencing insulin production and beta-cell apoptosis.30 The role of TXNIP in lipid metabolism was clearly demonstrated in mouse models, in which TXNIP deficient mice have increased levels of plasma lipids and triglycerides.31 Additionally, TXNIP inhibition is a potential target in the treatment of metabolic disorders,29 which might be interesting in light of epigenetic regulation of the TXNIP gene. For the TC, LDL-C, and HDL-C, we did not find epigenome-wide significant DMPs. However, we do believe that the top DMPs are potentially relevant associations, as they have previously been described in the regulation of lipids, weight, and glucose metabolism. For instance, for HDL-C, DNAm of cg00091964 annotated to the CTNNA2 gene has been reported to be associated with TC and LDL-C26 and genetic variation in the CTNNA2 gene has been associated with HDL-C,32 BMI33, 34, 35, 36 and coronary heart disease37 in multi-ethnic populations. In line with our findings DMP cg17058475 (CPT1A) and cg06500161 (ABCG1) have been linked to triglycerides and to lipid profile in general, BMI, and blood pressure.25,26,28 For LDL-C, the KLF7 gene has been linked to BMI,33,38,39 inflammation,40 and subcutaneous adipose tissue41 in European origin populations. Additionally, the pathway enrichment analysis showed that our top DMPs were involved in pathways of energy and lipid metabolism, transport and biosynthesis of lipids and cholesterol, and nitric oxide synthase signalling.42, 43, 44, 45, 46, 47, 48 Furthermore, the direction of effect and the strength of the associations were similar across all five geographical locations. This shows that despite different contextual factors, similar DMPs are at play in lipid metabolism. Moreover, lipids (independent variable) were not only associated with DNAm (dependent variable), but the methylation levels of the top DMPs (independent variable) were also significantly associated with lipid concentrations (dependent variable). Therefore, to confirm our findings, future research should aim for a larger sample size allowing more statistical power to detect epigenome-wide significant effects. We were able to replicate findings from Ghanaians in independent cohorts including South African Batswana, African American, and European descent populations, which supports that these DMPs (TXNIP, CPT1A, ABCG1, EPSTI1) are potentially relevant in the pathogenesis of dyslipidaemia and are universal across different ancestral groups. In contrast, the transferability of DMPs associated with lipid traits in South African Batswana, African American, and European origin populations to our Ghanaian study population was generally low, but especially limited for the findings in the South African populations. This implies the possible population specificity of these results, which are either based on genetic or environmental differences. Because of the large genetic diversity in SSA, it can be assumed that South African Batswana men are genetically different from Ghanaian population in genes regulating lipids or epigenetics,49 thereby making findings less generalisable between different ethnic populations in SSA. In contrast, admixed African Americans have up to $75\%$ shared ancestry with West Africans,50 and show a large percentage of European ancestry,51 thereby increasing the transferability of findings from African American to Ghanaians. Likewise, Ghanaian migrants residing in Europe share a more similar environment with African Americans and Europeans, whereas environmental factors differ between South African Batswana in South Africa, and Ghanaian migrants in Europe and non-migrants in Ghana, thereby affecting the transferability of findings between ethnicities. Previous findings from the RODAM study showed lower levels of HDL-C and higher levels of triglycerides in participants residing in rural Ghana, compared to those living in the other locations, and these differences were independent of common risk factors for dyslipidaemia.17 In this light, our finding of differences in methylation levels of CpGs between participants living in different geographical locations is interesting. Although we were not able to establish whether these differences in methylation levels are biologically relevant, it does highlight the importance of studying gene-environment interaction in different settings as DNAm is highly dynamic and potentially context-specific. Dyslipidaemia is strongly related to obesity and diabetes.52 This interrelatedness is also demonstrated by the observation that DMPs associated with triglycerides have previously been reported in EWAS on diabetes and obesity in the same Ghanaian study population.12,13 To rule out the potential interacting effects of obesity and diabetes, we adjusted our regression models for these factors. Additionally, in sensitivity analysis, we excluded participants with diabetes, which did not impact the effect size or direction of effect of the association. This indicates that the reported DMPs are potentially involved in lipid metabolism, independent of obesity and diabetes. The findings of this EWAS study on lipids in a West African population add to our knowledge of epigenetic associations with lipids in diverse populations. Highly standardised data collection across all five geographical locations allowed us to compare DNAm profiles in migrant and non-migrant Ghanaians, thereby assessing the impact of migration on DNAm. Additionally, we were able to perform the EWAS separately per geographical location before meta-analysing the findings, thereby minimising the confounding effect of unknown contextual factors on our results. Even though this study included the largest sample size of a West African population to date, our statistical power to detect epigenome-wide significant DMPs is assumed to be limited. Future studies should aim for a larger sample size, and more EWAS in different SSA populations can contribute to replication and pooling of the results. Because genotyping data were not available, we were not able to adjust our analysis for ancestry principal components. However, as $90\%$ of our study population was of a single ethnolinguistic group (Akan) who have been shown to be genetically homogenous,53 it is unlikely that our findings have been significantly affected by population stratification. We assessed DNAm extracted from whole blood samples. Even though lipids are a blood-based trait, methylation patterns can differ in target tissue where metabolism occurs, e.g. in adipocytes or hepatocytes, as our results from the EWAS Toolkit analyses showed. We conducted a cross–sectional association study, and conclusions related to the causal relation between DNAm and plasma lipid concentration should therefore be drawn with caution. For instance, Mendelian randomisation studies have shown lipid levels to be influenced by DNAm,54 but also that DNAm can influenced lipid levels.9 A longitudinal study design could help to establish temporality and direction of effect. In conclusion, we identified one epigenome-wide significant DMP associated with triglycerides (TXNIP) and several other lipid-associated DMPs (CPT1A, ABCG1) in this cohort of Ghanaians, loci which are also associated with lipids in populations of different ancestry. Several other identified CpGs are potentially relevant in lipid metabolism in Ghanaians but further work needs to be done to investigate their association with the observed favourable lipid profile of West African populations. Future studies including larger sample size, longitudinal study design, as well as translational studies - including different tissues and gene expression - can enlarge our pathophysiological understanding of dyslipidaemia among West African populations, thereby informing targeted strategies to curb the rising prevalence of dyslipidaemia in SSA populations. ## Contributors E.L.L., K.A.C.M., B.J.B., C.A., P.H. and A.A. conceived the study. C.A. and K.A.C.M. designed and carried out the recruitment and data collection. E.L.L., K.A.C.M., and A.A. were responsible for data analysis and interpretation. E.L.L., K.A.C.M. and A.V. verified the underlying data. E.L.L. wrote the article, supervised by K.A.C.M., P.H. and A.A., and in collaboration with F.C., C.H.B., S.B., K.K.G., A.V., B.J.B., and C.A. All authors read and approved the final version of the article. ## Data sharing statement Data are available upon reasonable request to the RODAM study coordinator (dr. Erik Beune, e.j.beune@amsterdamumc.nl). ## Declaration of interests E.L.L. is a voluntary member of the junior council of Amsterdam Public Health Research Institute, Global Health section. All other authors declared no conflicts. ## Supplementary data Supplementary Figs. S1–S4 and Tables S1–S5 ## References 1. Diseases G.B.D., 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) 396 1204-1222. PMID: 33069326 2. Bentley A.R., Rotimi C.N.. Interethnic differences in serum lipids and implications for cardiometabolic disease risk in african ancestry populations. Glob Heart (2017) 12 141-150. PMID: 28528248 3. Collaboration N.C.D.R.F.. Repositioning of the global epicentre of non-optimal cholesterol. Nature (2020) 582 73-77. PMID: 32494083 4. Jones A.C., Irvin M.R., Claas S.A., Arnett D.K.. Lipid phenotypes and DNA methylation: a Review of the literature. Curr Atheroscler Rep (2021) 23 71. PMID: 34468868 5. Coram M.A., Duan Q., Hoffmann T.J.. Genome-wide characterization of shared and distinct genetic components that influence blood lipid levels in ethnically diverse human populations. Am J Hum Genet (2013) 92 904-916. PMID: 23726366 6. Patil R.R.. Urbanization as a determinant of health: a socioepidemiological perspective. Soc Work Public Health (2014) 29 335-341. PMID: 24871771 7. Trerotola M., Relli V., Simeone P., Alberti S.. Epigenetic inheritance and the missing heritability. Hum Genom (2015) 9 17 8. Cavalli G., Heard E.. Advances in epigenetics link genetics to the environment and disease. Nature (2019) 571 489-499. PMID: 31341302 9. Jhun M.A., Mendelson M., Wilson R.. A multi-ethnic epigenome-wide association study of leukocyte DNA methylation and blood lipids. Nat Commun (2021) 12 3987. PMID: 34183656 10. Cronje H.T., Elliott H.R., Nienaber-Rousseau C., Pieters M.. Replication and expansion of epigenome-wide association literature in a black South African population. Clin Epigenet (2020) 12 6 11. Agyemang C., Beune E., Meeks K.. Rationale and cross-sectional study design of the Research on Obesity and type 2 Diabetes among African Migrants: the RODAM study. BMJ Open (2014) 4 12. Meeks K.A.C., Henneman P., Venema A.. An epigenome-wide association study in whole blood of measures of adiposity among Ghanaians: the RODAM study. Clin Epigenet (2017) 9 103 13. Meeks K.A.C., Henneman P., Venema A.. Epigenome-wide association study in whole blood on type 2 diabetes among sub-Saharan African individuals: findings from the RODAM study. Int J Epidemiol (2019) 48 58-70. PMID: 30107520 14. Chen Y.A., Lemire M., Choufani S.. Discovery of cross-reactive probes and polymorphic CpGs in the Illumina Infinium HumanMethylation450 microarray. Epigenetics (2013) 8 203-209. PMID: 23314698 15. Houseman E.A., Accomando W.P., Koestler D.C.. DNA methylation arrays as surrogate measures of cell mixture distribution. BMC Bioinformatics (2012) 13 86. PMID: 22568884 16. Du P., Zhang X., Huang C.C.. Comparison of Beta-value and M-value methods for quantifying methylation levels by microarray analysis. BMC Bioinformatics (2010) 11 587. PMID: 21118553 17. van der Linden E., Meeks K., Beune E.. Dyslipidaemia among Ghanaian migrants in three European countries and their compatriots in rural and urban Ghana: the RODAM study. Atherosclerosis (2019) 284 83-91. PMID: 30875497 18. 18Genetics of DNA Methylation ConsortiumGoDMC Database2022Available from:http://mqtldb.godmc.org.uk/index. (2022) 19. Safran M., Rosen N., Twik M., Abugessaisa I., Kasukawa T.. Practical Guide to Life Science Databases (2021) 22-56 20. Buniello A., MacArthur J.A.L., Cerezo M.. The NHGRI-EBI GWAS Catalog of published genome-wide association studies, targeted arrays and summary statistics 2019. Nucleic Acids Res (2019) 47 D1005-D1012. PMID: 30445434 21. Li M., Zou D., Li Z.. EWAS Atlas: a curated knowledgebase of epigenome-wide association studies. Nucleic Acids Res (2019) 47 D983-D988. PMID: 30364969 22. Komaki S., Shiwa Y., Furukawa R.. iMETHYL: an integrative database of human DNA methylation, gene expression, and genomic variation. Hum Genome Var (2018) 5 23. Xiong Z., Yang F., Li M.. EWAS Open Platform: integrated data, knowledge and toolkit for epigenome-wide association study. Nucleic Acids Res (2022) 50 D1004-D1009. PMID: 34718752 24. Kramer A., Green J., Pollard J., Tugendreich S.. Causal analysis approaches in ingenuity pathway analysis. Bioinformatics (2014) 30 523-530. PMID: 24336805 25. Sayols-Baixeras S., Subirana I., Lluis-Ganella C.. Identification and validation of seven new loci showing differential DNA methylation related to serum lipid profile: an epigenome-wide approach. The REGICOR study. Hum Mol Genet (2016) 25 4556-4565. PMID: 28173150 26. Hedman A.K., Mendelson M.M., Marioni R.E.. Epigenetic patterns in blood associated with lipid traits predict incident coronary heart disease events and are enriched for results from genome-wide association studies. Circ Cardiovasc Genet (2017) 10 27. Justice A.E., Chittoor G., Gondalia R.. Methylome-wide association study of central adiposity implicates genes involved in immune and endocrine systems. Epigenomics (2020) 12 1483-1499. PMID: 32901515 28. Richard M.A., Huan T., Ligthart S.. DNA methylation analysis identifies loci for blood pressure regulation. Am J Hum Genet (2017) 101 888-902. PMID: 29198723 29. Qayyum N., Haseeb M., Kim M.S., Choi S.. Role of thioredoxin-interacting protein in diseases and its therapeutic outlook. Int J Mol Sci (2021) 22 30. Pan M., Zhang F., Qu K., Liu C., Zhang J.. TXNIP: a double-edged sword in disease and therapeutic outlook. Oxid Med Cell Longev (2022) 2022 31. Sheth S.S., Castellani L.W., Chari S.. Thioredoxin-interacting protein deficiency disrupts the fasting-feeding metabolic transition. J Lipid Res (2005) 46 123-134. PMID: 15520447 32. Cheng H., Sewda A., Marquez-Luna C.. Genetic architecture of cardiometabolic risks in people living with HIV. BMC Med (2020) 18 288. PMID: 33109212 33. Pulit S.L., Stoneman C., Morris A.P.. Meta-analysis of genome-wide association studies for body fat distribution in 694 649 individuals of European ancestry. Hum Mol Genet (2019) 28 166-174. PMID: 30239722 34. Winkler T.W., Justice A.E., Graff M.. The influence of age and sex on genetic associations with adult body size and shape: a large-scale genome-wide interaction study. PLoS Genet (2015) 11 35. Hoffmann T.J., Choquet H., Yin J.. A large multiethnic genome-wide association study of adult body mass index identifies novel loci. Genetics (2018) 210 499-515. PMID: 30108127 36. Locke A.E., Kahali B., Berndt S.I.. Genetic studies of body mass index yield new insights for obesity biology. Nature (2015) 518 197-206. PMID: 25673413 37. Koyama S., Ito K., Terao C.. Population-specific and trans-ancestry genome-wide analyses identify distinct and shared genetic risk loci for coronary artery disease. Nat Genet (2020) 52 1169-1177. PMID: 33020668 38. Kichaev G., Bhatia G., Loh P.R.. Leveraging polygenic functional enrichment to improve GWAS power. Am J Hum Genet (2019) 104 65-75. PMID: 30595370 39. Zhu Z., Guo Y., Shi H.. Shared genetic and experimental links between obesity-related traits and asthma subtypes in UK Biobank. J Allergy Clin Immunol (2020) 145 537-549. PMID: 31669095 40. Ahola-Olli A.V., Wurtz P., Havulinna A.S.. Genome-wide association study identifies 27 loci influencing concentrations of circulating cytokines and growth factors. Am J Hum Genet (2017) 100 40-50. PMID: 27989323 41. Fox C.S., Liu Y., White C.C.. Genome-wide association for abdominal subcutaneous and visceral adipose reveals a novel locus for visceral fat in women. PLoS Genet (2012) 8 42. Yaku K., Okabe K., Nakagawa T.. NAD metabolism: implications in aging and longevity. Ageing Res Rev (2018) 47 1-17. PMID: 29883761 43. Zhou J., Lam B., Neogi S.G., Yeo G.S., Azizan E.A., Brown M.J.. Transcriptome pathway analysis of pathological and physiological aldosterone-producing human tissues. Hypertension (2016) 68 1424-1431. PMID: 27777363 44. Longo N., Frigeni M., Pasquali M.. Carnitine transport and fatty acid oxidation. Biochim Biophys Acta (2016) 1863 2422-2435. PMID: 26828774 45. Pavek P.. Pregnane X receptor (PXR)-Mediated gene repression and cross-talk of PXR with other nuclear receptors via coactivator interactions. Front Pharmacol (2016) 7 456. PMID: 27932985 46. Martinez-Sena T., Soluyanova P., Guzman C., Valdivielso J.M., Castell J.V., Jover R.. The vitamin D receptor regulates glycerolipid and phospholipid metabolism in human hepatocytes. Biomolecules (2020) 10 47. Ji N., Qi Z., Wang Y.. Pyroptosis: a new regulating mechanism in cardiovascular disease. J Inflamm Res (2021) 14 2647-2666. PMID: 34188515 48. stearoyl-CoA desaturase: GeneCards gene database. (2022) 49. Bentley A.R., Chen G., Doumatey A.P.. GWAS in Africans identifies novel lipids loci and demonstrates heterogenous association within Africa. Hum Mol Genet (2021) 30 2205-2214. PMID: 34196372 50. Auton A., Brooks L.D.. A global reference for human genetic variationA global reference for human genetic variation. Nature (2015) 526 68-74. PMID: 26432245 51. Bryc K., Durand E.Y., Macpherson J.M., Reich D., Mountain J.L.. The genetic ancestry of African Americans, Latinos, and European Americans across the United States. Am J Hum Genet (2015) 96 37-53. PMID: 25529636 52. Wu L., Parhofer K.G.. Diabetic dyslipidemia. Metabolism (2014) 63 1469-1479. PMID: 25242435 53. Adeyemo A.A., Chen G., Chen Y., Rotimi C.. Genetic structure in four West African population groups. BMC Genet (2005) 6 38. PMID: 15978124 54. Sayols-Baixeras S., Tiwari H.K., Aslibekyan S.W.. Disentangling associations between DNA methylation and blood lipids: a Mendelian randomization approach. BMC Proc (2018) 12 23. PMID: 30275879
--- title: Assessment and Optimization of Explainable Machine Learning Models Applied to Transcriptomic Data authors: - Yongbing Zhao - Jinfeng Shao - Yan W. Asmann journal: Genomics, Proteomics & Bioinformatics year: 2022 pmcid: PMC10025763 doi: 10.1016/j.gpb.2022.07.003 license: CC BY 4.0 --- # Assessment and Optimization of Explainable Machine Learning Models Applied to Transcriptomic Data ## Abstract Explainable artificial intelligence aims to interpret how machine learning models make decisions, and many model explainers have been developed in the computer vision field. However, understanding of the applicability of these model explainers to biological data is still lacking. In this study, we comprehensively evaluated multiple explainers by interpreting pre-trained models for predicting tissue types from transcriptomic data and by identifying the top contributing genes from each sample with the greatest impacts on model prediction. To improve the reproducibility and interpretability of results generated by model explainers, we proposed a series of optimization strategies for each explainer on two different model architectures of multilayer perceptron (MLP) and convolutional neural network (CNN). We observed three groups of explainer and model architecture combinations with high reproducibility. Group II, which contains three model explainers on aggregated MLP models, identified top contributing genes in different tissues that exhibited tissue-specific manifestation and were potential cancer biomarkers. In summary, our work provides novel insights and guidance for exploring biological mechanisms using explainable machine learning models. ## Introduction In recent years, many tools based on machine learning models have been developed and applied to biological studies, most of which are developed for predictions. For example, AlphaFold was developed to predict protein 3D structure from amino acid sequences [1], P-NET was used to predict cancer treatment-resistance state from molecular data [2], and CEFCIG can predict cell identity regulators from histone markers [3]. Additionally, machine learning models can predict different biological features from one single data type, depending on which feature is paired with the input data when training the model. For instance, a variety of models have been developed to predict ncRNA [4], nucleosome [5], and chromatin accessibility/activity/state [6], [7], [8], [9] from genome sequences. Although these tools have been greatly successful in various biological topics, biologists are still curious about how a machine learning model makes a decision and which features of the input data play important roles in the model output. To answer these questions, explainable artificial intelligence (XAI) programs have recently emerged to enable the development of models that can be understood by humans [10], [11]. These XAI methods can also be applied to interpret machine learning models obtained from biological data by quantifying feature contributions to model prediction [12], [13]. The two most popular approaches to estimate the contribution of each input feature to the model output are: 1) perturbing the input data and comparing outputs between the original and perturbated inputs; and 2) using backpropagation to measure the importance of each feature in the input data [14], [15], [16]. The former is intuitive but computationally expensive, especially when exhaustively estimating all input features, and there is also the risk of underestimating feature contribution [17]. By contrast, the latter can measure the contribution of all input features in “one shot”. Consequently, many model explainers based on backpropagation were proposed and developed in the field of computer science and computer vision [18]. Benefiting from these model explainers, computational biologists discovered the syntax of transcription factor (TF) binding motifs by interpreting models trained to predict chromatin accessibility [19], [20]; and screened for cancer marker genes from models of cancer type classification [21], [22], [23]. There is no doubt that these explorations have showcased the potential of interpretable models in discovering meaningful biological mechanisms. However, the remaining problem is that results from different model explainers are highly variable [21]. Since these model explainers were not specifically designed for biological data, it is critical to evaluate their applicability in biology. Currently, there is still a lack of comprehensive understanding of these explainers in biological studies. To fill this gap, we optimized and assessed the performance of different model explainers and analyzed their biological relevance. To minimize the impact of model performance on the assessment of explainers, we tested explainers on well-trained models for predicting tissue types and cancer/normal statuses from gene expression data. In summary, this study provides comprehensive guide for applying interpretable machine learning to biological studies. ## Overview of model interpretability In this study, we formulated a specific question to instantiate the application of interpretable models to biological data. Can we quantify the contribution of individual genes to tissue type and disease status? Two steps were implemented. First, we built neural network models and trained the models with transcriptomes as input and tissue type and disease status of the transcriptome sample as the prediction output. Models were built based on two types of neural networks, convolutional neural network (CNN) and multilayer perceptron (MLP), and model architectures are detailed in the methods section. In general, CNN is more complex than MLP. Next, we applied model explainers to compute a quantitative score of each gene’s contribution to the model’s prediction, which are named gene contribution scores. We tested eight popular model explainers and their variations commonly used in computer vision and assessed and compared their applicability and performances on each pre-trained model. These explainers are: gradients (Saliency), Input x Gradient (InputXGradient), guided backpropagation (GuidedBackprop), Integrated Gradients (IntegratedGradients), DeepLIFT (DeepLift), approximating SHAP values using DeepLIFT (DeepLiftShap), Guided Grad-CAM (GuidedGradCam), and Guided Grad-CAM++ (GuidedGradCam++) (Table S1) [17], [18], [24], [25], [26], [27], [28], [29]. Since GuidedGradCam and GuidedGradCam++ were developed for CNN specifically, only the first six explainers were tested on MLP. We used 27,417 RNA-seq samples from The Genotype-Tissue Expression (GTEx) and The Cancer Genome Atlas (TCGA) projects to train CNN or MLP-based models. These samples were from 82 distinct normal and cancer tissues and cell types (Table S2). After training, the prediction accuracy of all models was comparable, with a median value of $97.2\%$ for CNN and $97.8\%$ for MLP. The convolutional layers of the CNN models require that, as input data, gene expression values should be organized with a fixed gene order in a 2D matrix. Therefore, we tested various gene orders for the CNN-based models, e.g., sorting genes according to their genomic coordinates [22]. Results indicated that gene order did not affect model performance in terms of prediction accuracy. Although models were trained with all 82 different normal and cancer tissues, results from four normal tissues (liver, lung, ovary, and pancreas) are reported here to illustrate the applicability and performance of the eight explainers. ## Direct use of explainers from computer vision resulted in poor reproducibility Randomness is often challenging in machine learning, which is present in both model training and model interpretation [30], [31]. For this reason, we measured both intra-model and inter-model reproducibility of each explainer. During the testing, each explainer was applied to pre-trained models based on CNN or MLP, respectively. First, we tested intra-model reproducibility by applying an explainer to the same pre-trained model 5 times (5 replicates per model per explainer), and for each explainer, we checked correlations of gene contribution scores as well as the pair-wise overlap of the top 100 contributing genes between replicates. We found that intra-model reproducibility, in terms of Spearman’s correlation of gene contribution scores and overlap of the gene IDs among the top 100 contributing genes, is low for most explainers on both CNN- and MLP-based models (results from normal liver samples are shown in Figure 1; results from normal lung, ovary, and pancreas are shown Figure S1). The exception was GuidedGradCam++, which was previously used to identify cancer marker genes [21].Figure 1Performance of different model explainers without optimizationA. Spearman’s correlation of gene contribution scores (upper panel) and overlap in the top 100 contributing genes (lower panel) in liver among replicates from the same pre-trained model, different pre-trained models with the same gene order, and different pre-trained models with different gene orders on CNN-based models. B. Spearman’s correlation on gene contribution scores (upper panel) and overlap in the top 100 contributing genes (lower panel) in liver among replicates from the same pre-trained model and different pre-trained models on MLP-based models. CNN, convolutional neural network; MLP, multilayer perceptron; Saliency, gradients; InputXGradient, Input x Gradient; GuidedBackprop, guided backpropagation; IntegratedGradients, Integrated Gradients; DeepLift, DeepLIFT; DeepLiftShap, approximating SHAP values using DeepLIFT; GuidedGradCam, Guided Grad-CAM; GuidedGradCam++, Guided Grad-CAM++. As shown in Figure 1, we also tested inter-model reproducibility by applying each explainer to five different models with comparable prediction accuracies (5 models per explainer). These five models were trained using the same model architecture and training data set but with slightly different hyperparameters. For CNN models, we also tested the impact of different input gene orders since CNN requires organizing input genes in a 2D matrix. The inter-model reproducibility of all explainers, including GuidedGradCam++, was significantly lower than those of the intra-models. For CNN-based models, even though the gene order had little impact on prediction accuracy, it had a significant impact on the model reproducibility, especially regarding the overlap of top contributing genes. For MLP-based models, the reproducibility of intra-model and inter-model tests were similar; however, both Spearman’s correlation and overlap of the top 100 contributing genes were very low. In summary, gene contribution scores vary greatly intra- and inter-models for both CNN and MLP. These tests were performed per explainer. We expect that reproducibility across different explainers would be much worse. Therefore, model explainers developed for computer vision may not be directly applied to answering biological questions. Model interpretability in computer vision aims to identify visual features consisting of multiple similar pixels in an area, and variations within the area have a limited impact on the outcome. However, interpreting biological data such as the transcriptomes requires single-gene resolution since genes within an area were arbitrarily placed together, therefore, the results are very sensitive to random noise. ## Optimization of model interpretability Since it is not feasible to directly transfer model explainers from computer vision to biology, we tested whether these explainers can be optimized and adjusted for biological data. First, we borrowed a de-noising strategy widely used in computer vision, “SmoothGrad: removing noise by adding noise” [32]. Instead of estimating gene contribution scores in a sample in “one pass”, SmoothGrad calculates the contribution of a gene by averaging contribution scores from multiple explanation estimates per sample by adding random noise into the expression data each pass. Unfortunately, the strategy of SmoothGrad did not improve inter-model reproducibility. On the contrary, it lowered the performance of all explainers on both CNN and MLP except for Saliency on MLP and GuidedBackprop on CNN (Figure 2A, Figure S2). For Saliency on MLP, the improvement saturates when the number of repeat estimates reaches 50, while the performance of GuidedBackprop plateaued after 30 repeats. Next, we tested whether repeating the explanation without adding random noise into the expression data, which we defined as a simple repeat, would be beneficial. Results of the simulation indicated that repeats without adding noise significantly improved the performance of all explainers on both CNN and MLP, except for Saliency and GuidedGradCam++ on CNN (Figure 2B, Figure S3). For most explainers, improvement saturated after 20 repeats for CNN and after 40 repeats for MLP.Figure 2Optimization of different model explainersSpearman’s correlation of gene contribution scores in liver from CNN models (upper panels) and MLP (lower panels). A. Performance by averaging contribution scores from multiple estimates per sample by adding random noise into the expression data each pass. B. Performance from running the same model multiple times and averaging contribution scores without adding random noise (simple repeats). C. Performance of repeated reference zero for CNN, and a number of references randomly selected from 2000 simulated reference universal for MLP models, for three explainers that require reference samples. D. Performance of model aggregations. E. Spearman’s correlation on gene contribution scores (upper panel) and overlap in the top 100 contributing genes (lower panel) among replicates from different pre-trained models with different gene orders based on CNN-based models. The analyses were carried out with three different optimization strategies: without optimization, with optimized conditions for each explainer but without model aggregation, and with optimized conditions for each explainer and with model aggregation. F. Same as (E) but based on MLP-based models. w/o, without; w/, with. DeepLift, DeepLiftShap, and IntegratedGradients require a reference baseline when estimating gene contribution scores, where a reference is a synthetic, randomly generated transcriptome. In computer vision, a black image (all-zero input) or random pixel values are often used as references, and for motif identification of regulatory elements, the scrambled genomic sequence was demonstrated as good reference [17]. In this study, we compared four types of references named reference zero, normal, universal, and specific. Reference zero and normal are equivalent to black image and random pixel values, respectively. For reference universal and specific, we estimated the mean (µ) and standard deviation (σ) of each gene’s expression levels across samples and then randomly generated a value based on a truncated normal distribution N (µ, σ). Reference universal uses samples from all 82 tissue types, while reference specific uses samples from a specific tissue type. We tested the performance of these four kinds of references individually as well as in combination, to evaluate whether multiple references would improve reproducibility. First, for individual references, simulation results showed that reference zero is the best for CNN-based models, while universal is preferred for MLP-based models (Figure S4). The result is consistent with all the three explainers that required references. Next, we tested the impact of multiple references on reproducibility. Since the effect of using multiple references with zero is equivalent to that of simple repeat with single reference zero, these two kinds of optimization, using multiple references zero and simple repeat, cannot contribute to reproducibility additively. Therefore, we compared reproducibility by combing simple repeat with multiple references as reference normal, universal, or specific with the reproducibility by combing simple repeat with single reference zero (which is equivalent to multiple references zero without simple repeat). Interestingly, we found that reference zero still outperformed the other three kinds of references on CNN-based models (Figure S5). Similar as simple repeat, improvement saturates when the number of references zero reaches 20 on CNN-based models, while the number of references universal goes to 60 on MLP-based models (Figure 2C). Since the intra-model reproducibility was significantly improved by repeating the explanation process multiple times and averaging contribution scores from different estimates (Figure 2B), we next tested the benefits of inter-model aggregation. We applied optimized parameters of each explainer on CNN or MLP-based models (Table S1), estimated gene contribution scores on each pre-trained model individually, and then averaged inter-model results. Indeed, aggregating models significantly increased reproducibility (Figure 2D, Figure S6). Spearman’s correlations for DeepLift, DeepLiftShap, IntegratedGradients, and Saliency reached nearly 1.0 on MLP. In general, the reproducibility of all explainers was significantly improved on both CNN and MLP-based models after aggregating models (Figure 2D–F, Figure S7). Of note, the model aggregation had the most significant impact and improvement on the reproducibility of all explainers on CNN-based models in terms of overlaps between the top 100 contributing genes (Figure 2E, lower panel). For most explainers on MLP-based models, Spearman’s correlations on gene contribution scores from model aggregation were higher than 0.9, and over $90\%$ of the top 100 contributing genes overlapped between replicates on the same explainer (Figure 2F). To summarize, gene contribution scores were highly reproducible from the same explainer with optimized parameters. Reproducibility of the top 100 contributing genes was better on MLP-based models than those on CNN-based models. One possible reason is that CNN-based models are much more complex than MLP-based models and can be hard to be interpreted. ## Consistency across model explainers So far, we’ve tested the performance within each explainer. To test the consistency of gene contribution scores across different explainers, we checked the overlap of the top 100 contributing genes identified by different explainers with and without model aggregation (Table S3). Within CNN or MLP models, model aggregation did not only improve reproducibility within the same explainer, but also across explainers. However, the top 100 contributing genes from CNN-based models with model aggregation did not overlap with those from MLP-based models with or without model aggregation. Moreover, within CNN-based models, the top 100 contributing genes with model aggregation did not overlap with those without model aggregation, which suggests that model aggregation resulted in explainers identifying a completely different set of genes. By contrast, top contributing genes from MLP-based models were highly consistent with and without model aggregations. We further explored why model aggregation had different impacts on MLP and CNN. We first defined the top 100 contributing genes from optimized parameters with model aggregations as the baseline of comparison for each explainer. Next, we compared the top 100 contributing genes from each explainer without optimization to the baseline. It was found that, without optimization on MLP-based models, top contributing genes shared by two or more replicates had a higher overlap with the baseline than that between top contributing genes from individual replicates and the baseline (Figure S8). However, this conclusion was not seen on CNN-based models. This result suggests that the top contributing genes from different MLP models are convergent, while those from different CNN models are very divergent. Intriguingly, the measurement of reproducibility highlighted three representative groups, each with different explainers, model types (CNN or MLP), and aggregation status (Figure 3). The three groups are group I: DeepLift, DeepLiftShap, GuidedBackprop, InputXGradient, and IntegratedGradients on CNN-based models with model aggregation; group II: DeepLift, DeepLiftShap, InputXGradient, and IntegratedGradients on MLP-based models with model aggregation; and group III: GuidedBackprop and Saliency on MLP-based models with model aggregation. Next, we delved into the top contributing genes identified by these three groups. Figure 3Overlap of the top 100 contributing genes across explainers with and without model aggregationThree representative groups (I, II, and III) are marked by black bars. Aggregation of CNN models or MLP models is shown as CNN-Agg or MLP-Agg. ## Expression status of top contributing genes It is important to understand the biological relevance of the top contributing genes by different model architectures (MLP vs. CNN) and different explainers. Since contribution scores were derived from the input gene expression values, we first calculated Spearman’s correlation between gene contribution scores and expression levels (Figure S9). We expected a high correlation for genes identified by InputXGradient because gene expression level is a cofactor in computing gene contribution scores by InputXGradient. Indeed, we found weak correlations of all explainers in both groups II and III except for InputXGradient. Conversely, strong correlations were observed from 4 explainers in group I: InputXGradient, IntegartedGradients, DeepLift, and DeepLiftShap. However, it is puzzling that GuidedBackprop from group I showed negative correlations for unknown reasons. Additionally, we checked overlaps between the top 100 contributing genes and the top 100 expressed genes on both CNN- and MLP-based models (Figure S10). In liver, nearly $50\%$ of the top contributing genes from group II overlapped with top expressed genes, while the overlaps were less than $10\%$ in groups I and III (Figure 4A). Group II, as described above, are DeepLift, DeepLiftShap, InputXGradient, and IntegratedGradients on MLP-based models with model aggregation. Strikingly, model aggregation eliminated the already moderate overlaps in CNN-based models (group I). Another noticeable finding for group I is that though Spearman’s correlation between gene contribution scores and expression level was very high, the majority of the top contributing genes were not highly expressed. Figure 4Biological relevance of the top 100 contributing genesA. Overlaps between the top 100 contributing genes and the top 100 highest expressed genes in liver samples. Explainers that belong to groups I, II, and III are marked by horizontal black bars. B. Expression profiles of the top 100 contributing genes in liver, lung, ovary, and pancreas identified by DeepLift on CNN-based models with model aggregation (representative of group I), DeepLift on MLP-based models with model aggregation (group II), and Saliency on MLP-based models with model aggregation (group III). C. Overlaps in the top 100 contributing genes among liver, lung, ovary, and pancreas identified by DeepLift on CNN-based models with model aggregation (group I), DeepLift on MLP-based models with model aggregation (group II), and Saliency on MLP-based models with model aggregation (group III). D. Overlaps between the top 100 contributing genes and TS genes in liver samples. E. Overlaps between the top 100 contributing genes and HK genes in liver samples. F. Percentages of TF-coding genes among the top 100 contributing genes in liver samples (left panel) and ovary samples (right panel). Dashed lines indicate the overlap percentage by random chance. TS, tissue-specific; HK, housekeeping; TF, transcription factor. Since different tissues have distinct phenotypes, we wondered whether the top contributing genes of different tissue types exhibit distinct expression profiles. Heatmap of the top contributing genes clearly demonstrated tissue-specific (TS) manifestations for group II (represented by DeepLift on MLP with model aggregation), and the patterns were much weaker in both group I (represented by DeepLift on CNN with model aggregation) and group III (represented by Saliency on MLP with model aggregation) (Figure 4B). In addition, the total number of genes from group III is much lower than that of both groups I and II after removing redundant genes from the top 100 contributing genes across tissues. This suggests that the top 100 contributing genes were largely shared across tissues in group III, which was validated by comparison across tissue types in all explainers (Table S4). Among the three groups, the top contributing genes in both groups I and II are TS, while the top contributing genes in group III are highly shared across tissues (Figure 4C). Considering that the top contributing genes in groups I and II were mostly TS, we are curious how the top contributing genes are related to tissue specifically expressed genes (TS genes). For this purpose, we identified TS genes across all 82 tissues and cell types, which were used in model training. It was found that about $70\%$ of the top contributing genes overlap with TS genes in group II in liver (Figure 4D). The fractions vary across tissues (Figure S11A), since there are different numbers of TS genes in each tissue type (Figure S11B). The percentages drop to less than $10\%$ in both groups I and III. Interestingly, model aggregation also significantly reduced overlaps with TS genes in most explainers on CNN-based models. In addition, we observed that many of the top contributing genes (in group I particularly) are expressed at comparable levels across tissues. We investigated the relationships between the top contributing genes and housekeeping (HK) genes. Results showed that about $10\%$−$20\%$ of top contributing genes overlap with HK genes in group I, and the overlap was also further reduced by model aggregation (Figure 4E, Figure S12). Conversely, no overlap was found in both groups II and III, except for the explainer InputXGradient. ## Enrichment of top contributing genes in biological functions To understand the biological functions of the top contributing genes, we performed gene ontology (GO) enrichment analysis. No enrichment was found on genes identified by all explainers in group I. The enriched GO terms by genes from group II were mostly unique for each tissue type and TS functions (Table S5). For example, enriched GO terms in liver are molecular functions related to lipoprotein and lipoprotein lipase activities, while GO terms enriched in pancreas are associated with the binding of oligosaccharides, peptidoglycan and so on. Additionally, in group II, results among DeepLift, DeepLiftShap and IntegratedGradients are slightly more agreeable compared to that from InputXGradient. For group III, we expected similar GO terms enriched across tissue types since top contributing genes from different tissues highly overlapped. This turned out to be the case. GO enrichment analysis showed that the top contributing genes in group III are enriched in CCR7 chemokine receptor binding, neuropeptide hormone activity, neuropeptide receptor binding, and DNA-binding transcription activator activity across tissues. Next, we checked how top contributing genes are related to TFs and TF cofactors. We found about 20 genes overlapping with TFs in group III, which is more than 2-fold enrichment than by random chance (Figure 4F). By contrast, genes in group II showed depletion of TFs in liver, but 1.5-fold enrichment in ovary (Figure 4F, Figure S13). No enrichment or depletion was found in group I, except for genes identified by GuidedGradCam++. As for TF cofactors, there is low overlap in all three groups (Figure S14). ## Top contributing genes in cancers Group II’s top contributing genes are TS with TS manifestations of expression values. Therefore, we asked how the expression pattern of the top contributing genes changed from normal to cancer tissues. We compared normal and cancer samples of liver, lung, ovary, and pancreas from GTEx and TCGA, and studied the expression differences of top contributing genes. About $40\%$−$80\%$ of the top contributing genes were differentially expressed genes between normal and cancer tissues identified by DeepLift on MLP (group II), which is about twice more than the random chance (Figure 5A). The percentages ranged from $30\%$ to $60\%$ by Saliency on MLP (group III), which is about 1.5-fold over random chance. Group I was not included in this analysis since model aggregation eliminated many features in common from different explainers on CNN-based models, and no biological enrichments were found in the top contributing genes. Figure 5Top contributing genes in cancersA. Percentages of the top 100 contributing genes which were differentially expressed between normal and cancer tissues. Dashed lines indicate the percentage of differentially expressed genes by random chance. B. Expression levels of the top 100 contributing genes which were observed only in cancer samples, only in normal samples, and in both normal and cancer samples, of genes identified by DeepLift on aggregated MLP models (shown as MLP-Agg, representative of group II, left panel) and by Saliency on aggregated MLP models (representative of group III, right panel). C. Heatmap of the 1179 shared top contributing genes from 33 TCGA cancer types demonstrated tissue specificity. Interestingly, differentially expressed top contributing genes are segregated into two distinct populations in group II (Figure 5B, Figure S15). Specifically, the top contributing genes specific to normal tissues are downregulated in cancer, while those specific to cancer are upregulated in cancer. For example, Glypican-3 (GPC3), a member of the heparan sulfate proteoglycans family, is one of the top contributing genes in liver cancer but not in normal liver. GPC3 is often observed to be highly elevated in hepatocellular carcinoma and is a target for diagnosis and treatment of hepatocellular carcinoma [33]. However, similar segregation was not found in group III (Saliency on MLP-based models) because top contributing genes from group III were mostly shared between normal and cancer tissues. Together, the expression profiles suggest that the top contributing genes in group II might be potential cancer biomarkers. We identified the top 100 contributing genes in individual samples of all 33 different cancer types and named those shared by two or more samples of the same cancer type as shared top contributing genes. In total, 1179 genes were identified as shared top contributing genes. As expected, these shared top contributing genes are mostly TS (Figure 5C). Among these shared top contributing genes, we further studied the known oncogenes and tumor suppressor genes in OncoKB [34]. Heatmap analysis showed that some oncogenes and tumor suppressor genes are shared by multiple cancer types, such as SFRP2, while the others are specific to one or very few cancers (Figure S16). ## Discussion The beauty of interpreting machine learning models is that it converts the complex mathematical rules learned by neural networks into biological rules and provides new insights into biology. To facilitate the application of an interpretable machine learning model, we established a series of optimization steps and compared the biological relevance of different model explainers. Since the tests in this study were based on models that predict tissue types from transcriptomes, applications using other types of biological data or different model architectures may require further investigations. In addition, even though the current optimizations demonstrated good performance on MLP-based models for a subset of explainers, some important genes may still be missing from the top contributing genes. For example, a machine learning model might choose only one of two highly correlated genes to use for prediction. Alternatively, the contribution of two highly correlated genes might be diluted if the model chooses to use both genes, and thus, the contribution here might not reflect biological importance. These factors might partly explain the low reproducibility of individual single models and why improvement could be made by aggregation of models. Overall, we believe this study will provide novel insights to optimize interpretable machine learning in biological studies. A recent paper pointed out five potential pitfalls of applying machine learning in genomics: 1) distributional differences; 2) dependent examples; 3) confounding; 4) leaky pre-processing; and 5) unbalanced class [35]. These technical challenges of applying machine learning models to genomics data are nontrivial and should be paid close attention to in addition to the optimization strategies we laid out in this study. Typically, complicated models are not easily interpretable [36], which is also confirmed by the poor performance when interpreting CNN-based models. In this study, the optimized strategy significantly increased the interpretability on MLP-based models for a subset of explainers, but not on CNN-based models for any explainers. The aggregated CNN model approach should perhaps be categorized into a new modeling strategy, which is similar to the “averaging” of models. The “averaging” strategy indeed mitigated randomness to some extent but didn’t show biological relevance. Therefore, even if models of different architectures had comparable prediction performances, it’s probably preferable to use models with relatively simpler architectures for model interpretation. The top contributing genes detected by explainers in group II (DeepLift, DeepLiftShap, InputXGradient, and IntegratedGradients on MLP-based models with model aggregation) exhibited TS manifestation in both gene ontology and expression profile, which is expected based on prior knowledge about tissue specificity and cell identity [37], [38], [39], [40]. Therefore, explainers in group II are more suitable for biological study, especially when exploring biological questions based on transcriptomic data. In recent years, single-cell RNA-seq technology has been widely applied to different tissue and diseases, leading to the discovery of many well-defined sub-cell populations [41], [42]. Although this study assessed model interpretability on bulk RNA-seq transcriptomes, the optimization strategies proposed here can also be applied to single-cell transcriptomes to quantify individual gene contribution and identify important genes in each sub-population. It is expected that interpretable machine learning models will also benefit understanding of tissue heterogeneity, disease mechanisms, and cellular engineering at single-cell resolution. ## Human transcriptome collection and processing A total of 27,417 RNA-seq samples were used in our study, among which 17,329 and 10,088 samples were collected from GTEx and TCGA, respectively [43]. These samples are from 47 distinct primary normal tissues and 2 cell lines (with prefix GTEx_ in the tissue code) and 33 different primary tumors (with prefix TCGA_ in the tissue code). Pre-processed TCGA and GTEx RNA-seq gene expression level data were downloaded from GTEx Portal (phs000424.v8.p2) and Recount2 database [44], respectively. For TCGA data, only primary tumor samples were included. Names of tissue types (normal or cancer) remained the same as defined by TCGA and GTEx projects. For each sample, the expression levels of 19,241 protein-coding genes were normalized to log2 (TPM + 1), where TPM denotes Transcript Per Million, and then used for analyses. ## The architecture of CNN models We used a five-layer CNN to build the CNN models, which included three convolutional layers, one global average pooling layer, and one fully connected layer sequentially. Each layer included 64, 128, 256, 256, and 82 channels, respectively. The kernel sizes for the three convolutional layers were 5, 5, and 3, respectively, and each convolutional layer was followed by max-pooling with a kernel size of 2. Batch normalization and rectified linear unit activation function [ReLU, which can be presented as f(x) = max (0, x)] were applied immediately after max-pooling of each convolutional layer and global average pooling layer. As the input of the CNN model, normalized expression values of 19,241 protein-coding genes from a sample were transformed into a 144 × 144 matrix, and zero padding was used at the bottom of the matrix. The final fully connected layer produced a vector of 82-probability-like scores, each corresponding to one of the 82 tissue types (normal or cancer). ## The architecture of MLP models There was only one hidden layer in the MLP models with 128 units. Batch normalization and ReLU were applied immediately after the hidden layer. There were 19,241 variables in the input layer, each corresponding to one of the 19,241 genes. The output layer assigns a probability-like score for each of the 82 tissue types. ## Model training All samples in a tissue type were randomly partitioned at a 9:1 ratio, with $90\%$ of samples used as training data and the remaining $10\%$ as testing data. In each epoch, up-sampling was employed to avoid imbalance caused by different sample sizes between tissue types. Adam optimizer on cross-entropy loss was utilized to update the weights of the neural network. After hyperparameter optimization, an initial learning rate of 0.0006 was used for CNN models, and 0.001 was used for MLP models. A batch size of 256 was used for both CNN and MLP. If there was no improvement for 5 sequential epochs, the learning rate was reduced by 0.25. L2 regularization was applied with a lambda score of 0.001. A fixed dropout of 0.25 was applied before the output layer in the MLP models, while a dropout of 0.25 was applied before the global average pooling layer in the CNN models. To optimize reproducibility in model explanation, we selected 60 well-trained models with slightly different parameters but of similar performances. In the CNN model, genes were organized into 2D matrix with fixed orders as input. In this study, gene orders in the CNN model were also experimented. For testing of the same gene order, we selected 5 well well-trained models with slightly different parameters but of similar performances. To study different gene orders, we selected 60 well-trained models with slightly different parameters but of similar performances. ## Model performance estimation Five-fold cross-validation was used to estimate the model performance for both MLP and CNN. Five groups of datasets were prepared, and each included a training dataset and a test dataset. Dataset preparation for each group was as follows. First, we randomly split all samples in a tissue type into 5 parts. Each group used one of the 5 parts as a test dataset, and the remaining four parts were combined into the training dataset. The same hyperparameters were used to train models based on the training dataset of each group separately. The trained models were then used to estimate the test dataset of the same group. Estimated results from five groups were combined, and all metrics about performance were calculated based on the combined results. ## Model explanation To estimate how much each gene contributes to the model prediction, we used eight different model explainers and variations, which are DeepLift, DeepLiftShap, GuidedBackprop, GuidedGradCam, GuidedGradCam++, InputXGradient, IntegratedGradients, and Saliency. All these explainers were implemented based on the Captum package (https://github.com/pytorch/captum). All explanations were based on well-trained CNN and MLP models. In addition, dropout was enabled to increase the diversity of model architecture, which helps measure the impact of model variations and uncertainties during the model explanation. As output, each explainer estimated contribution scores for each of the 19,241 genes. ## Reference preparation In this study, we tested four kinds of references which are named as zero, normal, universal, and specific. 1) For reference zero, we assigned the expression level of each gene to 0. 2) For reference normal, the expression level of each gene was randomly generated from a truncated normal distribution N [0, 1], and all values were restricted between 0 and 1. 3) For reference universal, the expression level of each gene was randomly generated from a truncated normal distribution N (μ, σ), and all values are restricted between 0 and σ. Here, μ and σ were calculated based on expression values of this gene across all samples from all tissues, and σ is the standard deviation. 4) Reference specific was generated the same way as reference universal except that only samples of the same tissue types were used. In the reference testing, 2000 different references were generated for reference normal, universal, and specific separately. For zero, we repeated the same reference 2000 times. ## Simulation for an optimal number of pseudo-samples generated for each sample A simulation was performed on 5 different pre-trained models as follows, using sample X as an example. Step 1: we generated 50 pseudo-samples based on sample X by randomly adding noise to each gene’s expression level with normal distribution N [0, 1]. Step 2: we estimated gene contribution scores for all genes in each pseudo-sample. To estimate gene contribution scores based on n pseudo-samples, we randomly selected n replicates out of 50, and the final gene contribution score for a specific gene was calculated based on the mean of n scores. Step 3: repeat steps [1] and [2] on 5 different pre-trained models, respectively. Step 4: for sample X, there will be 5 replicates of gene contribution scores based on the same number of pseudo-samples. For the 5 replicates based on n ($$n = 1$$, 2, …, 100) pseudo-samples, we calculated Spearman’s correlation coefficient on gene contribution scores from any two replicates, and this operation was carried out on all C52 combinations. Based on the aforementioned method, we obtained the relationship between a number of pseudo-samples and the correlation coefficient on any two replicates. ## Simulation for optimal repeat number on the same model The simulation process was performed on 5 different pre-trained models as follows, using sample X as an example. Step 1: first, we estimated gene contribution scores for all genes in sample X 50 times respectively, and there were 50 replicates for sample X. To estimate gene contribution scores by n times repeats, we randomly selected n replicates out of 50, and the final gene contribution score for a specific gene was calculated based on the mean of n scores. Step 2: repeat step [1] on 5 different pre-trained models, respectively. Step 3: for sample X, there will be 5 replicates of gene contribution scores based on the same repeat number. For the 5 replicates based on n ($$n = 1$$, 2, …, 100) times of repeats, we calculated Spearman’s correlation coefficient on gene contribution scores from any two replicates, and this operation was carried out on all C52 combinations. Based on the aforementioned method, we obtained the relationship between repeat number and correlation coefficient on any two replicates. ## Simulation for an optimal number of references The simulation process was performed on 5 different pre-trained models as follows, using sample X as an example. Step 1: we estimated gene contribution scores for all genes in sample X with 1, 2, 3, …, 100 reference samples, respectively, and the reference samples were randomly selected from the 2000 background samples pool. Step 2: repeat step [1] on 5 different pre-trained models. Step 3: for sample X, there will be 5 replicates of gene contribution scores based on the same number of reference samples but different pre-trained models. For the 5 replicates based on n ($$n = 1$$, 2, …, 100) reference samples, we calculated Spearman’s correlation coefficient on gene contribution scores from any two replicates, and this operation was carried out on all C52 combinations. Based on the aforementioned method, we obtained the relationship between a number of reference samples and the correlation coefficient on any two replicates. For each type of reference, we repeated the aforementioned simulation process individually. ## Simulation for an optimal number of aggregated models The simulation process was performed on 60 different pre-trained models as follows, using sample X as an example. Step 1: we estimated gene contribution scores for all genes in sample X on each pre-trained model, respectively. Step 2: to estimate gene contribution scores by aggregating n ($$n = 1$$, 2, …, 20) models, we randomly selected n replicates out of 60, and the final gene contribution score for a specific gene was calculated based on the mean of n scores. Step 3: repeat step [2] K times, where K=max(60n,4). Step 4: for sample X, there will be K replicates of gene contribution scores based on the same number of aggregated models. For these K replicates based on n ($$n = 1$$, 2, …, 20) aggregated models, we calculated Spearman’s correlation coefficient on gene contribution scores from any two replicates, and this operation was carried out on all CK2 combinations. Based on the aforementioned method, we obtained the relationship between repeat number and correlation coefficient on any two replicates. ## Gene classification Tissue specifically expressed genes were identified by the tool TissueEnrich with the group “Tissue-Enhanced” [45]. In each tissue, the median expression level of each gene was calculated across all samples, and HK genes are defined as genes with TPM ≥ 1 and less than 2-fold change on median expression level among all tissue types [46]. ## Gene Ontology enrichment analysis Genes of interest were extracted and imported into the gene ontology online tool for GO enrichment analysis with the options “molecular function” or “biological process” and “Homo sapiens” checked [47], [48]. ## Annotation of TF and TF cofactors All TFs and TF cofactors were downloaded from animalTFDB [49]. In total, there were 1666 TFs and 1026 TF cofactors. ## Differentially expressed genes between normal and cancer Mann–Whitney U test (two-sided) was used to compare gene expression between normal and cancer tissues. Differentially expressed genes should satisfy the following criteria: false discovery rate (FDR) ≤ 0.001 and fold change ≥ 3. ## Code availability Scripts used to test model interpretability are based on Python and are freely available at https://github.com/zhaopage/model_interpretability. ## CRediT author statement Yongbing Zhao: Conceptualization, Methodology, Software, Formal analysis, Investigation, Writing - original draft, Writing - review & editing. Jinfeng Shao: Investigation, Writing - review & editing. Yan W. Asmann: Funding acquisition, Investigation, Writing - review & editing. All authors have read and approved the final manuscript. ## Competing interests The authors declare no competing interests. ## Supplementary material The following are the Supplementary material to this article:Supplementary Figure S1Performance of model neural network interpretability A. Spearman’s correlation on gene contribution scores and (B) overlap in the top 100 contributing genes in lung (upper panel), ovary (middle panel), and pancreas (lower panel) among replicates from the same pre-trained model, different pre-trained models with the same gene order, and different pre-trained models with different gene orders based on CNN. C. Spearman’s correlation on gene contribution scores and D. overlap in the top 100 contributing genes in lung (upper panel), ovary (middle panel), and pancreas (lower panel) among replicates from the same pre-trained model and different pre-trained models based on MLP. CNN, convolutional neural network; MLP, multilayer perceptron. Supplementary Figure S2Performance of repeat with adding noise Spearman’s correlation on gene contribution scores in lung (left panel), ovary (middle panel), and pancreas (right panel) among replicates from (A) different pre-trained models with different gene orders on CNN and from (B) different pre-trained models on MLP. Supplementary Figure S3Performance of simple repeat Spearman’s correlation on gene contribution scores in lung (left panel), ovary (middle panel), and pancreas (right panel) among replicates from (A) different pre-trained models with different gene orders on CNN and from (B) different pre-trained models on MLP. Supplementary Figure S4Performance of reference type Spearman’s correlation on gene contribution scores among replicates from (A) different pre-trained models with different gene orders on CNN and from (B) different pre-trained models on MLP. Supplementary Figure S5Performance of different reference types with simple repeat Spearman’s correlation on gene contribution scores among replicates from different pre-trained models with different gene orders on CNN. Supplementary Figure S6Performance of model aggregation Spearman’s correlation on gene contribution scores in lung (left panel), ovary (middle panel), and pancreas (right panel) among replicates from (A) different pre-trained models with different gene orders on CNN and from (B) different pre-trained models on MLP. Supplementary Figure S7Optimization on different model explainers A. Spearman’s correlation on gene contribution scores and (B) overlap in the top 100 contributing genes in lung (upper panel), ovary (middle panel), and pancreas (lower panel) among replicates from different pre-trained models with different gene orders on CNN. The analyses were carried out with different optimization strategies: without optimization, with optimized conditions for each explainer but without model aggregation, and with optimized conditions for each explainer and with model aggregation. C. and D. similar analysis as (A) and (B) but based on MLP. Supplementary Figure S8Overlap between the top contributing genes with and without optimization Comparing the top 100 contributing genes with an optimized method with the top 100 contributing genes in individual replicates without optimization (dark blue) and the top 100 contributing genes shared in 2 or more replicates without optimization (yellow). Supplementary Figure S9Spearman’s correlation between gene contribution scores and gene expression level Left panels are analyses based on CNN with and without model aggregation, while right panels are based on MLP. Supplementary Figure S10Overlap between the top 100 contributing genes and top 100 expressed genes Left panels are analyses based on CNN with and without model aggregation, while right panels are based on MLP. Supplementary Figure S11Overlap between the top 100 contributing genes and TS expressed genes A. Left panels are analyses based on CNN with and without model aggregation, while right panels are based on MLP. B. Number of TS genes in liver, lung, ovary, and pancreas. TS, tissue-specific. Supplementary Figure S12Overlap between the top 100 contributing genes and HK genes Left panels are analyses based on CNN with and without model aggregation, while right panels are based on MLP. HK, housekeeping. Supplementary Figure S13Overlap between the top 100 contributing genes and TFs Left panels are analyses based on CNN with and without model aggregation, while right panels are based on MLP. TF, transcription factor. Supplementary Figure S14Overlap between the top 100 contributing genes and TF cofactors Left panels are analyses based on CNN with and without model aggregation, while right panels are based on MLP. Supplementary Figure S15Expression level of the top 100 contributing genes between normal and cancer tissues Supplementary Figure S16Heatmap of 45 tumor oncogenes and suppressor genes identified as shared top contributing genes across 33 cancer types Supplementary Table S1Information for different model explainers Supplementary Table S2Information for 82 different tissues and cell types used in model training Supplementary Table S3Overlaps in the top 100 contributing genes across different explainers Supplementary Table S4Overlaps in the top 100 contributing genes across tissues in different explainers Supplementary Table S5Gene Ontology enrichment on molecular function on the top 100 genes in different explainers ## References 1. Jumper J., Evans R., Pritzel A., Green T., Figurnov M., Ronneberger O.. Highly accurate protein structure prediction with AlphaFold. Nature (2021.0) 596 583-589. PMID: 34265844 2. Elmarakeby H.A., Hwang J., Arafeh R., Crowdis J., Gang S., Liu D.. Biologically informed deep neural network for prostate cancer discovery. Nature (2021.0) 598 348-352. PMID: 34552244 3. Xia B., Zhao D., Wang G., Zhang M., Lv J., Tomoiaga A.S.. Machine learning uncovers cell identity regulator by histone code. Nat Commun (2020.0) 11 2696. PMID: 32483223 4. Chantsalnyam T., Lim D.Y., Tayara H., Chong K.T.. ncRDeep: non-coding RNA classification with convolutional neural network. Comput Biol Chem (2020.0) 88 107364. PMID: 32890916 5. Zhang J., Peng W., Wang L.. LeNup: learning nucleosome positioning from DNA sequences with improved convolutional neural networks. Bioinformatics (2018.0) 34 1705-1712. PMID: 29329398 6. Nair S., Kim D.S., Perricone J., Kundaje A.. Integrating regulatory DNA sequence and gene expression to predict genome-wide chromatin accessibility across cellular contexts. Bioinformatics (2019.0) 35 i108-i116. PMID: 31510655 7. Kelley D.R., Reshef Y.A., Bileschi M., Belanger D., McLean C.Y., Snoek J.. Sequential regulatory activity prediction across chromosomes with convolutional neural networks. Genome Res (2018.0) 28 739-750. PMID: 29588361 8. Kelley D.R., Snoek J., Rinn J.L.. Basset: learning the regulatory code of the accessible genome with deep convolutional neural networks. Genome Res (2016.0) 26 990-999. PMID: 27197224 9. Avsec Z., Agarwal V., Visentin D., Ledsam J.R., Grabska-Barwinska A., Taylor K.R.. Effective gene expression prediction from sequence by integrating long-range interactions. Nat Methods (2021.0) 18 1196-1203. PMID: 34608324 10. Angelov P.P., Soares E.A., Jiang R., Arnold N.I., Atkinson P.M.. Explainable artificial intelligence: an analytical review. Wiley Interdiscip Rev Data Min Knowl Disc (2021.0) 11 e1424 11. Barredo Arrieta A., Díaz-Rodríguez N., Del Ser J., Bennetot A., Tabik S., Barbado A.. Explainable artificial intelligence (XAI): concepts, taxonomies, opportunities and challenges toward responsible AI. Inf Fusion (2020.0) 58 82-115 12. Anguita-Ruiz A., Segura-Delgado A., Alcala R., Aguilera C.M., Alcala-Fdez J.. eXplainable artificial intelligence (XAI) for the identification of biologically relevant gene expression patterns in longitudinal human studies, insights from obesity research. PLoS Comput Biol (2020.0) 16 e1007792. PMID: 32275707 13. 13Alonso JM, Casalino G. Explainable artificial intelligence for human-centric data analysis in virtual learning environments. International Workshop on Higher Education Learning Methodologies and Technologies Online 2019:125–38. 14. Talukder A., Barham C., Li X., Hu H.. Interpretation of deep learning in genomics and epigenomics. Brief Bioinform (2020.0) 22 bbaa177 15. Zhou J., Troyanskaya O.G.. Predicting effects of noncoding variants with deep learning-based sequence model. Nat Methods (2015.0) 12 931-934. PMID: 26301843 16. Torng W., Altman R.B.. 3D deep convolutional neural networks for amino acid environment similarity analysis. BMC Bioinformatics (2017.0) 18 302. PMID: 28615003 17. Shrikumar A., Greenside P., Kundaje A.. Learning important features through propagating activation differences. Int Conf Mach Learn (2017.0) 3145-3153 18. Lundberg S.M., Lee S.I.. A unified approach to interpreting model predictions. Adv Neural Inf Process Syst (2017.0) 4765-4774 19. Avsec Z., Weilert M., Shrikumar A., Krueger S., Alexandari A., Dalal K.. Base-resolution models of transcription-factor binding reveal soft motif syntax. Nat Genet (2021.0) 53 354-366. PMID: 33603233 20. Kim D.S., Risca V.I., Reynolds D.L., Chappell J., Rubin A.J., Jung N.. The dynamic, combinatorial. Nat Genet (2021.0) 53 1564-1576. PMID: 34650237 21. 21Karim M, Cochez M, Beyan O, Decker S, Lange C. OncoNetExplainer: explainable predictions of cancer types based on gene expression data. arXiv 2019;1909.04169. 22. Lyu B., Haque A.. Deep learning based tumor type classification using gene expression data. Proceedings of the 2018 ACM International Conference on Bioinformatics, Computational Biology, and Health Informatics (2018.0) 89-96 23. Li Y., Kang K., Krahn J.M., Croutwater N., Lee K., Umbach D.M.. A comprehensive genomic pan-cancer classification using the cancer genome atlas gene expression data. BMC Genomics (2017.0) 18 508. PMID: 28673244 24. Simonyan K., Vedaldi A., Zisserman A.. Deep inside convolutional networks: visualising image classification models and saliency maps. arXiv (2013.0) 25. Shrikumar A., Greenside P., Shcherbina A., Kundaje A.. Not just a black box: learning important features through propagating activation differences. arXiv (2016.0) 1605.01713 26. 26Springenberg JT, Dosovitskiy A, Brox T, Riedmiller M. Striving for simplicity: the all convolutional net. arXiv 2014;1412.6806. 27. Sundararajan M., Taly A., Yan Q.. Axiomatic attribution for deep networks. Proc 34th Int Conf Mach Learn (2017.0) 40 3319-3328 28. Selvaraju R.R., Cogswell M., Das A., Vedantam R., Parikh D., Batra D.. Grad-CAM: visual explanations from deep networks via gradient-based localization. Proc IEEE Int Conf Comput Vis (2017.0) 2017 618-626 29. Chattopadhay A., Sarkar A., Howlader P., Balasubramanian V.N.. Grad-CAM++: generalized gradient-based visual explanations for deep convolutional networks. IEEE Winter Conf Appl Comput Vis (2018.0) 2018 839-847 30. Hartley M., Olsson T.S.G.. dtoolAI: reproducibility for deep learning. Patterns (N Y) (2020.0) 1 100073. PMID: 33205122 31. Fan F., Xiong J., Li M., Wang G.. On interpretability of artificial neural networks: a survey. IEEE Trans Radiat Plasma Med Sci (2021.0) 5 741-760. PMID: 35573928 32. 32Smilkov D, Thorat N, Kim B, Viégas F, Wattenberg M. Smoothgrad: removing noise by adding noise. arXiv 2017;1706.03825. 33. Guo M., Zhang H., Zheng J., Liu Y.. Glypican-3: a new target for diagnosis and treatment of hepatocellular carcinoma. J Cancer (2020.0) 11 2008-2021. PMID: 32127929 34. Chakravarty D., Gao J., Phillips S., Kundra R., Zhang H., Wang J.. OncoKB: a precision oncology knowledge base. JCO Precis Oncol (2017.0) 1 1-16 35. Whalen S., Schreiber J., Noble W.S., Pollard K.S.. Navigating the pitfalls of applying machine learning in genomics. Nat Rev Genet (2021.0) 23 169-181. PMID: 34837041 36. Carvalho D.V., Pereira E.M., Cardoso J.S.. Machine learning interpretability: a survey on methods and metrics. Electronics (2019.0) 8 832 37. Toyoda M., Hamatani T., Okada H., Matsumoto K., Saito H., Umezawa A.. Defining cell identity by comprehensive gene expression profiling. Curr Med Chem (2010.0) 17 3245-3252. PMID: 20666720 38. Ye Z., Sarkar C.A.. Towards a quantitative understanding of cell identity. Trends Cell Biol (2018.0) 28 1030-1048. PMID: 30309735 39. Sonawane A.R., Platig J., Fagny M., Chen C.Y., Paulson J.N., Lopes-Ramos C.M.. Understanding tissue-specific gene regulation. Cell Rep (2017.0) 21 1077-1088. PMID: 29069589 40. Uhlen M., Fagerberg L., Hallstrom B.M., Lindskog C., Oksvold P., Mardinoglu A.. Tissue-based map of the human proteome. Science (2015.0) 347 1260419. PMID: 25613900 41. Morris S.A.. The evolving concept of cell identity in the single cell era. Development (2019.0) 146 dev169748. PMID: 31249002 42. Stuart T., Satija R.. Integrative single-cell analysis. Nat Rev Genet (2019.0) 20 257-272. PMID: 30696980 43. The Genotype-Tissue Expression (GTEx) project. Nat Genet (2013.0) 45 580-585. PMID: 23715323 44. Collado-Torres L., Nellore A., Kammers K., Ellis S.E., Taub M.A., Hansen K.D.. Reproducible RNA-seq analysis using recount2. Nat Biotechnol (2017.0) 35 319-321. PMID: 28398307 45. Jain A., Tuteja G.. TissueEnrich: tissue-specific gene enrichment analysis. Bioinformatics (2019.0) 35 1966-1967. PMID: 30346488 46. Eisenberg E., Levanon E.Y.. Human housekeeping genes, revisited. Trends Genet (2013.0) 29 569-574. PMID: 23810203 47. Ashburner M., Ball C.A., Blake J.A., Botstein D., Butler H., Cherry J.M.. Gene Ontology: tool for the unification of biology. Nat Genet (2000.0) 25 25-29. PMID: 10802651 48. The Gene Ontology resource: 20 years and still GOing strong. Nucleic Acids Res (2019.0) 47 D330-D338. PMID: 30395331 49. Hu H., Miao Y.R., Jia L.H., Yu Q.Y., Zhang Q., Guo A.Y.. AnimalTFDB 3.0: a comprehensive resource for annotation and prediction of animal transcription factors. Nucleic Acids Res (2019.0) 47 D33-D38. PMID: 30204897
--- title: Relationships of habitual daily alcohol consumption with all-day and time-specific average glucose levels among non-diabetic population samples authors: - Maho Ishihara - Hironori Imano - Isao Muraki - Kazumasa Yamagishi - Koutatsu Maruyama - Mina Hayama-Terada - Mari Tanaka - Mikako Yasuoka - Tomomi Kihara - Masahiko Kiyama - Takeo Okada - Midori Takada - Yuji Shimizu - Tomotaka Sobue - Hiroyasu Iso journal: Environmental Health and Preventive Medicine year: 2023 pmcid: PMC10025860 doi: 10.1265/ehpm.22-00215 license: CC BY 4.0 --- # Relationships of habitual daily alcohol consumption with all-day and time-specific average glucose levels among non-diabetic population samples ## Abstract ### Background Alcohol consumption is a prevalent behavior that is bi-directionally related to the risk of type 2 diabetes. However, the effect of daily alcohol consumption on glucose levels in real-world situations in the general population has not been well elucidated. This study aimed to clarify the relationship between alcohol consumption and all-day and time-specific glucose levels among non-diabetic individuals. ### Methods We investigated 913 non-diabetic males and females, aged 40–69 years, during 2018–2020 from four communities across Japan. The daily alcohol consumption was assessed using a self-report questionnaire. All-day and time-specific average glucose levels were estimated from the interstitial glucose concentrations measured using the Flash glucose monitoring system for a median duration of 13 days. Furthermore, we investigated the association between all-day and time-specific average glucose levels and habitual daily alcohol consumption levels, using never drinkers as the reference, and performed multiple linear regression analyses after adjusting for age, community, and other diabetes risk factors for males and females separately. ### Results All-day average glucose levels did not vary according to alcohol consumption categories in both males and females. However, for males, the average glucose levels between 5:00 and 11:00 h and between 11:00 and 17:00 h were higher in moderate and heavy drinkers than in never drinkers, with the difference values of 4.6 and 4.7 mg/dL for moderate drinkers, and 5.7 and 6.8 mg/dL for heavy drinkers. Conversely, the average glucose levels between 17:00 and 24:00 h were lower in male moderate and heavy drinkers and female current drinkers than in never drinkers; the difference values of mean glucose levels were −5.8 for moderate drinkers, and −6.1 mg/dL for heavy drinkers in males and −2.7 mg/dL for female current drinkers. ### Conclusions Alcohol consumption was associated with glucose levels in a time-dependent biphasic pattern. ## Background The prevalence of diabetes mellitus (DM) and impaired glucose tolerance is rapidly increasing worldwide. According to the International Diabetes Federation (IDF), the worldwide diabetic population reached 537 million in 2021, indicating that one in ten adults has diabetes mellitus [1]. Longstanding high blood glucose levels can damage blood vessels, leading to various health problems, such as retinopathy, renal disease, peripheral neuropathy [2], coronary heart disease, stroke [3, 4], and deaths attributable to diabetes mellitus. Healthcare expenditure due to diabetes mellitus is also on the rise, creating a significant social, financial, and healthcare system burden worldwide [5, 6]. Therefore, establishing evidence-based preventive measures against diabetes mellitus is an urgent issue. Alcohol consumption is a prevalent behavior that is bi-directionally related to the risk of type 2 diabetes. Light drinking could lower this risk, whereas heavy drinking could increase it [7–9]. However, the effect of alcohol consumption on glucose metabolism in daily life has not been well elucidated. A clinical experimental study of five non-diabetic young adult males showed that the ingestion of 48 g ethanol reduced glycogenesis by $45\%$, although plasma glucose concentrations did not change [10]. A randomized controlled trial of 51 non-diabetic postmenopausal females demonstrated that the ingestion of 30 g/day ethanol for eight weeks lowered fasting serum insulin concentrations by $20\%$ but did not affect fasting plasma glucose levels compared with those in the placebo group [11]. However, these previous experimental studies did not capture the impact of daily alcohol consumption on glucose levels in real-world situations in the general population. Therefore, we aimed to clarify the relationship between habitual daily alcohol consumption and all-day and time-specific glucose levels in daily life using a Flash glucose monitoring system in community-based non-diabetic samples. ## Study subjects We included 1260 non-diabetic individuals (377 males and 883 females) who provided consent to participate in this study, were aged 40–69 years, and hailed from four communities, namely Ikawa town, Akita Prefecture (a northwestern rural community); Minami-Takayasu district, Yao City, Osaka Prefecture (a midwestern suburb); Kyowa district, Chikusei City, Ibaraki Prefecture (a mideastern rural community); and Kamisu City, Ibaraki Prefecture (an industrial area), in Japan. The first three communities were sub-cohorts of the Circulatory Risk in Communities Study (CIRCS), an ongoing dynamic cohort study on lifestyle-related diseases involving approximately 12000 [12]. In Kamisu City, approximately 8000 adults receive health checkups annually, and the subjects were selected from among them. Surveys of daily glucose monitoring were conducted in 2019 in Ikawa, 2018–2019 in Minami-Takayasu, 2018–2020 in Kyowa, and 2019–2020 in Kamisu. We excluded those participants whose hemoglobin A1c (HbA1c) data were missing ($$n = 10$$), whose HbA1c levels were ≥$6.0\%$ (42 mmol/mol) ($$n = 258$$), who were wearing the Flash glucose monitoring system for less than 3 days ($$n = 49$$), who declined to answer about regular alcohol consumption ($$n = 24$$), who declined to answer about usual exercise habits ($$n = 5$$), or who were pregnant ($$n = 1$$). In total, 913 participants (277 males and 636 females) were included in the present analysis (Fig. 1). Fig. 1: Flowchart of study participants of the present study ## Measurements Each participant’s habitual daily alcohol consumption was assessed using a questionnaire. Participants were asked combined questions 1. whether they drank alcohol, and for current drinkers, 2. the amount of alcohol consumed per day in go-units (a Japanese traditional unit of volume equivalent to 23 g of ethanol). Alcohol consumption was categorized into five groups (never, former, Light; current <23, Moderate; 23–45, and Heavy; ≥46 g/day ethanol) for males and into three groups (never, former, and Current) for females. The following covariates were also elicited by the questionnaire: the number of cigarettes smoked per day, their medical history, family history of diabetes mellitus, frequency and duration per occasion of physical activity during the past year (namely, for frequency: <1 /month, 1–3 /month, 1–2 /week, 3–4 /week, and every day; for time, <30 minutes, 30–59 minutes, 1–2 hours, 2–3 hours, and ≥4 hours), and skipping breakfast (yes or no). Blood was collected from the participants in plastic serum separator gel tubes. The serum was allowed to stand for 15–20 minutes after collection and centrifuged for 15 minutes at 3000 rpm in a centrifuge with a turning radius of 16 to 18 cm within 30 minutes. In Ikawa and Minami-Takayasu, serum samples were transported to the Osaka Center for Cancer and Cardiovascular Disease Prevention, and serum glucose levels were measured using the hexokinase and glucose-6-phosphate dehydrogenase methods and an automatic analyzer (TBA-2000FR; Toshiba, Otawara, Tochigi, Japan), whereas HbA1c levels were measured using high-performance liquid chromatography with an HLC-723 G8 (Tosoh, Minato-ku, Tokyo, Japan). In kyowa and Kamisu, serum samples were transported to the Ibaraki Health Service Association, and serum glucose levels were measured using the hexokinase and ultraviolet absorption spectrophotometric methods, while HbA1c levels were assessed via enzymatic methods using an automatic analyzer (JCA-BM9130; JEOL Ltd, Tokyo, Japan). The body mass index (BMI) was calculated as weight in light clothing (kg) divided by height squared in stocking feet (m2). The systolic and diastolic blood pressure levels were measured in the right arm by trained observers according to the unified epidemiological method, using automatic sphygmomanometers, except for participants in Minami-Takayasu in 2019, for whom standard mercury sphygmomanometers were used. ## Measurement of the daily glucose levels Glucose concentrations in the interstitial fluid were measured every 15 min for up to 15 days using a Flash glucose monitoring system (FGM; FreeStyle Libre Pro System, Abbott Diabetes Care, Inc. Alameda, CA, USA) in the upper left arm. We monitored glucose continuously, and analyzed date from day 2 through to the second-to-last day of recording. This was done because Freestyle Libre Pro is reported to be less accurate on the first day of measurement [13]. In addition, fewer data could be collected on the first and last days; hence, the average glucose values could not be calculated reliably. Finally, because of differences in the number of days of measurement among participants, the analyzed data amounted to one day in the shortest case and 13 days in the longest case. The distribution of the participants who had worn the FGM sensor for each span of days is shown in Table 1. For time-of-day classifications, because the nadirs of the daily glucose levels were found at 0:00, 5:00, 11:00, and 17:00 h (Fig. 2), we categorized the time range as follows: all-day, 0:00 to 5:00, 5:00 to 11:00, 11:00 to 17:00, and 17:00 to 24:00 h. The time-specific average glucose levels in each alcohol consumption category were defined as the averaged values of individual mean glucose levels, which were calculated by averaging the measurement data of each participant in each time range without adjustment for sex, age, and community. ## Statistical analysis We compared the time-specific average glucose levels according to usual alcohol consumption with reference to never drinking, using multiple linear regression for males and females separately. The adjustment values included age (continuous), community (Ikawa, Minami-Takayasu, Chikusei, Kamisu), and other confounding variables, such as habitual smoking (never, former, current), the BMI (continuous), habitual physical activity (exercise at least 30 minutes a day at least once a week, yes or no), family history of diabetes mellitus (yes or no), and skipping breakfast (yes or no). We conducted further adjustment for HbA1c levels at screening, after which HbA1c was categorized into two groups (≤5.5 and 5.6–$5.9\%$, ≤37 and 38–41 mmol/mol) and stratified the HbA1c groups. All analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC, USA). P-values < 0.05 were considered to indicate statistical significance in two-tailed analyses. ## Results Of all participants, 667 ($73\%$) wore the FGM sensor for full days. Table 2 shows the baseline characteristics according to the habitual alcohol consumption category among males and females. In males who were moderate and heavy alcohol consumers, the systolic and diastolic blood pressure, fasting and nonfasting blood glucose levels at screening, and the average glucose levels between 5:00 and 11:00 h and between 11:00 and 17:00 h were higher than those in others. Male heavy drinkers smoked more than others. Similarly, females who were current drinkers had higher fasting and nonfasting blood glucose levels as well as increased proportions of current smoking and skipping breakfast. Table 2 \| Unnamed: 0 \| Alcohol consumption \| Alcohol consumption.1 \| Alcohol consumption.2 \| Alcohol consumption.3 \| Alcohol consumption.4 \| Alcohol consumption.5 \| Alcohol consumption.6 \| Alcohol consumption.7 \| Alcohol consumption.8 \| Alcohol consumption.9 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| \| Never \| Never \| Former \| Former \| Light <23 \| Light <23 \| Moderate 23–45 \| Moderate 23–45 \| Heavy ≥46 \| Heavy ≥46 \| \| Males \| Males \| Males \| Males \| Males \| Males \| Males \| Males \| Males \| Males \| Males \| \| No. of participants, n \| 77 \| 77 \| 13 \| 13 \| 77 \| 77 \| 51 \| 51 \| 59 \| 59 \| \| Age, years \| 55.7 \| (9.1) \| 58.4 \| (8.9) \| 56.5 \| (9.1) \| 58.7 \| (8.0) \| 56.3 \| (9.1) \| \| Average glucose level, mg/dL \| Average glucose level, mg/dL \| Average glucose level, mg/dL \| Average glucose level, mg/dL \| Average glucose level, mg/dL \| Average glucose level, mg/dL \| Average glucose level, mg/dL \| Average glucose level, mg/dL \| Average glucose level, mg/dL \| Average glucose level, mg/dL \| Average glucose level, mg/dL \| \| All-day \| 100.5 \| (10.6) \| 98.6 \| (13.4) \| 98.2 \| (10.7) \| 101.5 \| (9.4) \| 99.3 \| (12.2) \| \| 0:00 to 5:00 h \| 88.9 \| (11.6) \| 85.1 \| (12.3) \| 85.6 \| (12.1) \| 89.8 \| (11.6) \| 86.7 \| (12.6) \| \| 5:00 to 11:00 h \| 97.6 \| (11.9) \| 94.7 \| (14.0) \| 97.1 \| (11.7) \| 102.7 \| (11.3) \| 100.3 \| (15.2) \| \| 11:00 to 17:00 h \| 106.9 \| (12.1) \| 106.2 \| (16.7) \| 105.3 \| (12.5) \| 112.0 \| (10.5) \| 114.4 \| (14.7) \| \| 17:00 to 24:00 h \| 105.9 \| (12.1) \| 105.0 \| (14.4) \| 101.9 \| (13.1) \| 99.9 \| (10.3) \| 97.2 \| (12.2) \| \| HbA1c at screening, % \| 5.5 \| (0.2) \| 5.6 \| (0.2) \| 5.6 \| (0.2) \| 5.6 \| (0.2) \| 5.5 \| (0.3) \| \| HbA1c at screening, mmol/mol \| 37 \| (2.6) \| 38 \| (2.4) \| 38 \| (2.5) \| 38 \| (2.2) \| 37 \| (2.6) \| \| ≤5.5 (≤37 mmol/mol), % \| 48.1 \| 48.1 \| 38.5 \| 38.5 \| 45.5 \| 45.5 \| 52.5 \| 52.5 \| 60.0 \| 60.0 \| \| 5.6–5.9 (38–41 mmol/mol), % \| 52.0 \| 52.0 \| 61.5 \| 61.5 \| 54.6 \| 54.6 \| 47.5 \| 47.5 \| 40.0 \| 40.0 \| \| Fasting glucose at screening, mg/dL \| 94.2 \| (6.9) \| 93.2 \| (8.1) \| 93.7 \| (6.8) \| 96.4 \| (9.5) \| 97.3 \| (9.1) \| \| Non-fasting glucose at screening, mg/dL \| 95.4 \| (12.6) \| 94.0 \| (11.6) \| 99.0 \| (16.1) \| 97.9 \| (16.6) \| 107.1 \| (28.7) \| \| Body mass index, kg/m2 \| 23.7 \| (3.3) \| 23.9 \| (2.4) \| 23.8 \| (3.2) \| 24.1 \| (2.6) \| 23.5 \| (3.1) \| \| Waist Circumstance, cm \| 84.2 \| (9.6) \| 84.1 \| (7.6) \| 84.4 \| (9.3) \| 85.6 \| (7.4) \| 84.5 \| (8.6) \| \| Systolic blood pressure, mmHg \| 122.8 \| (14.0) \| 122.2 \| (13.4) \| 124.6 \| (13.5) \| 130.6 \| (15.0) \| 130.9 \| (14.9) \| \| Diastolic blood pressure, mmHg \| 77.1 \| (9.9) \| 79.5 \| (9.9) \| 78.0 \| (9.6) \| 83.4 \| (9.0) \| 83.1 \| (9.6) \| \| Antihypertensive medication, % \| 18.2 \| 18.2 \| 23.1 \| 23.1 \| 24.7 \| 24.7 \| 29.4 \| 29.4 \| 27.1 \| 27.1 \| \| Smoking habit, % \| Smoking habit, % \| Smoking habit, % \| Smoking habit, % \| Smoking habit, % \| Smoking habit, % \| Smoking habit, % \| Smoking habit, % \| Smoking habit, % \| Smoking habit, % \| Smoking habit, % \| \| never \| 32.5 \| 32.5 \| 15.4 \| 15.4 \| 29.9 \| 29.9 \| 13.7 \| 13.7 \| 6.8 \| 6.8 \| \| past \| 44.2 \| 44.2 \| 76.9 \| 76.9 \| 52.0 \| 52.0 \| 66.7 \| 66.7 \| 61.0 \| 61.0 \| \| current \| 23.4 \| 23.4 \| 7.7 \| 7.7 \| 18.2 \| 18.2 \| 19.6 \| 19.6 \| 32.2 \| 32.2 \| \| Exercise, % \| 39.0 \| 39.0 \| 61.5 \| 61.5 \| 41.6 \| 41.6 \| 47.1 \| 47.1 \| 37.3 \| 37.3 \| \| Family history of diabetes mellitus, % \| 6.5 \| 6.5 \| 15.4 \| 15.4 \| 6.5 \| 6.5 \| 2.0 \| 2.0 \| 13.6 \| 13.6 \| \| Skipping breakfast, % \| 26.0 \| 26.0 \| 7.7 \| 7.7 \| 18.2 \| 18.2 \| 15.7 \| 15.7 \| 22.0 \| 22.0 \| \| \| Never \| Never \| Former \| Former \| Current \| Current \| \| \| \| \| \| Females \| Females \| Females \| Females \| Females \| Females \| Females \| \| \| \| \| \| No. of participants, n \| 399 \| 399 \| 42 \| 42 \| 195 \| 195 \| \| \| \| \| \| Age, years \| 56.5 \| (8.4) \| 53.1 \| (8.3) \| 54.7 \| (8.0) \| \| \| \| \| \| Average glucose level, mg/dL \| Average glucose level, mg/dL \| Average glucose level, mg/dL \| Average glucose level, mg/dL \| Average glucose level, mg/dL \| Average glucose level, mg/dL \| Average glucose level, mg/dL \| \| \| \| \| \| All-day \| 97.6 \| (10.8) \| 93.5 \| (12.7) \| 95.9 \| (10.6) \| \| \| \| \| \| 0:00 to 5:00 h \| 83.4 \| (11.5) \| 80.7 \| (14.2) \| 82.9 \| (11.4) \| \| \| \| \| \| 5:00 to 11:00 h \| 92.8 \| (11.6) \| 88.6 \| (13.1) \| 92.7 \| (11.8) \| \| \| \| \| \| 11:00 to 17:00 h \| 107.4 \| (13.0) \| 102.6 \| (13.4) \| 106.6 \| (13.7) \| \| \| \| \| \| 17:00 to 24:00 h \| 103.5 \| (12.4) \| 98.9 \| (14.2) \| 98.7 \| (12.1) \| \| \| \| \| \| HbA1c at screening, % \| 5.6 \| (0.2) \| 5.6 \| (0.2) \| 5.5 \| (0.3) \| \| \| \| \| \| HbA1c at screening, mmol/mol \| 37 \| (2.5) \| 38 \| (2.1) \| 37 \| (2.9) \| \| \| \| \| \| ≤5.5 (≤37 mmol/mol), % \| 41.6 \| 41.6 \| 30.2 \| 30.2 \| 48.2 \| 48.2 \| \| \| \| \| \| 5.6–5.9 (38–41 mmol/mol), % \| 58.4 \| 58.4 \| 69.8 \| 69.8 \| 51.8 \| 51.8 \| \| \| \| \| \| Fasting glucose at screening, mg/dL \| 91.4 \| (6.8) \| 92.0 \| (6.6) \| 93.6 \| (9.1) \| \| \| \| \| \| Non-fasting glucose at screening, mg/dL \| 92.5 \| (13.2) \| 90.8 \| (10.1) \| 96.4 \| (17.1) \| \| \| \| \| \| Body mass index, kg/m2 \| 22.3 \| (3.6) \| 22.8 \| (4.5) \| 22.4 \| (3.7) \| \| \| \| \| \| Waist Circumstance, cm \| 79.0 \| (10.2) \| 80.5 \| (11.3) \| 79.7 \| (11.1) \| \| \| \| \| \| Systolic blood pressure, mmHg \| 120.9 \| (17.3) \| 116.6 \| (16.4) \| 119.8 \| (15.3) \| \| \| \| \| \| Diastolic blood pressure, mmHg \| 73.1 \| (10.5) \| 71.6 \| (9.9) \| 74.1 \| (10.4) \| \| \| \| \| \| Antihypertensive medication, % \| 10.3 \| 10.3 \| 7.0 \| 7.0 \| 5.1 \| 5.1 \| \| \| \| \| \| Smoking habit, % \| Smoking habit, % \| Smoking habit, % \| Smoking habit, % \| Smoking habit, % \| Smoking habit, % \| Smoking habit, % \| \| \| \| \| \| never \| 83.7 \| 83.7 \| 60.5 \| 60.5 \| 62.6 \| 62.6 \| \| \| \| \| \| past \| 10.3 \| 10.3 \| 30.2 \| 30.2 \| 22.6 \| 22.6 \| \| \| \| \| \| current \| 6.0 \| 6.0 \| 9.3 \| 9.3 \| 14.9 \| 14.9 \| \| \| \| \| \| Exercise, % \| 45.4 \| 45.4 \| 34.9 \| 34.9 \| 40.0 \| 40.0 \| \| \| \| \| \| Family history of diabetes mellitus, % \| 9.0 \| 9.0 \| 32.6 \| 32.6 \| 12.8 \| 12.8 \| \| \| \| \| \| Skipping breakfast, % \| 13.5 \| 13.5 \| 7.0 \| 7.0 \| 19.0 \| 19.0 \| \| \| \| \| Figure 3 shows the time-specific variations of average glucose levels among alcohol consumption categories, and Table 3 shows the time-specific predicted differences in average glucose levels according to alcohol consumption categories among males and females. The all-day average glucose levels did not vary among alcohol consumption categories in either males or females. Among males, the time-specific average glucose levels between 5:00 and 11:00 h and between 11:00 and 17:00 h were approximately 5 to 7 mg/dL higher in moderate and heavy drinkers than in never drinkers, according to model 3 of the multivariable models. Contrastingly, the time-specific average glucose levels between 17:00 and 24:00 h were 6 mg/dL lower in male moderate and heavy drinkers and 3 mg/dL lower in female drinkers than in never drinkers, according to model 3. Fig. 3: Time-specific glucose levels according to alcohol consumption category among males and females. TABLE_PLACEHOLDER:Table 3 Table 4 shows the results stratified by HbA1c levels at screening. The higher time-specific average glucose levels between 5:00 and 11:00 h and between 11:00 and 17:00 h in male heavy drinkers were more evident among males with 5.6–$5.9\%$ (38–41 mmol/mol) HbA1c levels. The lower time-specific average glucose levels between 17:00 and 24:00 h in male moderate and heavy drinkers and female drinkers were more evident among individuals with HbA1c ≤ $5.5\%$ (≤37 mmol/mol). The lower time-specific average glucose levels between 17:00 and 24:00 h in male light drinkers were more evident among males with HbA1c levels of 5.6–$5.9\%$ (38–41 mmol/mol). Table 4 \| Unnamed: 0 \| Alcohol consumption \| Alcohol consumption.1 \| Alcohol consumption.2 \| Alcohol consumption.3 \| Alcohol consumption.4 \| Alcohol consumption.5 \| Alcohol consumption.6 \| Alcohol consumption.7 \| Alcohol consumption.8 \| Alcohol consumption.9 \| Alcohol consumption.10 \| Alcohol consumption.11 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| \| Never \| Former \| \| Light <23 \| \| \| Moderate 23–45 \| \| \| Heavy ≥46 \| \| \| \| \| \| β (95%CI) \| P-value \| β (95%CI) \| P-value \| P-value \| β (95%CI) \| P-value \| P-value \| β (95%CI) \| P-value \| P-value \| \| Males \| Males \| Males \| Males \| Males \| Males \| Males \| Males \| Males \| Males \| Males \| Males \| Males \| \| No. of participants, n \| 37 \| 5 \| \| 35 \| \| \| 20 \| \| \| 31 \| \| \| \| HbA1c ≤ 5.5% \| HbA1c ≤ 5.5% \| HbA1c ≤ 5.5% \| HbA1c ≤ 5.5% \| HbA1c ≤ 5.5% \| HbA1c ≤ 5.5% \| HbA1c ≤ 5.5% \| HbA1c ≤ 5.5% \| HbA1c ≤ 5.5% \| HbA1c ≤ 5.5% \| HbA1c ≤ 5.5% \| HbA1c ≤ 5.5% \| HbA1c ≤ 5.5% \| \| all-day \| all-day \| all-day \| all-day \| all-day \| all-day \| all-day \| all-day \| all-day \| all-day \| all-day \| all-day \| all-day \| \| model 1 \| Ref. \| 1.8(−8.2,11.8) \| .718 \| 0.1(−4.4,4.6) \| .952 \| \| −0.2(−5.8,5.4) \| .943 \| \| −0.9(−5.7,3.9) \| .724 \| \| \| model 2 \| Ref. \| −0.2(−10.4,10.0) \| .965 \| −0.4(−4.9,4.1) \| .863 \| \| −2.5(−8.2,3.3) \| .396 \| \| −3.0(−8.1,2.1) \| .246 \| \| \| model 3 \| Ref. \| −0.8(−10.8,9.2) \| .880 \| −1.0(−5.4,3.5) \| .665 \| \| −2.9(−8.5,2.7) \| .313 \| \| −2.1(−7.2,2.9) \| .398 \| \| \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| \| model 1 \| Ref. \| −3.6(−14.7,7.6) \| .531 \| −2.2(−7.3,2.8) \| .384 \| \| −4.7(−11.0,1.5) \| .138 \| \| −2.9(−8.2,2.5) \| .296 \| \| \| model 2 \| Ref. \| −5.7(−16.8,5.4) \| .313 \| −2.4(−7.4,2.5) \| .328 \| \| −7.3(−13.5,−1.1) \| .022 \| * \| −4.9(−10.4,0.6) \| .083 \| \| \| model 3 \| Ref. \| −6.4(−17.1,4.4) \| .244 \| −3.2(−8.0,1.6) \| .191 \| \| −7.8(−13.9,−1.8) \| .012 \| * \| −3.8(−9.2,1.6) \| .166 \| \| \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| \| model 1 \| Ref. \| 2.6(−9.0,14.2) \| .660 \| 1.7(−3.5,7.0) \| .511 \| \| 4.6(−1.9,11.0) \| .168 \| \| 3.9(−1.8,9.5) \| .176 \| \| \| model 2 \| Ref. \| −0.2(−12.2,11.8) \| .974 \| 0.7(−4.6,6.0) \| .799 \| \| 2.1(−4.7,8.8) \| .544 \| \| 1.6(−4.3,7.6) \| .585 \| \| \| model 3 \| Ref. \| −0.8(−12.6,11.0) \| .895 \| 0.0(−5.2,5.3) \| .988 \| \| 1.6(−5.0,8.2) \| .629 \| \| 2.6(−3.3,8.5) \| .388 \| \| \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| \| model 1 \| Ref. \| 3.4(−9.0,15.8) \| .591 \| 1.6(−4.0,7.2) \| .564 \| \| 5.6(−1.3,12.6) \| .112 \| \| 4.6(−1.4,10.6) \| .128 \| \| \| model 2 \| Ref. \| 1.1(−11.5,13.8) \| .859 \| 0.9(−4.7,6.5) \| .758 \| \| 3.5(−3.6,10.5) \| .336 \| \| 2.1(−4.1,8.4) \| .503 \| \| \| model 3 \| Ref. \| 0.8(−11.8,13.4) \| .904 \| 0.5(−5.2,6.1) \| .868 \| \| 3.2(−3.9,10.3) \| .376 \| \| 2.7(−3.6,9.0) \| .397 \| \| \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| \| model 1 \| Ref. \| 3.6(−7.4,14.5) \| .519 \| −0.9(−5.8,4.1) \| .727 \| \| −6.1(−12.2,0.0) \| .051 \| \| −8.1(−13.4,−2.9) \| .003 \| † \| \| model 2 \| Ref. \| 2.4(−8.8,13.5) \| .676 \| −1.0(−5.9,4.0) \| .702 \| \| −8.0(−14.3,−1.7) \| .013 \| * \| −10.0(−15.5,−4.4) \| .001 \| ‡ \| \| model 3 \| Ref. \| 1.9(−9.2,12.9) \| .740 \| −1.5(−6.4,3.4) \| .541 \| \| −8.4(−14.6,−2.2) \| .008 \| † \| −9.1(−14.7,−3.6) \| .001 \| † \| \| Hb A1c 5.6%–5.9% \| Hb A1c 5.6%–5.9% \| Hb A1c 5.6%–5.9% \| Hb A1c 5.6%–5.9% \| Hb A1c 5.6%–5.9% \| Hb A1c 5.6%–5.9% \| Hb A1c 5.6%–5.9% \| Hb A1c 5.6%–5.9% \| Hb A1c 5.6%–5.9% \| Hb A1c 5.6%–5.9% \| Hb A1c 5.6%–5.9% \| Hb A1c 5.6%–5.9% \| Hb A1c 5.6%–5.9% \| \| No. of participants, n \| 40 \| 8 \| \| 42 \| \| \| 31 \| \| \| 28 \| \| \| \| all-day \| all-day \| all-day \| all-day \| all-day \| all-day \| all-day \| all-day \| all-day \| all-day \| all-day \| all-day \| all-day \| \| model 1 \| Ref. \| 3.6(−4.3,11.6) \| .368 \| −3.8(−8.3,0.7) \| .094 \| \| 2.3(−2.7,7.2) \| .364 \| \| 2.9(−2.2,7.9) \| .267 \| \| \| model 2 \| Ref. \| 3.6(−4.6,11.9) \| .387 \| −3.8(−8.4,0.8) \| .108 \| \| 2.2(−2.9,7.2) \| .397 \| \| 2.9(−2.4,8.1) \| .281 \| \| \| model 3 \| Ref. \| 3.3(−4.9,11.6) \| .426 \| −3.7(−8.3,0.9) \| .113 \| \| 2.0(−3.0,7.1) \| .425 \| \| 2.9(−2.3,8.1) \| .270 \| \| \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| \| model 1 \| Ref. \| 4.2(−4.8,13.1) \| .358 \| −3.0(−8.1,2.0) \| .236 \| \| 6.7(1.2,12.2) \| .018 \| * \| 2.3(−3.3,8.0) \| .418 \| \| \| model 2 \| Ref. \| 4.4(−4.5,13.3) \| .331 \| −3.0(−7.9,2.0) \| .234 \| \| 6.3(0.9,11.8) \| .023 \| * \| 1.8(−3.8,7.5) \| .520 \| \| \| model 3 \| Ref. \| 4.2(−4.7,13.1) \| .354 \| −2.9(−7.9,2.0) \| .242 \| \| 6.3(0.8,11.7) \| .025 \| * \| 1.9(−3.8,7.5) \| .512 \| \| \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| \| model 1 \| Ref. \| 2.2(−7.1,11.5) \| .643 \| 1.8(−7.1,3.4) \| .488 \| \| 5.9(0.2,11.6) \| .044 \| * \| 6.8(0.9,12.7) \| .025 \| * \| \| model 2 \| Ref. \| 2.2(−7.4,11.8) \| .652 \| −1.6(−6.9,3.8) \| .564 \| \| 5.8(−0.1,11.7) \| .054 \| \| 7.2(1.1,13.2) \| .021 \| * \| \| model 3 \| Ref. \| 1.7(−7.8,11.1) \| .725 \| −1.4(−6.7,3.8) \| .590 \| \| 5.6(−0.2,11.3) \| .060 \| \| 7.3(1.3,13.2) \| .018 \| * \| \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| \| model 1 \| Ref. \| 4.3(−5.1,13.7) \| .368 \| −3.9(−9.2,1.4) \| .144 \| \| 4.5(−1.2,10.3) \| .122 \| \| 8.8(2.8,14.7) \| .004 \| † \| \| model 2 \| Ref. \| 3.8(−5.9,13.6) \| .438 \| −4.0(−9.4,1.4) \| .147 \| \| 4.2(−1.7,10.2) \| .162 \| \| 9.1(3.0,15.3) \| .004 \| † \| \| model 3 \| Ref. \| 3.6(−6.1,13.3) \| .467 \| −3.9(−9.3,1.5) \| .153 \| \| 4.1(−1.8,10.1) \| .172 \| \| 9.2(3.0,15.3) \| .004 \| † \| \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| \| model 1 \| Ref. \| 3.9(−5.4,13.2) \| .403 \| −6.1(−11.3,−0.8) \| .023 \| * \| −6.0(−11.7,−0.2) \| .041 \| * \| −5.2(−11.1,0.7) \| .083 \| \| \| model 2 \| Ref. \| 4.1(−5.5,13.8) \| .402 \| −6.1(−11.4,−0.7) \| .027 \| * \| −5.7(−11.6,0.2) \| .058 \| \| −5.4(−11.5,0.7) \| .080 \| \| \| model 3 \| Ref. \| 3.8(−5.8,13.5) \| .432 \| −6.0(−11.3,−0.6) \| .028 \| * \| −5.8(−11.7,0.1) \| .053 \| \| −5.4(−11.5,0.7) \| .082 \| \| \| \| Never \| Former \| \| Current \| \| \| \| \| \| \| \| \| \| \| \| β (95%CI) \| P-value \| β (95%CI) \| P-value \| P-value \| \| \| \| \| \| \| \| Females \| Females \| Females \| Females \| Females \| Females \| Females \| \| \| \| \| \| \| \| No. of participants, n \| 166 \| 13 \| \| 94 \| \| \| \| \| \| \| \| \| \| HbA1c ≤ 5.5% \| HbA1c ≤ 5.5% \| HbA1c ≤ 5.5% \| HbA1c ≤ 5.5% \| HbA1c ≤ 5.5% \| HbA1c ≤ 5.5% \| HbA1c ≤ 5.5% \| \| \| \| \| \| \| \| all-day \| all-day \| all-day \| all-day \| all-day \| all-day \| all-day \| \| \| \| \| \| \| \| model 1 \| Ref. \| 2.4(−2.6,7.3) \| .352 \| 0.3(−1.8,2.5) \| .771 \| \| \| \| \| \| \| \| \| model 2 \| Ref. \| 2.5(−2.6,7.6) \| .333 \| 0.2(−2.0,2.5) \| .841 \| \| \| \| \| \| \| \| \| model 3 \| Ref. \| 2.2(−2.9,7.3) \| .391 \| 0.4(−1.8,2.6) \| .708 \| \| \| \| \| \| \| \| \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| \| \| \| \| \| \| \| model 1 \| Ref. \| 0.5(−5.4,6.4) \| .871 \| 1.8(−0.7,4.4) \| .156 \| \| \| \| \| \| \| \| \| model 2 \| Ref. \| 0.3(−5.6,6.2) \| .922 \| 1.4(−1.2,4.0) \| .278 \| \| \| \| \| \| \| \| \| model 3 \| Ref. \| 0.1(−5.8,6.0) \| .974 \| 1.6(−1.0,4.2) \| .239 \| \| \| \| \| \| \| \| \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| \| \| \| \| \| \| \| model 1 \| Ref. \| 1.7(−3.9,7.3) \| .559 \| 1.3(−1.2,3.7) \| .310 \| \| \| \| \| \| \| \| \| model 2 \| Ref. \| 1.7(−4.1,7.4) \| .565 \| 1.4(−1.1,3.9) \| .259 \| \| \| \| \| \| \| \| \| model 3 \| Ref. \| 1.5(−4.3,7.2) \| .616 \| 1.6(−0.9,4.1) \| .216 \| \| \| \| \| \| \| \| \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| \| \| \| \| \| \| \| model 1 \| Ref. \| 4.0(−2.2,10.3) \| .208 \| 2.0(−0.8,4.7) \| .158 \| \| \| \| \| \| \| \| \| model 2 \| Ref. \| 4.4(−2.0,10.8) \| .176 \| 2.1(−0.7,4.9) \| .142 \| \| \| \| \| \| \| \| \| model 3 \| Ref. \| 4.1(−2.3,10.4) \| .206 \| 2.3(−0.5,5.0) \| .108 \| \| \| \| \| \| \| \| \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| \| \| \| \| \| \| \| model 1 \| Ref. \| 2.9(−2.8,8.6) \| .320 \| −3.0(−5.5,−0.5) \| .019 \| * \| \| \| \| \| \| \| \| model 2 \| Ref. \| 3.2(−2.6,9.0) \| .282 \| −3.3(−5.8,−0.7) \| .013 \| * \| \| \| \| \| \| \| \| model 3 \| Ref. \| 2.8(−3.0,8.5) \| .348 \| −3.0(−5.5,−0.4) \| .021 \| * \| \| \| \| \| \| \| \| HbA1c 5.6%–5.9% \| HbA1c 5.6%–5.9% \| HbA1c 5.6%–5.9% \| HbA1c 5.6%–5.9% \| HbA1c 5.6%–5.9% \| HbA1c 5.6%–5.9% \| HbA1c 5.6%–5.9% \| \| \| \| \| \| \| \| No. of participants, n \| 233 \| 29 \| \| 101 \| \| \| \| \| \| \| \| \| \| all-day \| all-day \| all-day \| all-day \| all-day \| all-day \| all-day \| \| \| \| \| \| \| \| model 1 \| Ref. \| 0.2(−3.8,4.2) \| .917 \| −0.2(−2.6,2.2) \| .896 \| \| \| \| \| \| \| \| \| model 2 \| Ref. \| 0.0(−4.2,4.2) \| .999 \| −0.1(−2.5,2.3) \| .931 \| \| \| \| \| \| \| \| \| model 3 \| Ref. \| −0.2(−4.4,4.0) \| .925 \| −0.2(−2.6,2.2) \| .867 \| \| \| \| \| \| \| \| \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| 0:00 to 5:00 h \| \| \| \| \| \| \| \| model 1 \| Ref. \| 0.7(−4.0,5.4) \| .782 \| −0.5(−3.3,2.3) \| .735 \| \| \| \| \| \| \| \| \| model 2 \| Ref. \| −0.4(−5.0,4.2) \| .868 \| −0.9(−3.6,1.8) \| .513 \| \| \| \| \| \| \| \| \| model 3 \| Ref. \| −0.5(−5.1,4.1) \| .821 \| −1.0(−3.6,1.7) \| .480 \| \| \| \| \| \| \| \| \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| 5:00 to 11:00 h \| \| \| \| \| \| \| \| model 1 \| Ref. \| 0.4(−4.0,4.8) \| .858 \| 2.0(−0.6,4.6) \| .136 \| \| \| \| \| \| \| \| \| model 2 \| Ref. \| −0.4(−4.9,4.1) \| .866 \| 1.9(−0.8,4.5) \| .164 \| \| \| \| \| \| \| \| \| model 3 \| Ref. \| −0.5(−5.1,4.0) \| .812 \| 1.8(−0.8,4.4) \| .183 \| \| \| \| \| \| \| \| \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| 11:00 to 17:00 h \| \| \| \| \| \| \| \| model 1 \| Ref. \| 0.0(−4.9,4.9) \| .998 \| 0.6(−2.3,3.6) \| .664 \| \| \| \| \| \| \| \| \| model 2 \| Ref. \| 0.2(−4.9,5.3) \| .939 \| 0.8(−2.2,3.8) \| .600 \| \| \| \| \| \| \| \| \| model 3 \| Ref. \| −0.1(−5.2,5.1) \| .984 \| 0.7(−2.3,3.7) \| .659 \| \| \| \| \| \| \| \| \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| 17:00 to 24:00 h \| \| \| \| \| \| \| \| model 1 \| Ref. \| −0.1(−4.7,4.5) \| .974 \| −2.4(−5.2,0.3) \| .081 \| \| \| \| \| \| \| \| \| model 2 \| Ref. \| 0.4(−4.3,5.2) \| .857 \| −2.0(−4.8,0.8) \| .155 \| \| \| \| \| \| \| \| \| model 3 \| Ref. \| 0.2(−4.5,4.9) \| .930 \| −2.1(−4.9,0.6) \| .132 \| \| \| \| \| \| \| \| ## Discussion This is the first study to show an association between habitual daily alcohol consumption and time-specific glucose levels in daily life. Among males, the time-specific average glucose levels between 5:00 and 11:00 h and between 11:00 and 17:00 h were higher in moderate and heavy drinkers than in never drinkers. The time-specific average glucose levels between 17:00 and 24:00 h were lower in male moderate and heavy drinkers than in never drinkers and female current drinkers. Possible mechanisms for the lowering of time-specific average glucose levels between 17:00 and 24:00 h among current drinkers could be the suppression of gluconeogenesis, caused by a decrease in the ratio of NAD to NADH [14] and the suppression of growth hormone secretion caused by the acute effect of alcohol consumption. The first mechanism can be explained as follows. Ethanol is oxidized by alcohol dehydrogenase to acetaldehyde and then metabolized to acetic acid by acetaldehyde dehydrogenase. NAD is consumed when ethanol is metabolized to acetaldehyde and acetic acid and thus reduced to NADH in the redox cycles. In an experimental study of five healthy adult males in fasting state, 48 g of ethanol consumption reduced gluconeogenesis by $45\%$ after 5 h compared to no ethanol ingestion [10]. Japanese people generally consume alcohol only at dinner or night; hence, it is speculated that the acute effects of alcohol intake are those that appear between 17:00 and 24:00 h. As the second mechanism, the acute suppression of growth hormone secretion by alcohol consumption resulted in increased insulin sensitivity and reduced blood glucose levels [15, 16]. Ingestion of 0.8 g/kg of alcohol reduces plasma growth hormone secretion at night by 70–$75\%$ of non-drinking baseline levels among healthy males aged 21–26 years [15]. A randomized double-blinded trial demonstrated that consuming gin alone or both gin and tonic, but not tonic alone, results in a marked reduction in plasma growth hormone levels and hypoglycemia within 5 h of drinking [17]. Thus, the abovementioned mechanisms may explain the lower average glucose levels between 17:00 and 24:00 h in current drinkers than in never drinkers in our study. Contrastingly, the average glucose levels between 5:00 and 11:00 h and between 11:00 and 17:00 h were higher in male moderate and heavy drinkers than in non-drinkers. This can be explained by different physiological processes during the 17:00 to 24:00 post-drinking interval, i.e., increased glucocorticoid levels and sympathetic excitation. Alcohol activates the hypothalamic-pituitary-adrenal axis, resulting in a dose-dependent increase in adrenocorticotropic hormone and glucocorticoid levels [18, 19]. In addition, acetaldehyde, a metabolite of alcohol, acts on the adrenal medulla and sympathetic ganglia to release catecholamines, resulting in sympathetic nerve excitation [20]. In a previous study of 539 Japanese males aged 35–65 years, heavy drinkers (≥46 g ethanol/day) had higher salivary cortisol concentrations and a higher prevalence of blood pressure surge in the morning and higher sympathetic nervous activity and an increased heart rate during both daytime and sleep, than non-drinkers [21]. Since the amount of alcohol metabolized per hour is approximately 100 mg/kg of body weight [22], these hormonal and sympathetic nervous effects can last for 8 h after heavy drinking. Indeed, the long-lasting effects of heavy drinking may become more apparent after the aforementioned acute effects of heavy drinking are attenuated. The higher average glucose levels between 5:00 and 11:00 h and between 11:00 and 17:00 h observed in male moderate and heavy drinkers were probably due to the alcohol-induced increase in glucocorticoid levels and sympathetic nervous system activity. Furthermore, suppressed glucose levels between 17:00 and 24:00 h were more pronounced in the individual with HbA1c ≤ $5.5\%$ (≤37 mmol/mol), while increased glucose levels between 5:00 and 11:00 h, between 11:00 and 17:00 h were more pronounced in those with HbA1c of 5.6–$5.9\%$ (38–41 mmol/mol). The reason may be due to differences in the basal insulin secretory capacity of the pancreas and the insulin resistance of skeletal muscle and liver. ## Strengths and limitations Our study has several strengths. First, the large sample size enabled us to conduct stratified analyses based on sex and HbA1c levels. Second, our study population was community-dwelling, which ensured the generalizability of our findings. Third, we monitored the daily variation of glucose levels for many days consecutively, which reduced the impact of inter-day variations in glucose levels on the observed association. The limitations of this study should also be discussed. First, our investigation was cross-sectional; hence, we could not determine the causality between drinking behavior and glucose levels or whether glucose levels affect drinking behavior. However, because our subjects were non-diabetic persons, it is unlikely that glucose levels could affect drinking behavior. Second, we could not examine the association between moderate and heavy drinking in females because of the limited sample size. Third, we did not have the data on alcohol beverage types and time of alcohol consumption, which warrant future studies. Forth, participants in our study may be more health conscious than non-participants, and thus selection bias may exist. Therefore, our results were likely to be generalizable, but not representative of the general population. ## Conclusions Alcohol consumption was associated with glucose levels in a time-dependent biphasic pattern. It remained uncertain how a time-dependent biphasic patterns of glucose levels by alcohol consumption lead to the development of diabetes mellitus, which will be examined in the future. Circadian changes in glucose levels among nondiabetic persons would provide useful information for the modification of lifestyle in the prevention of diabetes mellitus. ## Ethics approval and consent to participate The study protocol was approved by the ethics committees of Osaka University, the University of Tsukuba, and the Osaka Center for Cancer and Cardiovascular Disease Prevention. Written and oral explanations were provided to the participants and written informed consent was obtained. ## Consent for publication Not applicable. ## Availability of data and material Data cannot be shared for privacy or ethical reasons. ## Competing interests All authors declare no conflict of interest. ## Funding This study was supported by the JSPS (Japan Society for the Promotion of Science) KAKENHI (grant number A18H03050), the Japan Small- and Medium-Sized Enter-prise Welfare Foundation (FULLHAP), and a Seijinbyo igaku kenkyu jyosei A (Japanese) from Osaka foundation for the prevention of cancer and cardiovascular disease. ## Authors’ contributions M.I.: Designed the study, wrote original draft preparation, initial statistical analysis, investigation, and data curation H.Imano: Reviewed, editing, revised statistical analysis, revised the manuscript, conceptualization, and supervised. I.M.: Revised the manuscript and investigation. K.Y.: Revised the manuscript and investigation. K.M.: Data curation and software. M.T.H.: Conceptualization. M.Tanaka: Data curation, software, and investigation. M.Y., T.K., and M.Takada: Revised the manuscript and investigation. M.K., T.O., and Y.S.: Investigation. T.S.: Revised the manuscript and supervised. H.Iso: Reviewed, editing, revised statistical analysis, revised the manuscript, conceptualization, and supervised. All authors read and approved the final manuscript. M.I. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. ## Acknowledgments The authors thank the study physicians, clinical laboratory technologists, public health nurses, nutritionists, nurses, engineers, clerks, and officers of the CIRCS collaborating research institutes and the affiliated institutions in Ikawa, the Minami-Takayasu district of Yao, and Kyowa for their collaboration. We thank Kana Okamoto, Haruna Kawachi, and graduate students in the Department of Public Health, Osaka University for their assistance in collecting the data, and Jia-Yi Dong helping for in submitting the paper. ## References 1. 1.International Diabetes Federation. IDF Diabetes Atlas. 10th ed; 2021. 2. Lee WL, Cheung AM, Cape D, Zinman B. Impact of diabetes on coronary artery disease in women and men: a meta-analysis of prospective studies. Diabetes Care (2000.0) 23 962-8. PMID: 10895847 3. Peters SA, Huxley RR, Woodward M. Diabetes as a risk factor for stroke in women compared with men: A systematic review and meta-analysis of 64 cohorts, including 775,385 individuals and 12,539 strokes. Lancet (2014.0) 383 1973-80. PMID: 24613026 4. Sarwar N, Gao P, Kondapally Seshasai SR, Gobin R, Kaptoge S, Di Angelantonio E, Ingelsson E, Lawlor DA, Selvin E, Stampfer M. Diabetes mellitus, fasting blood glucose concentration, and risk of vascular disease: A collaborative meta-analysis of 102 prospective studies. Lancet (2010.0) 375 2215-22. PMID: 20609967 5. Worldwide trends in diabetes since 1980: a pooled analysis of 751 population-based studies with 4.4 million participants. Lancet (2016.0) 387 1513-30. PMID: 27061677 6. Wild S, Roglic G, Green A, Sicree R, King H. Global prevalence of diabetes: estimates for the year 2000 and projections for 2030. Diabetes Care (2004.0) 27 1047-53. PMID: 15111519 7. Koppes LL, Dekker JM, Hendriks HF, Bouter LM, Heine RJ. Moderate alcohol consumption lowers the risk of type 2 diabetes: a meta-analysis of prospective observational studies. Diabetes Care (2005.0) 28 719-25. PMID: 15735217 8. De Vegt F, Dekker JM, Groeneveld WJA, Nijpels G, Stehouwer CDA, Bouter LM, Heine RJ. Moderate alcohol consumption is associated with lower risk for incident diabetes and mortality: the Hoorn Study. Diabetes Res Clin Pract (2002.0) 57 53-60. PMID: 12007730 9. Conigrave KM, Hu BF, Camargo CA, Stampfer MJ, Willett WC, Rimm EB. A prospective study of drinking patterns in relation to risk of type 2 diabetes among men. Diabetes (2001.0) 50 2390-5. PMID: 11574424 10. Siler SQ, Neese RA, Christiansen MP, Hellerstein MK. The inhibition of gluconeogenesis following alcohol in humans. Am J Physiol (1998.0) 275 E897-907. PMID: 9815011 11. Davies MJ, Baer DJ, Judd JT, Brown ED, Campbell WS, Taylor PR. Effects of moderate alcohol intake on fasting insulin and glucose concentrations and insulin sensitivity in postmenopausal women: a randomized controlled trial. JAMA (2002.0) 287 2559-62. PMID: 12020337 12. Yamagishi K, Muraki I, Kubota Y, Hayama-Terada M, Imano H, Cui R, Umesawa M, Shimizu Y, Sankai T, Okada T. The circulatory risk in communities study (CIRCS): a long-term epidemiological study for lifestyle-related disease among Japanese men and women living in communities. J Epidemiol (2019.0) 29 83-91. PMID: 30584233 13. Bailey T, Bode BW, Christiansen MP, Klaff LJ, Alva S. The performance and usability of a factory-calibrated flash glucose monitoring system. Diabetes Technol Ther (2015.0) 17 787-94. PMID: 26171659 14. Krebs HA, Freedland RA, Hems R, Stubbs M. Inhibition of hepatic gluconeogenesis by ethanol. Biochem J (1969.0) 112 117-24. PMID: 5774487 15. Prinz PN, Roehrs TA, Vitaliano PP, Linnoila M, Weitzman ED. Effect of alcohol on sleep and nighttime plasma growth hormone and cortisol concentrations. J Clin Endocrinol Metab (1980.0) 51 759-64. PMID: 7419664 16. Emanuele MA, Kirsteins L, Reda D, Emanuele NV, Lawrence AM. The effect of in vitro ethanol exposure on basal growth hormone secretion. Endocr Res 14 283-91. PMID: 3250867 17. Flanagan D, Wood P, Sherwin R, Debrah K, Kerr D. Gin and tonic and reactive hypoglycemia: what is important-the gin, the tonic, or both?. J Clin Endocrinol Metab (1998.0) 83 796-800. PMID: 9506730 18. Thayer JF, Hall M, Sollers JJ, Fischer JE. Alcohol use, urinary cortisol, and heart rate variability in apparently healthy men: evidence for impaired inhibitory control of the HPA axis in heavy drinkers. Int J Psychophysiol (2006.0) 59 244-50. PMID: 16325293 19. Rachdaoui N, Sarkar DK. Pathophysiology of the effects of alcohol abuse on the endocrine system. Alcohol Res (2017.0) 38 255-76. PMID: 28988577 20. Eriksson CJP. The role of acetaldehyde in the actions of alcohol (update 2000). Alcohol Clin Exp Res (2001.0) 25 15S-32S. PMID: 11391045 21. Ohira T, Tanigawa T, Tabata M, Imano H, Kitamura A, Kiyama M, Sato S, Okamura T, Cui R, Koike KA. Effects of habitual alcohol intake on ambulatory blood pressure, heart rate, and its variability among Japanese men. Hypertension (2009.0) 53 13-9. PMID: 19029490 22. Mizoi Y, Kogame M, Fukunaga T, Ueno Y, Adachi J, Fujiwara S. Polymorphism of aldehyde dehydrogenase and ethanol elimination. Alcohol (1985.0) 2 393-6. PMID: 4026956
--- title: 'Association between meat, fish, and fatty acid intake and incidence of acute myeloid leukemia and myelodysplastic syndrome: the Japan Public Health Center-based Prospective Study' authors: - Yoshimitsu Shimomura - Tomotaka Sobue - Ling Zha - Tetsuhisa Kitamura - Motoki Iwasaki - Manami Inoue - Taiki Yamaji - Shoichiro Tsugane - Norie Sawada journal: Environmental Health and Preventive Medicine year: 2023 pmcid: PMC10025862 doi: 10.1265/ehpm.22-00233 license: CC BY 4.0 --- # Association between meat, fish, and fatty acid intake and incidence of acute myeloid leukemia and myelodysplastic syndrome: the Japan Public Health Center-based Prospective Study ## Abstract ### Background The association between meat, fish, or fatty acid intake and acute myeloid leukemia (AML) and myelodysplastic syndrome (MDS) has been investigated in a few studies, and the results were inconsistent. In addition, most studies are mainly based on the United States and European countries, in which the dietary patterns differ from that in Asia. Therefore, the risk of AML/MDS from meat, fish, or fatty acid intake in Asia requires further exploration. The aim of this study was to investigate the association between AML/MDS incidence and meat, fish, or fatty acid intake using the Japan Public Health Center–based prospective study. ### Methods The present study included 93,366 participants who were eligible for analysis and followed up from the 5-year survey date until December 2012. We estimated the impact of their intake on AML/MDS incidence using a Cox proportional hazards model. ### Results The study participants were followed up for 1,345,002 person-years. During the follow-up period, we identified 67 AML and 49 MDS cases. An increased intake of processed red meat was significantly associated with the incidence of AML/MDS, with a hazard ratio of 1.63 ($95\%$ confidence interval, 1.03–2.57) for the highest versus lowest tertile and a Ptrend of 0.04. Meanwhile, the intake of other foods and fatty acids was not associated with AML/MDS. ### Conclusion In this Japanese population, processed red meat was associated with an increased incidence of AML/MDS. ### Supplementary information The online version contains supplementary material available at https://doi.org/10.1265/ehpm.22-00233. ## Background Acute myeloid leukemia (AML) represent a genetically heterogeneous group of myeloid neoplasms that primarily affect older adults [1]. Myelodysplastic syndromes (MDS) shares a part of clinical, pathological feature with AML, and is distinguished from AML by lower percentage of blast [2]. An accurate understanding of the etiology of both diseases is essential for their prevention because they have high morbidity and mortality [1, 2]. Although the etiologies of AML/MDS are poorly understood, exposure to chemicals such as benzene, dioxin, pesticides, and herbicides is associated with the development of AML or MDS [1, 2]. Benzene is the most consistently-identified leukemogenic chemical and implicated in the development of AML and MDS via benzene-related oxidative stress and aryl hydrocarbon receptor dysregulation [3, 4]. Some studies revealed the possible association between dioxin exposure and AML development [5], pesticide exposure and AML [6] or MDS [7] development, and herbicides and AML [6] or MDS [8] development. Among the potential causes of diseases, food intakes may be related to the incidence of AML/MDS. For example, red meat and processed red meat intakes can increase the incidence of AML/MDS similar to other cancers due to carcinogens, such as polycyclic aromatic hydrocarbons and N-nitroso compounds [9–11]. Similarly, fish intakes, which is generally associated with a reduced risk of cancer, can increase the incidence of AML/MDS [12] due to its components, including sea pollution such as heavy metals [13] and dioxin [14, 15]. In addition, fatty acids, an important component of meat and fish, are associated with other hematological malignancies and are possibly associated with AML/MDS risk [16–19]. However, most epidemiological studies regarding association between meat and fish intakes and AML/MDS incidence are case-control studies, which induce publication bias and recall bias, there are a few good cohort studies with less bias, and the reported results were inconsistent [18, 20–29]. In addition, most studies were performed in the United States and European countries, in which the dietary patterns differed from that in Asia [12, 18, 20–29]. Therefore, the risk of AML/MDS from meat, fish, or fatty acid intake in Asia requires further exploration. Thus, this study aimed to investigate the association between meat, fish, or fatty acid intake and the incidence of AML/MDS using Japan public health center (JPHC)–based prospective study data. ## Study population The present study included 93,366 participants. All participants were registered in the JPHC. The details of the JPHC study are described elsewhere [30]. Briefly, the JPHC study comprised cohorts I and II, which started in 1990 and 1993, respectively. The cohort included 11 public health centers with a total of 140,420 inhabitants across Japan. This study excluded participants from Katsushika, Tokyo, who had no cancer incidence data ($$n = 7$$,097), who were not Japanese ($$n = 52$$), who emigrated before the start of the study ($$n = 188$$), whose recorded birth date was incorrect ($$n = 7$$), for whom registration data were duplicated ($$n = 12$$), who emigrated the original area before the 5-year survey ($$n = 5$$,060), who died before the 5-year survey ($$n = 6$$,598), and who were lost to follow-up before the 5-year survey ($$n = 73$$). Among remaining 121,333 participants, 98,616 participants returned the 5-year follow-up questionnaire, which yielded a response rate of $81.3\%$. We excluded the participants who did not participate in the 5-year follow-up questionnaire used as an exposure assessment because we used a more inclusive food frequency questionnaire (FFQ) in the 5-year follow-up questionnaire. Furthermore, we excluded participants who had a previous cancer history ($$n = 2$$,943, $3.0\%$), were lost to follow-up ($$n = 214$$, $0.2\%$), or reported an intake of over 5000 kcal/day or less than 500 kcal/day ($$n = 2$$,093, $2.1\%$). The study was approved by the ethics committees of the National Cancer Center (Approval number: 2015-085, date: November 25, 2021) and Osaka University (Approval number: 14020-10, date: February 10, 2021). ## Exposure assessment Daily food consumption (g/day) was calculated by multiplying the daily consumption frequency by the selected portion size using a FFQ obtained at the 5-year survey. The FFQ included 138 food and beverage items with standard portions/units (small, medium, and large portions) and nine frequency categories (almost never, 1–3 times/month, 1–2 times/week, 3–4 times/week, 5–6 times/week, 1 time/day, 2–3 times/day, 4–6 times/day, and ≥7 times/day). The average consumption of 16 meat items and 19 fish and shellfish items during the previous year was calculated from the standard portion and frequency categories obtained from the FFQ. We categorized these food items into the following groups: total meat including all meat items; red meat including all 10 unprocessed red meat items and all 4 processed red meat items; unprocessed red meat including steak, grilled beef, and stewed beef, stir-fried pork, deep-fried pork, Western-style stewed pork, Japanese-style stewed pork, pork in soup, and pork liver, and chicken liver, processed red meat including ham, sausage or Western-style sausage, bacon, and luncheon meat, poultry including grilled chicken and deep-fried chicken, fish including salted fish, dried fish, canned tuna, salmon or trout, bonito or tuna, codfish or flatfish, sea bream, horse mackerel or sardines, mackerel pike or mackerel, dried small fish, salted roe, eel, squid, octopus, prawns, short-necked clam, viviparidae, chikuwa, and kamaboko, big fish including salmon, skipjack/tuna, cod/flatfish, and sea bream, and small fish including horse mackerel/sardine, saury/mackerel, and eel [31]. In addition to meat and fish intake, the dietary intake of total fatty acid, saturated fatty acid, monounsaturated fatty acid, polyunsaturated fatty acid (PUFA), n-3 PUFA, and n-6 PUFA were calculated according to the fatty acid composition table developed by the substitute method based on fatty acid composition tables for Japanese foods [32]. The intake of meat, fish, and fatty acids was adjusted for total energy intake by the residual model [33]. The validity of the FFQ was evaluated using energy-adjusted Spearman’s correlation coefficients between the intake of interest derived from the FFQ and those derived from the 28-day or 14-day dietary records [34, 35]. The energy-adjusted Spearman’s correlation coefficients of meat were 0.50 for men and 0.47 for women, those of fish were 0.32 for men and 0.32 for women, and those of fatty acid were 0.52 for men and 0.47 for women [34, 35]. The reproducibility of the FFQ was evaluated by energy-adjusted Spearman’s correlation coefficients for the intake of interest derived from the two FFQ administered 1 year apart [36, 37]. The energy-adjusted Spearman’s correlation coefficients of meat were 0.52 for men and 0.52 for women, that of fish were 0.44 for men and 0.34 for women, and that of fatty acid were 0.47 for men and 0.52 for women [36, 37]. ## Case identification Newly diagnosed AML/MDS cases were determined from medical reports of major local hospitals and data linkage with population-based cancer registries. Death certificate information was used as a supplementary information source. Data linked between major local hospitals and population-based cancer registries were used to confirm newly diagnosed AML/MDS. The cases were coded according to the hospital-based cancer registries in Japan using the International Classification of Diseases for Oncology, Third Edition. The histologic codes were 9840, 9860, 9861, 9866, 9867, 9873, 9874, 9875, 9891, 9895, and 9896 for AML and 9980, 9982, 9983, 9985, and 9987–9989 for MDS. If two or more AML/MDS or other cancers were diagnosed in one participant, the first diagnosis was used in the analysis. ## Endpoints and statistical analysis The person-years of follow-up were counted for individual participants from the 5-year follow-up survey date to the end of follow-up, which was defined as move outside the study area, loss to follow-up, withdrawal from the study, death, diagnosis of AML/MDS, or last date of the follow-up period (December 31, 2012). Continuous variables are summarized as median and interquartile range (quartiles 1–3), while categorical variables are summarized as count and percentage. The study participants were subdivided into tertiles with respect to their energy-adjusted intakes of interest, including total meat, red meat, processed red meat, unprocessed meat, poultry, fish, total fatty acids, saturated fatty acids, monounsaturated fatty acids, PUFA, n-3 PUFAs, and n-6 PUFA. We estimated the impact of exposure on AML/MDS incidence using a multivariable Cox proportional hazard model. We described the hazard ratios (HRs), $95\%$ confidence intervals (CIs), and Ptrend values to test for a linear trend across tertiles as a rank variable. Adjusted covariates were associated with the incidence of AML/MDS. Information on covariates was collected from a 5-year follow-up questionnaire survey. Body mass index (BMI) was calculated as body weight (kg) divided by squared height (m2) and categorized as <23, 23–<25, 25–<27, and ≥27 kg/m2. We assessed and categorized smoking history into never smokers, past smokers, current smokers, and unknown. Alcohol consumption was assessed as the weekly consumption obtained by multiplying the weekly frequency and categorized as never, rarely, 1–3 times/month, 1–2 times/week, 3–4 times/week, >4 times/week, and unknown. Physical activity is expressed as metabolic equivalents/day and categorized as quartiles and unknown. In model 1, adjusted covariates were age (continuous), sex, and study area (10 public health center areas). In model 2 (primary result), adjusted covariates were age (continuous), sex, study area (10 public health center area), BMI, history of smoking, alcohol consumption frequency, and physical activity as metabolic equivalents/day. Covariates were selected based on previous studies [38, 39]. In model 3, the sensitivity analyses excluded participants diagnosed with AML/MDS in the first two years to remove the potential bias of having AML/MDS at the 5-year survey which is the start of follow-up (9 cases diagnosed before 5-year survey were excluded in the analysis). Stratified analyses were conducted by estimating the interaction between intake of interest and age (≤median and >median, the median age was 57 years) and sex. Similarly, we estimated the impact of exposure on AML/MDS incidence using the same methods mentioned above. All P values presented were two-sided, and values <0.05 were considered statistically significant. All statistical analyses were performed using Stata version 14 (StataCorp LLC, the United States). ## Results We showed the baseline characteristics divided by the tertiles of total meat intake in Table 1. The median age was 57 (interquartile range, 51–63) years, and $46.5\%$ of the participants were men. A total of 93,366 participants were included and followed up for 1,345,002 person-years. During the follow-up period, we identified 67 AML and 49 MDS cases. Table 1 \| Unnamed: 0 \| Tertile of energy adjusted total meat intake \| Tertile of energy adjusted total meat intake.1 \| Tertile of energy adjusted total meat intake.2 \| \| --- \| --- \| --- \| --- \| \| \| Tertile 1 \| Tertile 2 \| Tertile 3 \| \| Agea, year \| 58 (52–64) \| 56 (50–63) \| 56 (49–62) \| \| Men, % \| 49.8 \| 47.4 \| 42.4 \| \| Body mass indexa, kg/m2 \| 23.2 (21.4–25.2) \| 23.3 (21.5–25.3) \| 23.5 (21.5–25.6) \| \| Current smoker, % \| 24.2 \| 24 \| 21.5 \| \| Regular drinker, % \| 40.5 \| 39.5 \| 32.1 \| \| Metabolic equivalentsa \| 31.9 (27.1–36.1) \| 31.9 (27.1–35.5) \| 31.9 (27.1–35.5) \| \| Intakes \| Intakes \| Intakes \| Intakes \| \| Food \| Food \| Food \| Food \| \| Total meata, g/day \| 22.8 (13.7–30.0) \| 49.8 (43.1–57.1) \| 89.0 (75.3–112.2) \| \| Red meata, g/day \| 18.1 (10.3–24.7) \| 41.7 (35.5–48.7) \| 76.4 (64.1–97.9) \| \| Processed red meata, g/day \| 1.5 (0.4–3.5) \| 4.2 (1.9–7.8) \| 7.2 (3.3–13.7) \| \| Unprocessed red meata, g/day \| 15.3 (8.3–21.6) \| 36.1 (29.8–43.1) \| 67.1 (54.9–87.2) \| \| Poultrya, g/day \| 3.3 (0.6–6.0) \| 7.3 (4.2–11.1) \| 11.0 (6.1–18.1) \| \| Fisha, g/day \| 68.9 (42.1–107) \| 79.0 (54.4–111) \| 79.6 (53.9–113) \| \| Big fisha, g/day \| 14.2 (6.5–26.6) \| 17.5 (10.1–29.1) \| 19.0 (10.8–30.8) \| \| Small fisha, g/day \| 14.2 (6.5–26.6) \| 17.5 (10.1–29.1) \| 19.0 (10.8–30.8) \| \| Nutrient \| Nutrient \| Nutrient \| Nutrient \| \| Total energya, kcal/day \| 1912 (1464–2479) \| 1914 (1540–2377) \| 1890 (1537–2320) \| \| Total FAa, g/day \| 37.1 (28.8–46.0) \| 45.3 (38.2–52.6) \| 56.6 (48.9–64.6) \| \| Saturated FAa, g/day \| 12.2 (8.8–16.1) \| 15.3 (12.3–18.6) \| 19.6 (16.4–23.0) \| \| MUFAa, g/day \| 13.8 (10.7–17.1) \| 17.7 (15.1–20.5) \| 23.1 (20.0–26.6) \| \| PUFAa, g/day \| 10.4 (8.1–12.9) \| 11.7 (9.8–13.8) \| 13.5 (11.5–15.6) \| \| n-3 PUFAa, g/day \| 2.1 (1.6–2.8) \| 2.3 (1.9–2.9) \| 2.5 (2.0–3.0) \| \| n-6 PUFAa, g/day \| 8.1 (6.4–10.1) \| 9.3 (7.8–11.0) \| 10.9 (9.3–12.6) \| The adjusted HR, $95\%$ CI, and Ptrend of the incidence of AML/MDS according to the intake of dietary factors of interest were shown in Table 2. An increased intake of processed red meat was significantly associated with the incidence of AML/MDS, with an HR of 1.63 ($95\%$ CI, 1.03–2.57) for the highest versus lowest tertile and Ptrend of 0.041. Meanwhile, red meat (HR [highest vs lowest]: 1.35, $95\%$CI: 0.85–2.14), unprocessed meat (HR: 1.08, $95\%$CI: 0.68–1.71), poultry (HR: 1.35, $95\%$CI: 0.85–2.13), fish (HR: 1.13, $95\%$CI: 0.71–1.79), big fish (HR: 1.03, $95\%$CI: 0.64–1.67) and small fish (HR: 1.32, $95\%$CI: 0.82–2.14) intake were not significantly associated with the incidence of AML/MDS. The intakes of total fatty acid (HR: 1.31, $95\%$CI: 0.81–1.57), saturated fatty acid (HR: 1.01, $95\%$CI: 0.62–1.63), monounsaturated fatty acid (HR: 1.03, $95\%$CI: 0.63–1.69), PUFA (HR: 1.42, $95\%$CI: 0.87–2.33), n-3 PUFA (HR: 1.16, $95\%$CI: 0.73–1.84), and n-6 PUFA (HR: 1.16, $95\%$CI: 0.69–1.93) were not significantly associated with the incidence of AML/MDS. Additionally, associations between the intake of interest and the incidence of AML/MDS did not show evidence of heterogeneity by age and sex, except for heterogeneity of small fish intake and sex (P for interaction: 0.048) (Supplemental Table 1). Table 2 \| Unnamed: 0 \| Tertile of energy-adjusted food intake \| Tertile of energy-adjusted food intake.1 \| Tertile of energy-adjusted food intake.2 \| Unnamed: 4 \| \| --- \| --- \| --- \| --- \| --- \| \| \| Tertile 1 \| Tertile 2 \| Tertile 3 \| Ptrendd \| \| Total meat \| Total meat \| Total meat \| Total meat \| Total meat \| \| Median intake, g/day \| 22.8 (<36.6) \| 49.8 (36.6–65.4) \| 89.0 (>65.4) \| \| \| Person-years, year \| 444386 \| 449493 \| 451123 \| \| \| Cases, n \| 41 \| 32 \| 43 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 0.89 (0.56–1.42) \| 1.33 (0.85–2.07) \| 0.224 \| \| HR (95%CI)b \| 1.00 (reference) \| 0.89 (0.56–1.42) \| 1.33 (0.85–2.08) \| 0.225 \| \| HR (95%CI)c \| 1.00 (reference) \| 1.02 (0.63–1.65) \| 1.40 (0.87–2.24) \| 0.169 \| \| Red meat \| Red meat \| Red meat \| Red meat \| Red meat \| \| Median intake, g/day \| 17.8 (<29.7) \| 41.3 (21.7–55.1) \| 76.4 (>55.1) \| \| \| Person-years, year \| 444781 \| 449404 \| 450816 \| \| \| Cases, n \| 39 \| 36 \| 41 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.07 (0.68–1.68) \| 1.35 (0.85–2.13) \| 0.209 \| \| HR (95%CI)b \| 1.00 (reference) \| 1.07 (0.68–1.69) \| 1.35 (0.85–2.14) \| 0.208 \| \| HR (95%CI)c \| 1.00 (reference) \| 1.17 (0.73–1.88) \| 1.3 (0.85–2.25) \| 0.187 \| \| Processed red meat \| Processed red meat \| Processed red meat \| Processed red meat \| Processed red meat \| \| Median intake, g/day \| 0.5 (<2.1) \| 3.7 (2.1–6.1) \| 11.1 (>6.1) \| \| \| Person-years, year \| 440133 \| 448860 \| 456008 \| \| \| Cases, n \| 40 \| 32 \| 44 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.00 (0.63–1.60) \| 1.61 (1.02–2.54) \| 0.044 \| \| HR (95%CI)b \| 1.00 (reference) \| 1.01 (0.63–1.61) \| 1.63 (1.03–2.57) \| 0.041 \| \| HR (95%CI)c \| 1.00 (reference) \| 1.09 (0.67–1.78) \| 1.67 (1.04–2.70) \| 0.038 \| \| Unprocessed red meat \| Unprocessed red meat \| Unprocessed red meat \| Unprocessed red meat \| Unprocessed red meat \| \| Median intake, g/day \| 14.8 (<25.2) \| 35.4 (25.2–47.7) \| 67.1 (>47.7) \| \| \| Person-years, year \| 445439 \| 449569 \| 449993 \| \| \| Cases, n \| 41 \| 38 \| 37 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.03 (0.66–1.61) \| 1.08 (0.68–1.71) \| 0.744 \| \| HR (95%CI)b \| 1.00 (reference) \| 1.04 (0.67–1.62) \| 1.08 (0.68–1.71) \| 0.747 \| \| HR (95%CI)c \| 1.00 (reference) \| 1.10 (0.69–1.74) \| 1.12 (0.69–1.81) \| 0.638 \| \| Poultry \| Poultry \| Poultry \| Poultry \| Poultry \| \| Median intake, g/day \| 1.2 (<4.2) \| 6.5 (4.2–9.5) \| 14.6 (>9.5) \| \| \| Person-years, year \| 442015 \| 450843 \| 452143 \| \| \| Cases, n \| 34 \| 40 \| 42 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.22 (0.77–1.93) \| 1.34 (0.85–2.12) \| 0.211 \| \| HR (95%CI)b \| 1.00 (reference) \| 1.22 (0.77–1.93) \| 1.35 (0.85–2.13) \| 0.207 \| \| HR (95%CI)c \| 1.00 (reference) \| 1.33 (0.82–2.17) \| 1.47 (0.90–2.38) \| 0.125 \| \| Fish \| Fish \| Fish \| Fish \| Fish \| \| Median intake, g/day \| 40.4 (<58.9) \| 76.2 (58.9–97.0) \| 127.8 (>97.0) \| \| \| Person-years, year \| 444689 \| 449074 \| 451238 \| \| \| Cases, n \| 37 \| 35 \| 44 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 0.97 (0.60–1.55) \| 1.13 (0.72–1.79) \| 0.584 \| \| HR (95%CI)b \| 1.00 (reference) \| 0.97 (0.61–1.56) \| 1.13 (0.71–1.79) \| 0.586 \| \| HR (95%CI)c \| 1.00 (reference) \| 0.99 (0.60–1.62) \| 1.19 (0.74–1.93) \| 0.454 \| \| Big Fish \| Big Fish \| Big Fish \| Big Fish \| Big Fish \| \| Median intake, g/day \| 6.3 (<11.6) \| 17.0 (11.6–24.2) \| 35.9 (>24.2) \| \| \| Person-years, year \| 435909 \| 451396 \| 457696 \| \| \| Cases, n \| 39 \| 38 \| 39 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.07 (0.67–1.73) \| 1.03 (0.64–1.67) \| 0.914 \| \| HR (95%CI)b \| 1.00 (reference) \| 1.09 (0.67–1.75) \| 1.03 (0.64–1.67) \| 0.917 \| \| HR (95%CI)c \| 1.00 (reference) \| 1.19 (0.71–1.97) \| 1.23 (0.74–2.05) \| 0.428 \| \| Small Fish \| Small Fish \| Small Fish \| Small Fish \| Small Fish \| \| Median intake, g/day \| 6.8 (<11.4) \| 16.3 (11.4–24.0) \| 37.4 (>24.0) \| \| \| Person-years, year \| 450314 \| 448042 \| 446644 \| \| \| Cases, n \| 32 \| 39 \| 45 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.27 (0.79–2.05) \| 1.32 (0.81–2.13) \| 0.277 \| \| HR (95%CI)b \| 1.00 (reference) \| 1.26 (0.78–2.04) \| 1.32 (0.82–2.14) \| 0.269 \| \| HR (95%CI)c \| 1.00 (reference) \| 1.34 (0.81–2.22) \| 1.37 (0.83–2.28) \| 0.454 \| \| Total FAs \| Total FAs \| Total FAs \| Total FAs \| Total FAs \| \| Median intake, g/day \| 33.0 (<40.6) \| 46.6 (40.6–52.7) \| 60.2 (>52.7) \| \| \| Person-years, year \| 444236 \| 450036 \| 450728 \| \| \| Cases, n \| 43 \| 35 \| 38 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 0.98 (0.62–1.54) \| 1.29 (0.81–2.06) \| 0.307 \| \| HR (95%CI)b \| 1.00 (reference) \| 1.00 (0.63–1.57) \| 1.31 (0.81–1.57) \| 0.289 \| \| HR (95%CI)c \| 1.00 (reference) \| 1.05 (0.65–1.68) \| 1.38 (0.83–2.28) \| 0.225 \| \| Saturated FAs \| Saturated FAs \| Saturated FAs \| Saturated FAs \| Saturated FAs \| \| Median intake, g/day \| 10.5 (<13.4) \| 13.4 (15.9–18.5) \| 22.0 (>18.5) \| \| \| Person-years, year \| 446481 \| 450463 \| 448057 \| \| \| Cases, n \| 47 \| 36 \| 33 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 0.91 (0.59–1.42) \| 1.01 (0.63–1.61) \| 0.978 \| \| HR (95%CI)b \| 1.00 (reference) \| 0.92 (0.59–1.43) \| 1.01 (0.62–1.63) \| 0.984 \| \| HR (95%CI)c \| 1.00 (reference) \| 0.88 (0.55–1.39) \| 0.93 (0.56–1.55) \| 0.746 \| \| MUFAs \| MUFAs \| MUFAs \| MUFAs \| MUFAs \| \| Median intake, g/day \| 12.6 (<15.7) \| 18.2 (15.7–20.7) \| 24.1 (>20.7) \| \| \| Person-years, year \| 444418 \| 449733 \| 450850 \| \| \| Cases, n \| 43 \| 42 \| 31 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.18 (0.76–1.81) \| 1.02 (0.63–1.67) \| 0.867 \| \| HR (95%CI)b \| 1.00 (reference) \| 1.18 (0.76–1.83) \| 1.03 (0.63–1.69) \| 0.858 \| \| HR (95%CI)c \| 1.00 (reference) \| 1.2 (0.78–1.93) \| 1.09 (0.65–1.83) \| 0.683 \| \| PUFAs \| PUFAs \| PUFAs \| PUFAs \| PUFAs \| \| Median intake, g/day \| 8.7 (<10.5) \| 12.0 (10.5–13.5) \| 15.4 (>13.5) \| \| \| Person-years, year \| 438631 \| 449622 \| 456748 \| \| \| Cases, n \| 34 \| 43 \| 39 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.42 (0.90–2.24) \| 1.39 (0.86–2.26) \| 0.176 \| \| HR (95%CI)b \| 1.00 (reference) \| 1.45 (0.92–2.29) \| 1.42 (0.87–2.33) \| 0.157 \| \| HR (95%CI)c \| 1.00 (reference) \| 1.51 (0.93–2.46) \| 1.63 (0.98–2.73) \| 0.060 \| \| n-3 PUFA \| n-3 PUFA \| n-3 PUFA \| n-3 PUFA \| n-3 PUFA \| \| Median intake, g/day \| 1.6 (<2.0) \| 2.3 (2.0–2.7) \| 3.2 (>2.7) \| \| \| Person-years, year \| 441029 \| 449351 \| 453621 \| \| \| Cases, n \| 37 \| 38 \| 41 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.10 (0.69–1.73) \| 1.15 (0.72–1.82) \| 0.562 \| \| HR (95%CI)b \| 1.00 (reference) \| 1.11 (0.70–1.75) \| 1.16 (0.73–1.84) \| 0.538 \| \| HR (95%CI)c \| 1.00 (reference) \| 1.15 (0.70–1.87) \| 1.35 (0.83–2.19) \| 0.223 \| \| n-6 PUFA \| n-6 PUFA \| n-6 PUFA \| n-6 PUFA \| n-6 PUFA \| \| Median intake, g/day \| 7.0 (<8.4) \| 9.5 (8.4–10.7) \| 12.3 (>10.7) \| \| \| Person-years, year \| 438156 \| 449894 \| 456951 \| \| \| Cases, n \| 36 \| 47 \| 33 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.49 (0.96–2.31) \| 1.14 (0.69–1.88) \| 0.562 \| \| HR (95%CI)b \| 1.00 (reference) \| 1.51 (0.97–2.36) \| 1.16 (0.69–1.93) \| 0.527 \| \| HR (95%CI)c \| 1.00 (reference) \| 1.58 (0.99–2.53) \| 1.32 (0.78–2.24) \| 0.277 \| The HRs, $95\%$ CIs, and Ptrend of AML were shown in Table 3. None of the dietary factors of interest was significantly associated with the incidence of AML (Table 3). Among them, total meat (HR: 1.54, $95\%$CI: 0.88–2.69), red meat (HR: 1.74, $95\%$CI: 0.97–3.12), processed red meat (HR: 1.58, $95\%$CI: 0.89–2.82), and total fatty acids (HR: 1.77, $95\%$CI: 0.93–3.36) tend to be associated with an increased risk of AML. Table 3 \| Unnamed: 0 \| Tertile of energy adjusted intake of interests \| Tertile of energy adjusted intake of interests.1 \| Tertile of energy adjusted intake of interests.2 \| Unnamed: 4 \| \| --- \| --- \| --- \| --- \| --- \| \| \| Tertile 1 \| Tertile 2 \| Tertile 3 \| Ptrendb \| \| Total meat \| Total meat \| Total meat \| Total meat \| Total meat \| \| Person year. year \| 444386 \| 449493 \| 451123 \| \| \| Case, n \| 24 \| 15 \| 28 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 0.71 (0.37–1.36) \| 1.54 (0.88–2.69) \| 0.141 \| \| Red meat \| Red meat \| Red meat \| Red meat \| Red meat \| \| Person year. year \| 444781 \| 449404 \| 450816 \| \| \| Case, n \| 21 \| 19 \| 27 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.05 (0.56–1.97) \| 1.74 (0.97–3.12) \| 0.067 \| \| Processed red meat \| Processed red meat \| Processed red meat \| Processed red meat \| Processed red meat \| \| Person year. year \| 440133 \| 448860 \| 456008 \| \| \| Case, n \| 24 \| 17 \| 26 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 0.87 (0.46–1.63) \| 1.58 (0.89–2.82) \| 0.130 \| \| Unprocessed red meat \| Unprocessed red meat \| Unprocessed red meat \| Unprocessed red meat \| Unprocessed red meat \| \| Person year. year \| 445439 \| 449569 \| 449993 \| \| \| Case, n \| 21 \| 23 \| 23 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.25 (0.69–2.27) \| 1.38 (0.76–2.53) \| 0.291 \| \| Poultry \| Poultry \| Poultry \| Poultry \| Poultry \| \| Person year. year \| 442015 \| 450843 \| 452143 \| \| \| Case, n \| 19 \| 24 \| 24 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.33 (0.72–2.44) \| 1.35 (0.73–2.50) \| 0.342 \| \| Fish \| Fish \| Fish \| Fish \| Fish \| \| Person year. year \| 444689 \| 449074 \| 451238 \| \| \| Case, n \| 18 \| 21 \| 28 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.11 (0.58–2.10) \| 1.33 (0.72–2.46) \| 0.349 \| \| Big Fish \| Big Fish \| Big Fish \| Big Fish \| Big Fish \| \| Person year. year \| 435909 \| 451394 \| 457696 \| \| \| Case, n \| 21 \| 23 \| 23 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.05 (0.56–1.95) \| 0.91 (0.48–1.69) \| 0.931 \| \| Small Fish \| Small Fish \| Small Fish \| Small Fish \| Small Fish \| \| Person year. year \| 450314 \| 448043 \| 446644 \| \| \| Case, n \| 17 \| 25 \| 25 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.47 (0.78–2.76) \| 1.37 (0.72–2.63) \| 0.367 \| \| Total FAs \| Total FAs \| Total FAs \| Total FAs \| Total FAs \| \| Person year. year \| 444236 \| 450036 \| 450728 \| \| \| Case, n \| 21 \| 22 \| 24 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.30 (0.70–2.41) \| 1.77 (0.93–3.36) \| 0.082 \| \| Saturated FAs \| Saturated FAs \| Saturated FAs \| Saturated FAs \| Saturated FAs \| \| Person year. year \| 446481 \| 450463 \| 448057 \| \| \| Case, n \| 23 \| 22 \| 22 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.18 (0.65–2.15) \| 1.48 (0.79–2.79) \| 0.224 \| \| MUFAs \| MUFAs \| MUFAs \| MUFAs \| MUFAs \| \| Person year. year \| 444418 \| 449733 \| 450850 \| \| \| Case, n \| 21 \| 27 \| 19 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.57 (0.87–2.82) \| 1.35 (0.69–2.61) \| 0.343 \| \| PUFAs \| PUFAs \| PUFAs \| PUFAs \| PUFAs \| \| Person year. year \| 438631 \| 449622 \| 456748 \| \| \| Case, n \| 20 \| 24 \| 23 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.35 (0.73–2.47) \| 1.35 (0.71–2.58) \| 0.363 \| \| n-3 PUFA \| n-3 PUFA \| n-3 PUFA \| n-3 PUFA \| n-3 PUFA \| \| Person year. year \| 441029 \| 449351 \| 453621 \| \| \| Case, n \| 20 \| 22 \| 25 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.13 (0.61–2.10) \| 1.20 (0.65–2.22) \| 0.569 \| \| n-6 PUFA \| n-6 PUFA \| n-6 PUFA \| n-6 PUFA \| n-6 PUFA \| \| Person year. year \| 438156 \| 449894 \| 456951 \| \| \| Case, n \| 20 \| 24 \| 23 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.37 (0.74–2.52) \| 1.42 (0.74–2.73) \| 0.287 \| The HR, $95\%$ CI, and Ptrend of MDS were shown in Table 4. None of the dietary factors of interest was significantly associated with the incidence of MDS (Table 4). In contrast to AML, only processed red meat intake tended to associate with the incidence of MDS (HR: 1.66, $95\%$CI: 0.80–3.47). Table 4 \| Unnamed: 0 \| Tertile of energy-adjusted intake of interest \| Tertile of energy-adjusted intake of interest.1 \| Tertile of energy-adjusted intake of interest.2 \| Unnamed: 4 \| \| --- \| --- \| --- \| --- \| --- \| \| \| Tertile 1 \| Tertile 2 \| Tertile 3 \| Ptrendb \| \| Total meat \| Total meat \| Total meat \| Total meat \| Total meat \| \| Person year. year \| 444386 \| 449492 \| 451123 \| \| \| Case, n \| 17 \| 17 \| 15 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.14 (0.58–2.24) \| 1.02 (0.49–2.14) \| 0.934 \| \| Red meat \| Red meat \| Red meat \| Red meat \| Red meat \| \| Person year. year \| 444781 \| 449404 \| 450816 \| \| \| Case, n \| 18 \| 17 \| 14 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.09 (0.58–2.12) \| 0.88 (0.41–1.86) \| 0.764 \| \| Processed red meat \| Processed red meat \| Processed red meat \| Processed red meat \| Processed red meat \| \| Person year. year \| 440133 \| 448860 \| 456008 \| \| \| Case, n \| 16 \| 15 \| 18 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.23 (0.60–2.51) \| 1.66 (0.80–3.47) \| 0.179 \| \| Unprocessed red meat \| Unprocessed red meat \| Unprocessed red meat \| Unprocessed red meat \| Unprocessed red meat \| \| Person year. year \| 445439 \| 449569 \| 449993 \| \| \| Case, n \| 20 \| 15 \| 14 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 0.81 (0.41–1.59) \| 0.75 (0.36–1.55) \| 0.422 \| \| Poultry \| Poultry \| Poultry \| Poultry \| Poultry \| \| Person year. year \| 442015 \| 450843 \| 452143 \| \| \| Case, n \| 15 \| 16 \| 18 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.09 (0.53–2.21) \| 1.33 (0.66–2.66) \| 0.422 \| \| Fish \| Fish \| Fish \| Fish \| Fish \| \| Person year. year \| 444689 \| 449074 \| 451238 \| \| \| Case, n \| 19 \| 14 \| 16 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 0.83 (0.41–1.69) \| 0.91 (0.45–1.84) \| 0.778 \| \| Big Fish \| Big Fish \| Big Fish \| Big Fish \| Big Fish \| \| Person year. year \| 435909 \| 451394 \| 457696 \| \| \| Case, n \| 18 \| 15 \| 16 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.16 (0.55–2.44) \| 1.25 (0.59–2.65) \| 0.569 \| \| Small Fish \| Small Fish \| Small Fish \| Small Fish \| Small Fish \| \| Person year. year \| 450314 \| 448043 \| 446644 \| \| \| Case, n \| 15 \| 14 \| 20 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.02 (0.48–2.16) \| 1.25 (0.60–2.59) \| 0.541 \| \| Total FAs \| Total FAs \| Total FAs \| Total FAs \| Total FAs \| \| Person year. year \| 444236 \| 450036 \| 450729 \| \| \| Case, n \| 22 \| 13 \| 14 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 0.71 (0.35–1.43) \| 0.87 (0.41–1.81) \| 0.631 \| \| Saturated FAs \| Saturated FAs \| Saturated FAs \| Saturated FAs \| Saturated FAs \| \| Person year. year \| 446481 \| 450463 \| 448057 \| \| \| Case, n \| 24 \| 14 \| 11 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 0.66 (0.34–1.30) \| 0.57 (0.26–1.23) \| 0.128 \| \| MUFAs \| MUFAs \| MUFAs \| MUFAs \| MUFAs \| \| Person year. year \| 444418 \| 4499733 \| 450850 \| \| \| Case, n \| 22 \| 15 \| 12 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 0.81 (0.42–1.60) \| 0.71 (0.33–1.52) \| 0.366 \| \| PUFAs \| PUFAs \| PUFAs \| PUFAs \| PUFAs \| \| Person year. year \| 438631 \| 449622 \| 456748 \| \| \| Case, n \| 14 \| 19 \| 16 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.59 (0.79–3.21) \| 1.49 (0.70–3.20) \| 0.294 \| \| n-3 PUFA \| n-3 PUFA \| n-3 PUFA \| n-3 PUFA \| n-3 PUFA \| \| Person year. year \| 441029 \| 449351 \| 454621 \| \| \| Case, n \| 17 \| 16 \| 16 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.07 (0.54–2.14) \| 1.10 (0.54–2.23) \| 0.796 \| \| n-6 PUFA \| n-6 PUFA \| n-6 PUFA \| n-6 PUFA \| n-6 PUFA \| \| Person year. year \| 438156 \| 449894 \| 456951 \| \| \| Case, n \| 16 \| 23 \| 10 \| \| \| HR (95%CI)a \| 1.00 (reference) \| 1.69 (0.88–3.25) \| 0.79 (0.34–1.84) \| 0.752 \| ## Discussion Here we investigated the association between meat, fish, or fatty acid intake and the incidence of AML/MDS. Our results showed that a higher processed red meat intake was associated with an increased incidence of AML/MDS. On the other hand, other intakes of interest had a null association with the incidence of AML/MDS. Since there is little evidence regarding the association between meat, fish, or fatty acid intake and the incidence of AML/MDS, our results provide additional information regarding the etiology of AML and MDS. Red meat and processed red meat are reportedly associated with an increased risk of several cancers through the development of carcinogens [11]. However, little epidemiological evidence is available regarding the association between meat intakes and the incidence of AML/MDS because of the limited number of studies, low incidence of AML/MDS, and differing definitions of leukemia, although leukemia subtypes have different etiologies. A cohort study from Europe revealed the null association between red meat or processed red meat and incidence of AML [24]. Similarly, two cohort studies from the United States revealed the null association between red meat or processed red meat and incidence of AML [22, 23]. For MDS, there have been even fewer epidemiological studies. A case-control study revealed that an increased intake of meat was associated with an increased incidence of MDS; however, a cohort study from the United States revealed a null association [25, 28]. The present study revealed that an increased intake of processed red meat was associated with an increased incidence of AML/MDS. Subtype analysis showed similar results, although the statistical significance was not maintained. Our results were inconsistent with those of previous studies conducted in the United States or Europe. A possible reason of the inconsistent result was that the intake of red meat and processed meat in Western countries were higher than that in Japan [40, 41]. The differing amounts of processed red meat intakes, which was smaller in our study than in previous studies, could have influenced the results if a smaller amount of processed red meat intake was associated with a decreased incidence of AML/MDS [22–24]. The chemicals contained in processed red meat such as antioxidants and preservatives could be associated with the incidence of AML/MDS [42]. Other possibilities were the difference in the geographic distribution of AML subtypes between the West and Asia and unmeasured confounders, such as professional exposure to chemicals. On the other hand, the intakes of saturated fatty acids, which is one of the component of meat, was not associated with the incidence of AML among epidemiological data, although the number of studies was limited and most of the studies were case-control in design [18, 26, 27]. Our results revealed that saturated fatty acid intake was not associated with the incidence of AML/MDS with HR of 1.01–1.42, a finding that was consistent with those of previous studies [18, 26, 27]. Further research is needed to assess the detailed etiology of the association between processed red meat intakes and the incidence of AML/MDS. A recent meta-analysis that combined two cohort studies revealed that fish intake might increase the risk of the development of AML with an HR of 1.73 ($95\%$ CI, 1.22–2.47) for the highest versus lowest intake group [12], although fish are generally associated with reduced cancer risks via suppressing mutations, enhancing cell apoptosis, and inhibiting cell growth [43–45]. The results of the meta-analysis are considered due to fish containing some carcinogens such as sea pollution and dioxins that are related to AML development [12]. One of the European study included in the meta-analysis revealed that the median fish intake estimated from FFQ was 21.5 g/day, which was lower than our study and fish intake had trend to increase risk of AML with an HR of 1.50 (0.99–2.26) [24]. In contrast with the meta-analysis, our results indicated a null association between total fish, big and small fish intake and AML/MDS incidence, including subtype analysis. The different results from some previous reports might be due to the difference in amounts of fish intakes, the types of fish considering the small fish intake trend to higher risk of AML/MDS in our study, differences in environmental pollution of producing area, and eating habits of fish, and differences in AML/MDS subtypes. Our study has some limitations. First, the effects of meat cooking methods, doneness levels, and additives were not considered due to a lack of data. Therefore, we have no suggestions or speculations regarding possible carcinogens such as heterocyclic amine, nitrites and so on [46, 47]. Second, it was unable to account for changes in exposure over the course of study period in this study. Third, unmeasured confounding factors may have influenced the intake of interest. Finally, the statistical power for the analysis might not be sufficient for the total outcome, especially subtypes. In addition, a detailed subtype analysis was not performed because of the insufficient incidence of AML. Therefore, larger studies and meta-analysis are required to confirm the results of this study. Readers should interpret our results cautiously while bearing these limitations in mind. ## Conclusion We revealed that processed red meat was associated with an increased incidence of AML/MDS in the Japanese population. ## Ethics approval and consent to participate The study was approved by the ethics committees of the National Cancer Center (Approval number: 2015-085, date: November 25, 2021) and Osaka University (Approval number: 14020-10, date: February 10, 2021). ## Consent for publish Not applicable. ## Availability of data and materials For information on how to gain access to JPHC data, follow the instructions at https://epi.ncc.go.jp/en/jphc/$\frac{805}{8155.}$html. ## Competing interests The authors declare no competing financial interests in this study. ## Funding This study was supported by a grant from the Food Safety Commission, Cabinet Office, Government of Japan (Research Program for Risk Assessment Study on Food Safety, no. 1503; principal investigator, TS), the National Cancer Center Research and Development Fund (since 2011; principal investigators, ST [∼2019] and NS [2020∼], and a Grant-in-Aid for Cancer Research from the Ministry of Health, Labour and Welfare of Japan (from 1989 to 2010; principal investigator [1997–2010], ST). ## Author’s contribution YS, TS, TK, and LZ designed the study. NS and TS acquired data. YS analyzed the data, performed the statistical analysis, and wrote the first draft of the manuscript. TS had primary responsibility for final content. All authors contributed critical review of the analysis data and the manuscript and approved the final version. ## Acknowledgements JPHC members are listed at the following site (as of March 2021): https://epi.ncc.go.jp/en/jphc/$\frac{781}{8510.}$html. We are indebted to the Aomori, Akita, Iwate, Ibaraki, Niigata, Osaka, Kochi, Nagasaki, and Okinawa Cancer Registries for providing their incidence data. ## References 1. Shallis RM, Wang R, Davidoff A, Ma X, Zeidan AM. Epidemiology of acute myeloid leukemia: Recent progress and enduring challenges. Blood Rev (2019.0) 36 70-87. PMID: 31101526 2. Zeidan AM, Shallis RM, Wang R, Davidoff A, Ma X. Epidemiology of myelodysplastic syndromes: Why characterizing the beast is a prerequisite to taming it. Blood Rev (2019.0) 34 1-15. PMID: 30314642 3. McHale CM, Zhang L, Smith MT. Current understanding of the mechanism of benzene-induced leukemia in humans: implications for risk assessment. Carcinogenesis (2012.0) 33 240-52. PMID: 22166497 4. Lv L, Lin G, Gao X, Wu C, Dai J, Yang Y. Case-control study of risk factors of myelodysplastic syndromes according to World Health Organization classification in a Chinese population. Am J Hematol (2011.0) 86 163-9. PMID: 21264898 5. Pesatori AC, Consonni D, Rubagotti M, Grillo P, Bertazzi PA. Cancer incidence in the population exposed to dioxin after the “Seveso accident”: twenty years of follow-up. Environ Health Glob Access Sci Source (2009.0) 8 39 6. Van Maele-Fabry G, Duhayon S, Mertens C, Lison D. Risk of leukaemia among pesticide manufacturing workers: a review and meta-analysis of cohort studies. Environ Res (2008.0) 106 121-37. PMID: 18028905 7. Jin J, Yu M, Hu C, Ye L, Xie L, Jin J. Pesticide Exposure as a Risk Factor for Myelodysplastic Syndromes: A Meta-Analysis Based on 1,942 Cases and 5,359 Controls. PLoS One (2014.0) 9 e110850. PMID: 25335083 8. Strom SS, Gu Y, Gruschkus SK, Pierce SA, Estey EH. Risk factors of myelodysplastic syndromes: a case-control study. Leukemia (2005.0) 19 1912-8. PMID: 16167059 9. Haorah J, Zhou L, Wang X, Xu G, Mirvish SS. Determination of total N-nitroso compounds and their precursors in frankfurters, fresh meat, dried salted fish, sauces, tobacco, and tobacco smoke particulates. J Agric Food Chem (2001.0) 49 6068-78. PMID: 11743810 10. Knize MG, Salmon CP, Pais P, Felton JS. Food heating and the formation of heterocyclic aromatic amine and polycyclic aromatic hydrocarbon mutagens/carcinogens. Adv Exp Med Biol (1999.0) 459 179-93. PMID: 10335376 11. Huang Y, Cao D, Chen Z, Chen B, Li J, Guo J. Red and processed meat consumption and cancer outcomes: Umbrella review. Food Chem (2021.0) 356 129697. PMID: 33838606 12. Sergentanis TN, Ntanasis-Stathopoulos I, Tzanninis IG, Gavriatopoulou M, Sergentanis IN, Dimopoulos MA. Meat, fish, dairy products and risk of hematological malignancies in adults - a systematic review and meta-analysis of prospective studies. Leuk Lymphoma (2019.0) 60 1978-90. PMID: 30912696 13. Bosch AC, O’Neill B, Sigge GO, Kerwath SE, Hoffman LC. Heavy metals in marine fish meat and consumer health: a review. J Sci Food Agric (2016.0) 96 32-48. PMID: 26238481 14. Zhang Q, Yang L, Wang WX. Bioaccumulation and trophic transfer of dioxins in marine copepods and fish. Environ Pollut Barking Essex 1987 (2011.0) 159 3390-7 15. Bertazzi PA, Consonni D, Bachetti S, Rubagotti M, Baccarelli A, Zocchetti C. Health effects of dioxin exposure: a 20-year mortality study. Am J Epidemiol (2001.0) 153 1031-44. PMID: 11390319 16. MacLean CH, Newberry SJ, Mojica WA, Khanna P, Issa AM, Suttorp MJ. Effects of omega-3 fatty acids on cancer risk: a systematic review. JAMA (2006.0) 295 403-15. PMID: 16434631 17. Charbonneau B, O’Connor HM, Wang AH, Liebow M, Thompson CA, Fredericksen ZS, Macon WR, Slager SL, Call TG, Habermann TM. Trans fatty acid intake is associated with increased risk and n3 fatty acid intake with reduced risk of non-hodgkin lymphoma. J Nutr (2013.0) 143 672-81. PMID: 23486982 18. Laake I, Carlsen MH, Pedersen JI, Weiderpass E, Selmer R, Kirkhus B. Intake of trans fatty acids from partially hydrogenated vegetable and fish oils and ruminant fat in relation to cancer risk. Int J Cancer (2013.0) 132 1389-403. PMID: 22821174 19. Shimomura Y, Sobue T, Zha L, Kitamura T, Iwasaki M, Inoue M. Association between Meat, Fish, and Fatty Acid Intake and Non-Hodgkin Lymphoma Incidence: The Japan Public Health Center-Based Prospective Study. J Nutr (2022.0) 152 1895-906. PMID: 35675313 20. Cross AJ, Leitzmann MF, Gail MH, Hollenbeck AR, Schatzkin A, Sinha R. A Prospective Study of Red and Processed Meat Intake in Relation to Cancer Risk. PLoS Med (2007.0) 4 e325. PMID: 18076279 21. Daniel CR, Cross AJ, Graubard BI, Hollenbeck AR, Park Y, Sinha R. Prospective investigation of poultry and fish intake in relation to cancer risk. Cancer Prev Res (Phila) (2011.0) 4 1903-11. PMID: 21803982 22. Ma X, Park Y, Mayne ST, Wang R, Sinha R, Hollenbeck AR. Diet, Lifestyle, and Acute Myeloid Leukemia in the NIH-AARP Cohort. Am J Epidemiol (2010.0) 171 312-22. PMID: 20042434 23. Ross JA, Kasum CM, Davies SM, Jacobs DR, Folsom AR, Potter JD. Diet and risk of leukemia in the Iowa Women’s Health Study. Cancer Epidemiol Biomark Prev Publ Am Assoc Cancer Res Cosponsored Am Soc Prev Oncol (2002.0) 11 777-81 24. Saberi Hosnijeh F, Peeters P, Romieu I, Kelly R, Riboli E, Olsen A. Dietary Intakes and Risk of Lymphoid and Myeloid Leukemia in the European Prospective Investigation into Cancer and Nutrition (EPIC). Nutr Cancer (2014.0) 66 14-28. PMID: 24279598 25. Ma X, Lim U, Park Y, Mayne ST, Wang R, Hartge P. Obesity, Lifestyle Factors, and Risk of Myelodysplastic Syndromes in a Large US Cohort. Am J Epidemiol (2009.0) 169 1492-9. PMID: 19395696 26. Fritschi L, Ambrosini GL, Kliewer EV, Johnson KC. Dietary fish intake and risk of leukaemia, multiple myeloma, and non-Hodgkin lymphoma. Cancer Epidemiol Biomark Prev Publ Am Assoc Cancer Res Cosponsored Am Soc Prev Oncol (2004.0) 13 532-7 27. Hu J, Vecchia CL, de Groh M, Negri E, Morrison H, Mery L. Dietary transfatty acids and cancer risk. Eur J Cancer Prev (2011.0) 20 530-8. PMID: 21701388 28. Avgerinou C, Giannezi I, Theodoropoulou S, Lazaris V, Kolliopoulou G, Zikos P. Occupational, dietary, and other risk factors for myelodysplastic syndromes in Western Greece. Hematology (2017.0) 22 419-29. PMID: 28102107 29. Hu J, Vecchia CL, DesMeules M, Negri E, Mery L. Meat and Fish Consumption and Cancer in Canada. Nutr Cancer (2008.0) 60 313-24. PMID: 18444165 30. Tsugane S, Sawada N. The JPHC study: design and some findings on the typical Japanese diet. Jpn J Clin Oncol (2014.0) 44 777-82. PMID: 25104790 31. Nanri A, Mizoue T, Noda M, Takahashi Y, Matsushita Y, Poudel-Tandukar K. Fish intake and type 2 diabetes in Japanese men and women: the Japan Public Health Center-based Prospective Study. Am J Clin Nutr (2011.0) 94 884-91. PMID: 21775559 32. 32.Science and Technology Agency. Standard Tables of Food Composition in Japan. 7th Revised and Enlarged edn. Tokyo: Printing Bureau of the Ministry of Finance, 2015. https://www.mext.go.jp/en/policy/science_technology/policy/title01/detail01/1374030.htm Accessed 1 Sep, 2022. 33. W W, Mj S. Total energy intake: implications for epidemiologic analyses. Am J Epidemiol (1986.0) 124 17-27. PMID: 3521261 34. Sasaki S, Kobayashi M, Tsugane S. Validity of a Self-administered Food Frequency Questionnaire Used in the 5-year Follow-up Survey of the JPHC Study Cohort I: Comparison with Dietary Records for Food Groups. J Epidemiol (2003.0) 13 57-63 35. Kobayashi M, Sasaki S, Kawabata T, Hasegawa K, Tsugane S. Validity of a Self-administered Food Frequency Questionnaire Used in the 5-year Follow-up Survey of the JPHC Study Cohort I to Assess Fatty Acid Intake: Comparison with Dietary Records and Serum Phospholipid Level. J Epidemiol (2003.0) 13 64-81 36. Ishihara J, Sobue T, Yamamoto S, Yoshimi I, Sasaki S, Kobayashi M. Validity and Reproducibility of a Self-administered Food Frequency Questionnaire in the JPHC Study Cohort II: Study Design, Participant Profile and Results in Comparison with Cohort I. J Epidemiol (2003.0) 13 134-47 37. Sasaki S, Ishihara J, Tsugane S. Reproducibility of a self-administered food frequency questionnaire used in the 5-year follow-up survey of the JPHC Study Cohort I to assess food and nutrient intake. J Epidemiol (2003.0) 13 S115-24. PMID: 12701639 38. Ugai T, Matsuo K, Sawada N, Iwasaki M, Yamaji T, Shimazu T. Coffee and green tea consumption and subsequent risk of acute myeloid leukemia and myelodysplastic syndromes in Japan. Int J Cancer (2018.0) 142 1130-8. PMID: 29076523 39. Ugai T, Matsuo K, Oze I, Ito H, Wakai K, Wada K. Smoking and subsequent risk of acute myeloid leukaemia: A pooled analysis of 9 cohort studies in Japan. Hematol Oncol (2018.0) 36 262-8. PMID: 28681440 40. 40.Ministry of Health, Labour and Welfare. Ministry of Health, Labour and Welfare (2021) https://www.mhlw.go.jp/index.html Accessed 10 Jan, 2023. 41. 41.National Center for Health Statistics (NCHS) (2021): National Health and Nutrition Examination Survey (NHANES) https://www.cdc.gov/nchs/nhanes/index.htm Accessed 10 Jan, 2023. 42. Zengin N, Yüzbaşıoğlu D, Unal F, Yılmaz S, Aksoy H. The evaluation of the genotoxicity of two food preservatives: sodium benzoate and potassium benzoate. Food Chem Toxicol Int J Publ Br Ind Biol Res Assoc (2011.0) 49 763-9 43. Yang L, Shi WY, Xu XH, Wang XF, Zhou L, Wu DP. Fish consumption and risk of non-Hodgkin lymphoma: A meta-analysis of observational studies. Hematology (2020.0) 25 194-202. PMID: 25428434 44. Lee KH, Seong HJ, Kim G, Jeong GH, Kim JY, Park H. Consumption of Fish and ω-3 Fatty Acids and Cancer Risk: An Umbrella Review of Meta-Analyses of Observational Studies. Adv Nutr Bethesda Md (2020.0) 11 1134-49 45. Simopoulos AP. Omega-3 fatty acids in health and disease and in growth and development. Am J Clin Nutr (1991.0) 54 438-63. PMID: 1908631 46. Sinha R, Knize MG, Salmon CP, Brown ED, Rhodes D, Felton JS. Heterocyclic amine content of pork products cooked by different methods and to varying degrees of doneness. Food Chem Toxicol Int J Publ Br Ind Biol Res Assoc (1998.0) 36 289-97 47. Chazelas E, Pierre F, Druesne-Pecollo N, Esseddik Y, Szabo de Edelenyi F, Agaesse C. Nitrites and nitrates from food additives and natural sources and cancer risk: results from the NutriNet-Santé cohort. Int J Epidemiol (2022.0) dyac046
--- title: 'The effect of depressive symptoms on disability-free survival in healthy older adults: A prospective cohort study' authors: - Greg Roebuck - Mojtaba Lotfaliany - Bruno Agustini - Malcolm Forbes - Mohammadreza Mohebbi - John McNeil - Robyn L. Woods - Christopher M. Reid - Mark R. Nelson - Raj C. Shah - Joanne Ryan - Anne B. Newman - Alice Owen - Rosanne Freak-Poli - Nigel Stocks - Michael Berk journal: Acta psychiatrica Scandinavica year: 2022 pmcid: PMC10026010 doi: 10.1111/acps.13513 license: CC BY 4.0 --- # The effect of depressive symptoms on disability-free survival in healthy older adults: A prospective cohort study ## Abstract ### Background: Gerontology and ageing research are increasingly focussing on healthy life span (healthspan), the period of life lived free of serious disease and disability. Late-life depression (LLD) is believed to impact adversely on physical health. However, no studies have examined its effect on healthspan. This study investigated the effect of LLD and subthreshold depression on disability-free survival, a widely accepted measure of healthspan. ### Methods: This prospective cohort study used data from the ASPirin in Reducing Events in the Elderly study. Participants were aged ≥70 years (or ≥65 years for African-American and Hispanic participants) and free of dementia, physical disability and cardiovascular disease. Depressive symptoms were measured using the 10-item Centre for Epidemiological Studies Depression Scale (CES-D-10). LLD and subthreshold depression were defined as CES-D-10 scores ≥8 and 3–7, respectively. Disability-free survival was defined as survival free of dementia and persistent physical disability. ### Results: A total of 19,110 participants were followed up for a maximum of 7.3 years. In female participants, LLD was associated with lower disability-free survival adjusting for sociodemographic and lifestyle factors, medical comorbidities, polypharmacy, physical function and antidepressant use (HR, 1.50; $95\%$ CI, 1.23–1.82). In male participants, LLD was associated with lower disability-free survival adjusting for sociodemographic and lifestyle factors (HR, 1.30; $95\%$ CI, 1.03–1.64). Subthreshold depression was also associated with lower disability-free survival in both sexes. ### Conclusions: LLD may be a common and important risk factor for shortened healthspan. ## INTRODUCTION Life expectancy has increased dramatically since the mid-twentieth century. Between 1950 and 2017, global life expectancy at birth rose from 52.9 to 75.6 years for women and from 48.1 to 70.5 years for men.1 This ageing of the global population has been associated with a substantial increase in the prevalence of chronic diseases. In the developed world, $87\%$ of people aged over 65 years suffer from at least one chronic illness and $66\%$ suffer from two or more such illnesses.2 Given the suffering and disability caused by age-related chronic diseases, medicine is increasingly focussing on prolonging healthy life span, or “healthspan”. This is the period of life lived free of serious illness and disability.3 There is growing interest in the lifestyle and other factors that determine healthspan and the ability of clinical interventions to extend healthspan.4–6 Late-life depression (LLD) can be defined as major depressive disorder occurring in adults aged 60 years or older. It is common, with an estimated global prevalence of $13.3\%$.7 A substantial body of evidence suggests that LLD impacts adversely on older adults’ physical health. Longitudinal studies have found that it is associated with higher rates of all-cause and cardiovascular mortality.8 LLD also appears to be an independent risk factor for developing cardiovascular disease (CVD) and for poor outcomes in established CVD.9 Finally, LLD prospectively predicts the development of common geriatric syndromes, including frailty, dementia and disability.10–12 These findings suggest that LLD is likely to shorten healthspan. There is some evidence from studies not limited to geriatric populations that depression is associated with lower healthy life expectancy.13,14 A recent study of 9761 individuals aged 50 years and older in the United Kingdom found that the presence of depressive symptoms was associated with lower disability-free life expectancy across all age groups.14 Consulting a general practitioner about depressive symptoms for the first time after the age of 60 years also predicts shorter disability-free life expectancy.15 As far as we are aware, however, no large observational studies have investigated the effect of LLD, assessed using a validated measure of depressive symptoms, on healthspan. The aim of the current study was to examine the effect of LLD and subthreshold depressive symptoms on disability-free survival in physically healthy older adults. Disability-free survival is a widely accepted measure of healthspan or health expectancy.6,16 In this study, it was defined as survival free of dementia and persistent physical disability. Persistent physical disability was defined in turn as self-reported severe difficulty or inability to perform independently one or more of six basic activities of daily life (ADLs) (walking across a room, bathing, dressing, transferring from a bed or chair, toileting and eating) for at least 6 months.17 We hypothesised that both LLD and subthreshold depression would be associated with lower disability-free survival in female and male participants. We further hypothesised that there would be a dose-response relationship, with LLD associated with a larger reduction than subthreshold depression. Secondarily, we aimed to explore the effects of LLD on the individual outcomes of all-cause mortality, incident dementia and incident persistent physical disability. ## Study population This prospective cohort study used data from the ASPirin in Reducing Events in the Elderly (ASPREE) study, which was a large, multi-centre, population-based randomised controlled trial that investigated the effect of aspirin on disability-free survival and other outcomes in healthy older adults. Participants were recruited from Australia and the United States between 2010 and 2014. In Australia, recruitment was mostly based in general practise. In the United States, prospective participants were identified through clinic-based mailing lists, electronic medical records and responses to media advertisements. The inclusion criteria required participants to be aged 70 years or over (or 65 years or over for African-American and Hispanic participants in the United States) and able and willing to give informed consent. Exclusion criteria included a history of CVD or atrial fibrillation, a clinical diagnosis of dementia or score of <78 on the Modified Mini-Mental State Examination (3MS),18 physical disability, anaemia, a current or recurrent condition with a high risk of major bleeding, uncontrolled hypertension, absolute contraindication or allergy to aspirin, current continuous use of aspirin for secondary prevention, current continuous use of other antiplatelet or anticoagulant medication and any medical condition likely to cause death within 5 years. The design of the ASPREE study has been reported elsewhere.4 ## Procedures and measures Participants attended for assessments at baseline and then annually for the duration of their participation in the study. They also had telephone contact with the investigators every 3 months between visits. At each visit, participants were asked about demographic and lifestyle factors, their medical history and their medication use. They completed the LIFE disability questionnaire19 and other self-report measures. Their weight, height, waist circumference, blood pressure and heart rate were measured. A blood sample was taken to measure haemoglobin, creatinine, fasting glucose and fasting lipids and a urine sample was taken to measure urinary albumin:creatinine ratio. Every second year, hand grip strength was measured using a dynamometer and time to walk 3 m was measured. ## Measurement of depressive symptoms Depressive symptoms were measured at baseline using the 10-item version of the Centre for Epidemiological Studies Depression Scale (CES-D-10).20 The CES-D-10 is a self-report questionnaire that assesses the severity of depressive symptoms over the previous week. It consists of 10 items rated on a 4-point Likert-type scale from 0 (“rarely or none of the time”) to 3 (“all of the time”). Using a cut-off score of ≥10, it has very tight agreement (κ = 0.97) with the full 20-item CES-D.20 A recent meta-analysis found that the 20-item CES-D has a sensitivity of $81\%$ and a specificity of $78\%$ for identifying LLD.21 LLD and subthreshold depression were defined as CES-D-10 scores of ≥8 and 3–7, respectively. Participants with a CES-D-10 score ≤2 were considered to have minimal or no depressive symptoms. The cut-off of ≥8 for LLD was consistent with previous studies using the CES-D-10.22 The ranges for the subthreshold depression and no depression groups were based on an earlier analysis by our group that showed that ASPREE participants’ CES-D-10 scores tended to follow four different trajectories: consistently low, consistently moderate, consistently high and initially low and then increasing scores.23 These ranges were derived from the interquartile ranges for the consistently moderate and consistently low classes in this analysis, respectively. ## Primary and secondary outcomes Disability-free survival was defined as survival free of dementia or persistent physical disability and assessed as a composite of the first event of death, dementia or persistent physical disability. Dementia was defined according to the criteria in the fourth edition of the American Psychiatric Association’s Diagnostic and Statistical Manual of Mental Disorders24 (DSM-IV). Triggers for a dementia assessment included a 3MS score of <78, a decrease in 3MS score of >10.15 points from the participant’s baseline score after adjustment for age and level of education, a clinical diagnosis of dementia, a report of cognitive concerns to a specialist and the prescription of a cholinesterase inhibitor (for Australian participants). Dementia assessments involved additional cognitive and functional testing by research staff, the collection of ancillary data, including laboratory tests and neuroimaging, and a review of the participant’s medical records. The Dementia Adjudication Panel then determined whether the participant met DSM-IV criteria for dementia.4 The individual endpoints of all-cause mortality, incident dementia and incident persistent physical disability were treated as secondary endpoints. For participants who met the primary composite endpoint, follow up continued with respect to the remaining secondary endpoints. ## Statistical analysis Differences in the baseline characteristics of the LLD, subthreshold depression and no depression groups were explored using a three-way analysis of variance for continuous variables and the chi-square test of independence for frequencies. For each endpoint, survival time was coded as the time between the baseline assessment and the occurrence of the endpoint or, if the endpoint did not occur, the final study visit. The Kaplan–Meier method was used to estimate survival functions for the three groups for each endpoint. The relationship between group and survival time was explored using Cox proportional hazards regression models. To explore the factors contributing to any associations between depressive symptoms and the endpoints, including mediators, moderators and confounding factors, five different regression models were tested. Each model adjusted for an additional set of covariates. Model 1 adjusted for sociodemographic factors (age and race). Model 2 further adjusted for lifestyle factors (body mass index [BMI], alcohol use history, smoking history, level of education and accommodation status). Model 3 further adjusted for the presence of common medical conditions (hypertension, dyslipidaemia, diabetes mellitus, chronic kidney disease (CKD), respiratory disease, gastroesophageal reflux disease (GORD), gout, Parkinson’s disease and a cancer history) and polypharmacy (simultaneous use of ≥5 medications). Model 4 further adjusted for measures of physical function (grip strength, gait speed and self-reported longest time walking without rest). Model 5 further adjusted for use of antidepressant medications at baseline. An F-test of overall significance was performed to calculate p-values for the three-way comparisons between the groups for each endpoint in Model 5. The threshold for statistical significance was set at 0.05 and the Benjamini-Hochberg procedure was used to adjust for multiple comparisons with the false discovery rate set to 0.05.25 Data for female and male participants were analysed separately. ## Baseline characteristics of three groups and incidence of endpoints A total of 19,110 participants completed the CES-D-10 at baseline and were included in the study: 10,799 female participants and 8311 male participants. Table 1 shows the sociodemographic and lifestyle characteristics, rates of common medical conditions and polypharmacy, performance on physical function measures, rates of antidepressant medication use and mean CES-D-10 scores at baseline for the LLD, subthreshold depression and no depression groups. Table 2 shows the incidences of the primary and secondary endpoints for the three groups during the follow-up period. ## Disability-free survival After a median (range) follow-up period of 4.7 (0–7.3) years and 4.5 (0–7.3) years, 918 female participants and 917 male participants had died or developed dementia or persistent physical disability as a first event. Adjusting for all covariates, the LLD, subthreshold depression and no depression groups differed significantly for female participants ($p \leq 0.001$) but not male participants ($$p \leq 0.746$$) (Table 3 and Figure 1). For female participants, LLD was associated with lower disability-free survival adjusting for all covariates (hazard ratio (HR), 1.50; $95\%$ confidence interval (CI), 1.23–1.82). Subthreshold depression was associated with lower disability-free survival adjusting for sociodemographic and lifestyle factors, medical comorbidities and polypharmacy (HR, 1.19; $95\%$ CI, 1.03–1.38). The association became non-significant after adjustment was made for physical function, however. For male participants, LLD and subthreshold depression were associated with lower disability-free survival adjusting for sociodemographic and lifestyle factors (LLD: HR, 1.30; $95\%$ CI, 1.03–1.64; subthreshold depression: HR, 1.30; $95\%$ CI, 1.03–1.36). The associations became non-significant after adjustment was made for medical comorbidities and polypharmacy, however. The full results of the Cox proportional hazards regression analysis for Model 5, which included all covariates, are set out in Table S1 in the Supplementary Materials. ## All-cause mortality After a median (range) follow-up period of 4.8 (0–7.3) years and 4.6 (0–7.3) years, 469 female participants and 583 male participants had died. Adjusting for all covariates, the three groups did not differ significantly for female participants ($$p \leq 0.029$$) or male participants ($$p \leq 0.501$$) when correction was made for multiple comparisons (Table 3 and Figure S1, Supplementary Materials). For female participants, LLD was associated with higher mortality adjusting for all covariates (HR, 1.44; $95\%$ CI, 1.08–1.91). Subthreshold depression was associated with higher mortality adjusting for sociodemographic and lifestyle factors, medical comorbidities and polypharmacy (HR, 1.26; $95\%$ CI, 1.03–1.55). The association became non-significant when adjustment was made for physical function, however. For male participants, subthreshold depression was associated with higher mortality adjusting for sociodemographic factors (HR, 1.19; $95\%$ CI, 1.01–1.42). There was otherwise no association between depressive symptoms and mortality. ## Dementia After a median (range) follow-up period of 4.6 (0–7.3) years and 4.5 (0–7.3) years, 302 female participants and 273 male participants had developed dementia. Adjusting for all covariates, the three groups did not differ significantly for female participants ($$p \leq 0.147$$) or male participants ($$p \leq 0.562$$) (Table 3 and Figure S2, Supplementary Materials). For female participants, LLD was associated with dementia adjusting for all covariates (HR, 1.43; $95\%$ CI, 1.01–2.03). There were otherwise no associations between depressive symptoms and dementia for female or male participants. ## Persistent physical disability After a median (range) follow-up period of 4.0 (0–7.0) years and 4.0 (0–7.0) years, 238 female participants and 174 male participants had developed persistent physical disability. Adjusting for all covariates, there was a significant difference between the three groups for female participants ($$p \leq 0.006$$) but not male participants ($$p \leq 0.045$$) following correction for multiple comparisons (Table 3 and Figure S3, Supplementary Materials). For female participants, LLD was associated with persistent physical disability adjusting for all covariates (HR, 1.87; $95\%$ CI, 1.29–2.73). Subthreshold depression was associated with persistent physical disability adjusting for sociodemographic and lifestyle factors (HR, 1.35; $95\%$ CI, 1.01–1.81). The association became non-significant when adjustment was made for medical comorbidities and polypharmacy. For male participants, LLD was associated with persistent physical disability adjusting for sociodemographic and lifestyle factors, medical comorbidities, polypharmacy and physical function (HR, 1.75; $95\%$ CI, 1.06–2.90). The association became non-significant when adjustment was made for antidepressant medication use at baseline. Subthreshold depression was associated with persistent physical disability adjusting for all covariates (HR, 1.47; $95\%$ CI, 1.05–2.06). ## DISCUSSION To our knowledge, this is the first study to investigate the effect of LLD on disability-free survival. Consistent with our hypotheses, LLD was associated with lower disability-free survival in both sexes. In female participants, this association remained significant after adjustment for sociodemographic and lifestyle factors, medical comorbidities, polypharmacy, physical function measures and antidepressant medication use at baseline. Adjusting for these factors, the risk of dying or developing dementia or persistent physical disability during the follow-up period was $50\%$ higher for women with LLD compared with women with minimal or no depressive symptoms. In male participants, the association between LLD and lower disability-free survival was significant adjusting for sociodemographic and lifestyle factors but became non-significant after adjustment was made for medical comorbidities and polypharmacy. Subthreshold depressive symptoms were also associated with lower disability-free survival in both sexes. The magnitudes of these associations were smaller than those for LLD, suggesting the existence of a dose-response relationship regarding the effect of depressive symptoms on disability-free survival. Taken together, these findings suggest that LLD may have a causal, and not merely associational relationship, with lower disability-free survival. The study also partially replicated previous findings that LLD is prospectively associated with all-cause mortality, dementia and physical disability.8,10,11 In female participants, LLD was associated with both all-cause mortality and dementia adjusting for all covariates. In male participants, subthreshold depression was associated with all-cause mortality adjusting for age and race. In both sexes, LLD was associated with persistent physical disability. This was by far the strongest relationship that existed between LLD and the endpoints. Adjusting for all covariates, the risk of developing persistent physical disability during the follow-up period was $87\%$ higher for female participants with LLD compared with those with minimal or no depressive symptoms. The associations between depressive symptoms and the endpoints were in almost all cases stronger for female participants than male participants. This was unanticipated because previous studies have not generally found that LLD has more severe physical health impacts in females. It is possible that males’ shorter life expectancy and earlier onset of age-related morbidity meant that the inclusion criteria, which required participants to be free of physical disability and cardiovascular disease, resulted in a cohort of male participants who were healthier and therefore less susceptible to the effects of LLD than their female counterparts. The fact that female participants had a higher rate of polypharmacy at baseline than male participants ($23.6\%$ vs. $15.6\%$) lends some support to this possibility. Alternatively, the disparity may have been due to the study’s greater statistical power for women. The numbers (percentages) of female participants in the LLD and subthreshold depression groups were 1248 ($11.6\%$) and 4378 ($40.6\%$), respectively, compared with 631 ($7.6\%$) and 2981 ($35.9\%$) male participants. However, the widths of the confidence intervals for most endpoints were similar for both sexes, indicating that statistical power is at most only part of the explanation. There are likely multiple biopsychosocial mechanisms underlying the associations found between LLD and the endpoints. Depression has a complex and bidirectional relationship with physical illness and numerous biological, behavioural and psychosocial mechanisms are believed to contribute to this relationship. One intriguing but speculative hypothesis is that depression and other psychiatric disorders involve common pathophysiological processes that contribute concurrently to the development of physical morbidity. Such processes may include the activation of immune/inflammatory pathways, increased oxidative and nitrosative stress and mitochondrial dysfunction.27 Given the strong relationship that was found to exist between LLD and persistent physical disability, it is interesting to consider the mechanisms underlying this relationship. Little is known regarding these mechanisms, although proposed causal pathways include: (i) amotivation causing physical inactivity, leading to deconditioning and frailty, (ii) decreased appetite and poor nutrition contributing to skeletal muscle loss and sarcopenia, (iii) self-neglect leading to risky health behaviours such as alcohol abuse and thereby causing medical comorbidities that lead to disability and (iv) depression-related deficits in executive function impairing a depressed person’s ability to perform the cognitive aspects of ADLs.11 The results of this study suggest that these mechanisms may indeed mediate part of the relationship between LLD and physical disability. For example, the HRs for the endpoint of persistent physical disability decreased from 2.22 to 1.97 for women and from 2.17 to 1.75 for men after adjustment was made for physical function measures. This is consistent with factors such as inactivity, frailty and skeletal muscle loss partly mediating the relationship between LLD and physical disability. However, the fact that a strong relationship persisted even after adjustment for these measures suggests that there are other important mechanisms contributing to the relationship. One possibility is that psychological mechanisms are operative. A psychological variable that may partially mediate the relationship between LLD and physical disability is apathy. Apathy is a common symptom of LLD, occurring in nearly $40\%$ of cases.28 There is evidence that it is related to physical disability independently of other depressive symptoms.28 Apathy also appears to be associated with a decline in self-report measures of physical function but not in objective measures such as gait speed or handgrip strength.29 It is possible that apathy affects depressed older adults’ perceptions of their functional capacities and thereby contributes to an impairment of these capacities. If psychological variables partially mediate the relationship between LLD and physical disability, this suggests that depression-related disability, even if persistent, may be amenable to treatment. This study has a number of strengths, including its large sample size, its prospective design, the use of a validated measure of depressive symptoms and the collection of data according to a well-designed randomised controlled trial protocol with rigorous procedures for assessing endpoints. Rates of attrition and missing data were low. The study also has some limitations. First, the CES-D-10 is a screening instrument and is not equivalent to diagnosis by a trained clinician using a semi-structured diagnostic instrument, the gold standard for psychiatric diagnosis. Second, a single cross-sectional CES-D-10 score was used to assign participants to the three groups. This may have inflated the LLD and subthreshold depression groups by, for instance, including participants experiencing transient mood symptoms in these groups. Third, the median follow-up period for the primary composite endpoint was only 4.7 and 4.5 years for female and male participants, respectively. This means that associations between LLD and lower healthspan could have been due to undiagnosed subclinical or prodromal conditions causing depressive symptoms. For instance, there is some evidence that prodromal dementia can cause depressive symptoms.30 Fourth, the study did not adjust for all potential confounders. For example, low subjective social status and physical pain are risk factors for LLD that were not controlled for.31 This is also consistent with LLD being merely a marker of risk factors for lower healthspan and not itself having any causal effects on healthspan. Finally, study participants were healthier than the overall population of older adults, which may limit generalisability. ## CONCLUSION This study suggests that LLD may be a common risk factor for shortened healthspan. This underscores the importance of identifying and treating depression for healthy ageing. ## DATA AVAILABILITY STATEMENT Research data are not shared. ## FUNDING INFORMATION ASPREE was supported by grants from the National Institute on Ageing and the National Cancer Institute at the U.S. National Institutes of Health (U01AG029824 and U19AG062682); the National Health and Medical Research Council of Australia (NHMRC) (334047, 1081901 and 1127060); Monash University (Australia) and the Victorian Cancer Agency (Australia). JM is supported by a NHMRC Investigator Grant [1173690], CMR by a NHMRC Principal Research Fellowship [1136372], JR by a NHMRC Boosting Dementia Research Leader Fellowship [1135727], RFP by a National Heart Foundation of Australia Post-Doctoral Fellowship [101927] and MB by a NHMRC Senior Principal Research Fellowship [1156072]. We would like to thank the ASPREE participants who volunteered for this study, the general practitioners and staff of the medical clinics who support the study participants and the trial staff and management team of the ASPREE study in Australia and the United States (www.aspree.org). ## Funding information U.S. National Institutes of Health, Grant/Award Numbers: U19AG062682, U01AG029824; National Health and Medical Research Council of Australia, Grant/Award Numbers: 1127060, 1081901, 334047; Monash University; Victorian Cancer Agency; NHMRC Investigator Grant, Grant/Award Number: 1173690; NHMRC Principal Research Fellowship, Grant/Award Number: 1136372; NHMRC Boosting Dementia Research Leader Fellowship, Grant/Award Number: 1135727; National Heart Foundation of Australia Post-Doctoral Fellowship, Grant/Award Number: 101927; NHMRC Senior Principal Research Fellowship, Grant/Award Number: 1156072 ## References 1. Dicker D, Nguyen G, Abate D. Global, regional, and national age-sex-specific mortality and life expectancy, 1950–2017: a systematic analysis for the Global Burden of Disease Study 2017. Lancet (2018) 392 1684-1735. DOI: 10.1016/S0140-6736(18)31891-9 2. Ofori-Asenso R, Chin KL, Curtis AJ, Zomer E, Zoungas S, Liew D. Recent patterns of multimorbidity among older adults in high-income countries. Popul Health Manag (2019) 22 127-137. DOI: 10.1089/POP.2018.0069/SUPPL_FILE/SUPP_TABLE4.PDF 3. Crimmins EM. Lifespan and healthspan: past, present, and promise. Gerontologist (2015) 55 901-911. DOI: 10.1093/GERONT/GNV130 4. McNeil JJ, Woods RL, Nelson MR. Effect of aspirin on disability-free survival in the healthy elderly. N Engl J Med (2018) 379 1499-1508. DOI: 10.1056/NEJMOA1800722/SUPPL_FILE/NEJMOA1800722_DATA-SHARING.PDF 5. Stenholm S, Head J, Kivimaki M. Smoking, physical inactivity and obesity as predictors of healthy and disease-free life expectancy between ages 50 and 75: a multicohort study. Int J Epidemiol (2016) 45 1260-1270. DOI: 10.1093/IJE/DYW126 6. Zaninotto P, Head J, Steptoe A. Behavioural risk factors and healthy life expectancy: evidence from two longitudinal studies of ageing in England and the US. Sci Rep (2020) 10 1-9. DOI: 10.1038/s41598-020-63843-6 7. Abdoli N, Salari N, Darvishi N. The global prevalence of major depressive disorder (MDD) among the elderly: a systematic review and meta-analysis. Neurosci Biobehav Rev (2021) 132 1067-1073. DOI: 10.1016/J.NEUBIOREV.2021.10.041 8. Wei J, Hou R, Zhang X. The association of late-life depression with all-cause and cardiovascular mortality among community-dwelling older adults: systematic review and meta-analysis. Br J Psychiatry J Ment Sci (2019) 215 449-455. DOI: 10.1192/BJP.2019.74 9. Nemeroff CB, Goldschmidt-Clermont PJ. Heartache and heartbreak—the link between depression and cardiovascular disease. Nat Rev Cardiol (2012) 9 526-539. DOI: 10.1038/nrcardio.2012.91 10. Diniz B, Butters M, Albert S, Dew M, Reynolds C. Late-life depression and risk of vascular dementia and Alzheimer’s disease: systematic review and meta-analysis of community-based cohort studies. Br J Psychiatry J Ment Sci (2013) 202 329-335. DOI: 10.1192/BJP.BP.112.118307 11. Lenze EJ, Rogers JC, Martire LM. The association of late-life depression and anxiety with physical disability: a review of the literature and prospectus for future research. Am J Geriatr Psychiatry (2001) 9 113-135. DOI: 10.1097/00019442-200105000-00004 12. Vaughan L, Corbin A, Goveas J. Depression and frailty in later life: a systematic review. Clin Interv Aging (2015) 10 1947-1958. DOI: 10.2147/CIA.S69632 13. Steensma C, Loukine L, Orpana H. Describing the population health burden of depression: health-adjusted life expectancy by depression status in Canada. Health Promot Chronic Dis Prev Can Res Pol Pract (2016) 36 205. DOI: 10.24095/HPCDP.36.10.01 14. Zaninotto P, Steptoe A. Association between subjective well-being and living longer without disability or illness. JAMA Netw Open (2019) 2 e196870. DOI: 10.1001/JAMANETWORKOPEN.2019.6870 15. Pérès K, Jagger C, Matthews F. Impact of late-life self-reported emotional problems on disability-free life expectancy: results from the MRC cognitive function and ageing study. Int J Geriatr Psychiatry (2008) 23 643-649. DOI: 10.1002/GPS.1955 16. Pongiglione B, de Stavola BL, Ploubidis GB. A systematic literature review of studies analyzing inequalities in health expectancy among the older population. PLoS One (2015) 10 e0130747. DOI: 10.1371/JOURNAL.PONE.0130747 17. Woods RL, Espinoza S, Thao LTP. Effect of aspirin on activities of daily living disability in community-dwelling older adults. J Gerontol Ser A Biol Sci Med Sci (2021) 76 2007-2014. DOI: 10.1093/GERONA/GLAA316 18. Teng E, Chui H. The Modified Mini-Mental State (3MS) examination. J Clin Psychiatry (1987) 48 314-318. PMID: 3611032 19. Rejeski W, Fielding R, Blair S. The lifestyle interventions and independence for elders (LIFE) pilot study: design and methods. Contemp Clin Trials (2005) 26 141-154. DOI: 10.1016/J.CCT.2004.12.005 20. Andresen E, Malmgren J, Carter W, Patrick D. Screening for depression in well older adults: evaluation of a short form of the CES-D (Center for Epidemiologic Studies Depression Scale). Am J Prev Med (1994) 10 77-84. PMID: 8037935 21. Park SH, Lee H. Is the center for epidemiologic studies depression scale as useful as the geriatric depression scale in screening for late-life depression? A systematic review. J Affect Disord (2021) 292 454-463. DOI: 10.1016/J.JAD.2021.05.120 22. Mohebbi M, Agustini B, Woods RL. Prevalence of depressive symptoms and its associated factors among healthy community-dwelling older adults living in Australia and the United States. Int J Geriatr Psychiatry (2019) 34 1208-1216. DOI: 10.1002/GPS.5119 23. Agustini B, Lotfaliany M, Mohebbi M. Trajectories of depressive symptoms in older adults and associated health outcomes. Nature Aging (2022) 2 295-302 24. 24.APA. (1994). Diagnostic and Statistical Manual of Mental Disorders.. Diagnostic and Statistical Manual of Mental Disorders (1994) 25. Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc B Methodol (1995) 57 289-300. DOI: 10.2307/2346101 26. Agustini B, Lotfaliany M, Woods RL. Patterns of association between depressive symptoms and chronic medical morbidities in older adults. J Am Geriatr Soc (2020) 68 1834-1841. DOI: 10.1111/JGS.16468 27. Morris G, Puri BK, Walker AJ. Shared pathways for neuroprogression and somatoprogression in neuropsychiatric disorders. Neurosci Biobehav Rev (2019) 107 862-882. DOI: 10.1016/J.NEUBIOREV.2019.09.025 28. Yuen GS, Bhutani S, Lucas BJ. Apathy in late-life depression: common, persistent, and disabling. Am J Geriatr Psychiatry Off J Am Assoc Geriatr Psychiatry (2015) 23 488-494. DOI: 10.1016/J.JAGP.2014.06.005 29. Henstra MJ, Feenstra TC, van der Velde N. Apathy is associated with greater decline in subjective, but not in objective measures of physical functioning in older people without dementia. J Gerontol Ser A (2019) 74 254-260. DOI: 10.1093/GERONA/GLY014 30. Wiels W, Baeken C, Engelborghs S. Depressive symptoms in the elderly—an early symptom of dementia? A systematic review. Front Pharmacol (2020) 11 34. DOI: 10.3389/FPHAR.2020.00034/BIBTEX 31. Chang SC, Pan A, Kawachi I, Okereke OI. Risk factors for late-life depression: a prospective cohort study among older women. Prev Med (2016) 91 144-151. DOI: 10.1016/J.YPMED.2016.08.014
--- title: 'Cell shape characteristics of human skeletal muscle cells as a predictor of myogenic competency: A new paradigm towards precision cell therapy' authors: - Charlotte Desprez - Davide Danovi - Charles H Knowles - Richard M Day journal: Journal of Tissue Engineering year: 2023 pmcid: PMC10026113 doi: 10.1177/20417314221139794 license: CC BY 4.0 --- # Cell shape characteristics of human skeletal muscle cells as a predictor of myogenic competency: A new paradigm towards precision cell therapy ## Abstract Skeletal muscle-derived cells (SMDC) hold tremendous potential for replenishing dysfunctional muscle lost due to disease or trauma. Current therapeutic usage of SMDC relies on harvesting autologous cells from muscle biopsies that are subsequently expanded in vitro before re-implantation into the patient. Heterogeneity can arise from multiple factors including quality of the starting biopsy, age and comorbidity affecting the processed SMDC. Quality attributes intended for clinical use often focus on minimum levels of myogenic cell marker expression. Such approaches do not evaluate the likelihood of SMDC to differentiate and form myofibres when implanted in vivo, which ultimately determines the likelihood of muscle regeneration. Predicting the therapeutic potency of SMDC in vitro prior to implantation is key to developing successful therapeutics in regenerative medicine and reducing implementation costs. Here, we report on the development of a novel SMDC profiling tool to examine populations of cells in vitro derived from different donors. We developed an image-based pipeline to quantify morphological features and extracted cell shape descriptors. We investigated whether these could predict heterogeneity in the formation of myotubes and correlate with the myogenic fusion index. Several of the early cell shape characteristics were found to negatively correlate with the fusion index. These included total area occupied by cells, area shape, bounding box area, compactness, equivalent diameter, minimum ferret diameter, minor axis length and perimeter of SMDC at 24 h after initiating culture. The information extracted with our approach indicates live cell imaging can detect a range of cell phenotypes based on cell-shape alone and preserving cell integrity could be used to predict propensity to form myotubes in vitro and functional tissue in vivo. ## Introduction Cell-based therapy using autologous skeletal muscle derived cells (SMDC) has been technically feasible in principle for several decades. Clinical investigations conducted to date have attempted to treat a variety of conditions including muscular dystrophy, cardiac heart failure, and incontinence.1–6 Faecal incontinence (FI) is a condition that would particularly benefit from cell-based regenerative medicine to restore functional muscle. FI affects a large portion of the general population, with an estimated prevalence of ($2\%$–$15\%$).7 Sphincter muscle lesions are the most frequent pathophysiological alteration found in patients with this condition.8 Although sacral neuromodulation is currently the first-line therapy recommended after failure of conservative treatment,9 long-term efficacy at 10-years is achieved in less than half of patients.10 More definitive surgical repair of the sphincter muscle has also shown disappointing results at long-term follow-up.11 In terms of potential benefit, SMDC-based treatment of FI could deliver tremendous impact to the quality of life of a significant proportion of the general population affected by this common disorder. Autologous SMDC-based therapy for FI, consisting of a mixed population of cells (but predominantly myoblasts) isolated from a biopsy taken from skeletal muscle, is reported to be safe, with 1-year efficacy in up to $80\%$ of patients in open label studies3,5 and in $60\%$ of patients in the only randomised-controlled placebo trial available to date.6 While these results are encouraging, the level of efficacy reported in previous studies suggests that heterogeneity in the proposed therapy may contribute to inconsistent clinical outcomes. Improved methods for patient stratification to identify patients who are (un)likely to respond well to regenerative therapies, are needed to ensure the potential benefits of cell-based regenerative medicine can be fully exploited. To date, myogenic quality attributes of SMDC used in clinical studies for the treatment of FI have relied mainly on the use of flow cytometry analysis of the expression of the myogenic cell marker CD56 expressed throughout satellite cells and their descendants3,6,12,13 and muscle stem cell markers Pax7 and Myf5.14,15 Cell marker expression has been complemented with the use of ex vivo myotube formation assays to demonstrate competency necessary for differentiation and myofibre formation.15,16 More recently, the potency of SMDC has been correlated with the expression of CD56 and acetylcholinesterase activity (AChE),16 with AChE activity found to reflect in vitro differentiation of SMDC and the clinical potency of cells used for the treatment of FI. While these assays provide useful metrics for the batch release of SMDC products for clinical use, they involve lengthy protocols and result in the analysed cells being unusable for clinical delivery. Such approaches are not compatible with continuous monitoring of cell phenotypes in real-time, especially when limited quantities of autologous cells are available. None of the clinical or pre-clinical investigations to date exploring SMDC therapies for FI have directly evaluated the phenotypic attributes of the cells being transplanted on an individual basis using non-destructive methods that does not impinge the subsequent use of the analysed cells. Collecting this information during upstream bioprocessing and correlating it with clinical outcomes could provide invaluable information on target quality attributes during manufacture of the cell-based products. The true extent to which isolated autologous SMDC reflect phenotypic characteristics of the donor tissue and subsequent functionality following implantation has yet to be fully elucidated. Human SMDC expansion in vitro is affected by many factors. Firstly, not all of the cells isolated from the biopsy may be compatible with ex vivo culture, resulting in a selective population of cells that preferentially grow on 2D tissue culture vessels. The culture conditions, especially culture medium, as well as background (patho)physiological characteristics of the donor, such as co-morbidity, age and gender of the donor are also likely to influence the composition of the isolated cell population.17 The impact of age on functionality was reported for engineered muscle derived from donors divided into three different age groups, with cells from young female donors achieving the fastest-growing SMDC in vitro and optimum contractile output of the engineered construct.18 These findings may shed light on how well transplanted cells function following in vivo implantation of SMDC and warrant further consideration, given that musculoskeletal conditions, such as FI, are more prevalent in the older population but also affect a smaller, much younger female population with obstetric anal sphincter injuries. If phenotypic features can be identified from in vitro behaviour of cells, it is conceivable that new methods to stratify patients most likely to benefit from cell-based therapy could be devised that would avoid treating patients in whom the regenerative response is unlikely to restore muscle function. Other factors related to the cell supply chain and manufacturing may also influence the potency of the therapy. For example, how the cell product is handled immediately before transplantation into the recipient may subsequently affect how well cells perform in vivo. Thawing of cells at point-of-care immediately before delivery is a method widely used in cell therapy studies. However, in vitro studies with skeletal muscle and other cell types indicate that achieving cell recovery with sufficient quality after cryopreservation ideally should take into consideration the optimisation of the rate of freezing and thawing to prevent osmotic shock during thawing; the state of cell differentiation at the time of cryopreservation; and inclusion of post-thaw recovery interval involving in vitro culture to increase the likelihood of subsequent good cell viability, attachment and differentiation.19–22 We have recently described novel analysis tools using dynamic imaging of cells in culture to provide phenotypic information at the sub-cellular, single-cell and population level using high-content image analysis.23 We hypothesise that a similar approach can be adopted for the development and quality control of SMDC being used for therapeutic applications. To our knowledge, the use of non-destructive multi-parametric imaging-based phenotypic characterisation of SMDC has not been attempted to date as means to assess and quantify the myogenic potency of SMDC derived from different donors. The primary aim of the study was thus precisely to evaluate the applicability of multi-parametric imaging-based phenotypic characterisation to distinguish the myogenic potency of SMDC obtained from different donors. The secondary aim was to determine whether the approach could be applied to detect phenotypic differences in myogenic potential of freshly thawed cells compared with cells that were established in culture for a short period of time in order to simulate different approaches used to handle cells prior to transplantation. ## Cell culture and cryopreservation Commercially available human SMDC (Sk-1111) from 14 individual human donors were supplied by Cook MyoSite (Pennsylvania, USA). The cells were isolated from muscle biopsies (abdominal rectus or vastus lateralis) acquired from cadavers. Cells used in the experiments were derived from single donors and consisted of myoblast-like, non-differentiated primary human muscle cells characterised by the supplier via immunocytological analyses and/or flow cytometry of desmin and myosin heavy chain, with cells ⩾$70\%$ cells positive for desmin. The cells were cultured until Passage 2 in Ham’s F-10 Nutrient Mix (ThermoFisher Scientific) supplemented with $20\%$ foetal bovine serum (FBS; Life Technologies Limited), 0.1 mg/ml streptomycin and 0.25 μg/ml amphotericin B (Merck), 1 µM dexamethasone (Merck) and 10 ng/ml human fibroblast growth factor (FGF)-basic (PeproTech) (proliferation medium). The culture medium was changed every 2–3 days. For subculture and harvesting, cells were rinsed once for 1 min with Dulbecco’s phosphate buffered saline (PBS; Merck) and incubated in trypsin-ethylenediaminetetraacetic acid (EDTA) solution (Merck) for 5 min at 37°C. Cells were washed in proliferation medium and centrifugated at 1000×g for 5 min. The supernatant was discarded and the cell pellet resuspended in 3 mL of proliferation medium before distributing into fresh tissue culture flasks. For cryopreservation at passage 2, the cell pellet was resuspended in proliferation medium before dropwise addition of an equal volume of cryomedium ($20\%$ dimethyl sulphoxide, $40\%$ FBS, $40\%$ proliferation medium) and transferred into cryovials. The cell suspension was initially frozen to −80°C over a period of 24 h using a Mr Frosty™ freezing containing (ThermoFisher Scientific) before being transferred to the vapour phase of a liquid nitrogen cryostorage vessel. ## Comparison of SMDC growth Cryopreserved SMDC at Passage 2 were thawed at 37°C in a water bath. The cell suspension was then plated in proliferation medium under the following conditions: [1] 5000 cells/cm2 per well (9.07 cm²) in 6 well tissue culture plates (Sarstedt) or [2] 13,000 cells/cm2 in a 75 cm2 tissue culture flasks. The culture medium was changed every 2–3 days and the cells incubated at 37°C, $5\%$ CO2. Cells in the 6-well plate [1] were cultured until they reached $80\%$–$90\%$ confluence, at which point the culture medium was switched to ‘differentiation medium’ consisting of Dulbecco’s Modified Eagle Medium (DMEM)/F12 (ThermoFisher Scientific), $1\%$ insulin-transferrin-selenium solution (ThermoFisher Scientific), $1\%$ N2 supplement (ThermoFisher Scientific), 2 mM L-Glutamine (Merck) and 100 units penicillin and 0.1 mg streptomycin/mL (Merck). Cells in the 75 cm2 tissue culture flask [2] were cultured until they reached ~$75\%$ confluence and sub-cultured (Passage 3). The cells were replated at a density of 5000 cells/cm2 in a 6 well tissue culture plate (‘Passage 3 cells’) and cultured in proliferation medium until they reached $80\%$–$90\%$ confluence, followed by differentiation for 5 days in differentiation medium. In summary, ‘Passage 2 cells’ were considered as ‘freshly thawed cells’ and ‘Passage 3 cells’ were considered as ‘sub-cultured cells’. For each donor, cells cultured in 6 well tissue culture plates at Passage 2 (‘freshly thawed cells’) and Passage 3 (‘sub-cultured cells’) were used for imaging, myotube formation assays and fusion index, beta-galactosidase (β-Gal) and AChE activity assay studies. Cells at Passage 3 were used for confluence studies and flow cytometry analysis of cell markers. ## Time-lapse imaging of SMDC growth Image acquisition of SMDC growth was performed in an incubator at 37°C, $5\%$ CO2 using a CytoSMART Lux2 microscope (CytoSMART Technologies B.V., Netherlands) at 20X magnification. Images of the same field of view were acquired every 5 min over a period of 3 days during the growing stage for both Passage 2 and Passage 3 cells. Each 6 well tissue culture plate was placed in the same position on the Lux2 microscope to enable imaging of the central area of the well that contained at least 20 individual cells (area of 0.92 mm × 0.92 mm, 960 × 960 pixels). ## Image analysis pipeline CellProfiler™ software (www.cellprofiler.org) version 4.1.324 was used to create an image analysis pipeline that enabled the analysis of different characteristics of the cells’ shape (Table 1, Figure 1). To avoid inaccurate object identification within images, the number of objects was monitored for each image during the building of the pipeline. We first used the ‘EnhanceOrSuppressFeatures’ module with enhance neurites function to enhance the structure of the longitudinal cells and the ‘IdentifyPrimaryObjects’ module to identify the cells. For this module, the best results were obtained using 13 as minimal diameter of objects, the Otsu three classes automatic image thresholding method with pixels of the middle intensity belonging to the foreground, threshold smoothing scale of 1.3488 and without threshold correction. The lower bound of the threshold was usually 0.74 but could be adapted in case of variation of the illumination in the input image. The ‘SplitOrMergeObjects’ module was then used to merge objects within a distance of 4 pixels and the ‘FilterObjects’ module used to filter objects with major axis length under 15 pixels. Lastly, characteristics were measured using ‘MeasureObjectNeighbors’, ‘MeasureObjectSizeShape’ and ‘MeasureAreaOccupied’. The pipeline used can be found in Supporting Information Table S1. For tracking cells, a similar pipeline was built and the ‘TrackObjects’ module was used with the ‘follow neighbors’ function with an average cell diameter of 45.0 pixels and a maximal pixel distance to consider match of 70 (Supporting Information Table S1). Two hours of tracking was selected at 12 and 24 h of imaging to allow good accuracy of the pipeline to follow cells. For each characteristic studied, Konstanz Information Miner (KNIME; open-source data analytics software version 4.3.2) was used to calculate the mean and standard deviation (SD) of the characteristic for all the objects identified in one image. These data were then used for the correlation analyses. ## Confluence analysis The confluence study was conducted using a CytoSMART Omni microscope at 10X magnification (CytoSMART Technologies B.V., Netherlands). Briefly, cryopreserved cells at Passage 3 were thawed and plated in a 6 well tissue culture plate at a density of 5000 cells/cm2 for 5 days in proliferation medium. The plates were imaged for 5 days and growth medium was changed every 2 days. Confluence of cells for each well was automatically calculated by the Cytosmart Omni using full plate scanning and image stitching. ## Immunostaining for myogenic markers After 5 days of culture in differentiation medium, cells were fixed in $4\%$ formaldehyde for 10 min. After permeabilisation ($0.1\%$ Triton-X 100 in PBS) and blocking with $5\%$ goat serum (Merck), cells were incubated overnight with Alexa Fluor 647 conjugated anti-NCAM1 (CD56) IgG rabbit monoclonal antibody (1:500; ab237456, Abcam, United Kingdom) and MF-20 anti-sarcomeric myosin heavy chain IIa IgG mouse antibody supernatant (1:20; Developmental Studies Hydroma Bank, Iowa) in $5\%$ goat serum. Cells were then incubated in the dark with secondary antibody (1:350 Alexa Fluor 488 IgG goat anti-mouse in $5\%$ Goat serum). Finally, the nuclei of cells were counter stained with 4′,6-diamidino-2-phenylindole (DAPI). Immunofluorescence images were acquired with 3 channels (blue, green, red) using Leica Dmi8 microscope (Leica, United Kingdom) at 10X magnification and reconstructed using LAS X Life Science software (Leica). For each set of donor cells the navigation mode of the software was used to acquire one image at the centre of the well and four images at cardinal points from the centre of the well. The fusion index in myotube formation assays for each donor was calculated as the ratio of nuclei number in myocytes with two or more nuclei divided by the total number of nuclei, as previously described.25 For each donor, five images were acquired, with one image at the centre of the well and four images at cardinal points from the centre of the well. The fusion index was calculated for each image and was generated from at least 500 randomly chosen MHC-positive cells or myotubes. The mean fusion index for each donor was calculated from the fusion index of the five images. To analyse intracellular expression of NCAM1/CD56, a pipeline was created using CellProfiler™. The three colour channels (red, blue and green) for the acquired fluorescence microscopy images were separated using the ‘ColorToGray’ module and nuclei stained with DAPI were identified using the ‘IdentifyPrimaryObjects’ module on the blue channel. The distance between nuclei was then calculated using the ‘MeasureObjectsNeighbors’ module. The cytoplasmic intensity of NCAM1/CD56 expression was calculated for each cell using the ‘IdentifySecondaryObjects’ module and the ‘MeasureObjectIntensity’ module on the red channel. The mean NCAM1/CD56 intensity per nuclei was calculated for cultures from each donor. The correlation between the NCAM1/CD56 intensity by nucleus and the corresponding distance to the closest neighbour (=nucleus) was also calculated for each donor using Spearman correlation. ## Flow cytometry for myogenic markers Cells cryopreserved at Passage 3 from each donor were plated in proliferation medium in 6 well tissue culture plates at a density of 5000 cells/cm2 and cultured until they reached $80\%$–$90\%$ confluence. Cells were harvested from the wells, as previously detailed, centrifugated at 1000×g and fixed using $4\%$ formaldehyde. Cells were stained with FITC-conjugated anti-CD56 (Clone AF-7H3, Milteny Biotec), BV421-conjugated anti-CD34 (Clone 581, BD Biosciences) and APC-conjugated anti-CD90 (Clone 5E10, BD Biosciences) for 20 min in the dark at 4°C. Controls for non-specific staining were included consisting of isotype controls for each antibody (FITC-conjugated IS5-21F5 mouse IgG1 (Milteny Biotec, United Kingdom), BV421-conjugated anti-KLH (BD Biosciences) and APC-conjugated mouse IgG1 clone MOPC-21 (BD Biosciences). Cells were washed 2 times before the analysis. Each acquisition file included 10,000 events. A forward scatter (FSC) threshold was set to avoid debris from list mode data and for each sample. Flow cytometry analysis was conducted using BD LSRFortessa™ Cell Analyser (BD Biosciences) and BD FACSDiva™ Software for acquisition. Results were analysed using FlowJo™ version 10 software (BD Biosciences). ## SMDC senescence Senescence of the SMDC was assessed using the senescence β-Gal staining kit (Cell Signalling Technology). Briefly, cells were cultured in proliferation medium at a density of 5000 cells/cm2 in a 6 well tissue culture plate for 3 days and then fixed for 15 min. The β-Gal staining solution was added to each well and the plate incubated in the dark at 37°C overnight. To ensure consistency for the experiment between all donors, cells from all donors were assayed at the same time with pH of the β-Gal staining solution controlled before use (final pH of 6.0). Assessment of positive β-Gal staining in cells was determined by the presence of blue staining colour at 24 h. Microscopy images were acquired using a ZEISS Primovert optical microscope (Zeiss) at 10X magnification with 5 fields of view per well. The number of positive cells and the total number of cells were counted manually for each image. To facilitate identification of cell staining, images were opened using ImageJ software and the Colour Threshold function applied to provide better contrast of positively stained cells. The ratio of positively stained cells divided by the total number of cells per field of view was calculated for cells from each donor. ## Acetylcholinesterase assay Cells were plated at a density of 5000 cells/cm2 in 6 well tissue culture plates at Passage 2 and Passage 3 in proliferation medium until $80\%$–$90\%$ confluence was reached. The culture medium was switched to differentiation medium for a further 5 days culture. Analysis of AChE activity in cells from different donors was conducted using Amplite™ colorimetric AChE activity assay kit (AAT Bioquest®) according to manufacturer’s instructions. Briefly, 20X DTNB stock solution, 20X acetylthiocholine stock solution and AChE standard solution (50 U/mL) were prepared immediately before the experiment. The AChE working solution was prepared using 20X DNTB stock solution, 20X acetylthiocholine stock solution and assay buffer. The AChE standard solution at 50 U/mL was used to generate an AChE standard solution at 1000 mU/mL. The latter standard solution was then used to generate 7 serially diluted AChE standards (ranging from 1000 to 1.37 mU/mL) by performing 1:3 serial dilution in Assay buffer. The differentiation medium was carefully transferred from each well into a 96 well plate (50 μL per well) with one duplicate for each sample. The assay buffer solution alone was used as blank control. 50 μL of AChE working solution was added in each well of AChE standards, blank control, and test samples to make the total AChE assay volume of 100 μL/well. The reaction was incubated in the dark for 30 min at room temperature followed by the OD measurement at 405 nm on a Multiskan FC microplate absorbance reader (Thermo Scientific). For analysis, the reading obtained from the blank control well was used as negative control and this value was subtracted from the other readings to obtain the baseline corrected values. The OD measurements from the standard samples were plotted to obtain a standard curve from which AChE concentration (mU/mL) for each sample was interpolated using GraphPad Prism version 9 software. ## Statistical analyses Mann-Whitney test for continuous variables and Chi-square test for categorical data were used for comparison among groups. Wilcoxon test was used to compare continuous paired variables. Correlation studies used the Spearman test with r < −0.5 or >+0.5 indicating correlation. Results are expressed as mean ± standard deviation. Data were analysed using GraphPad Prism version 9 for Windows (GraphPad Software Inc). ## Donor demographics for human SMDC Cells isolated from 14 donors were studied and the characteristics of each donor (M/F: $\frac{5}{9}$, mean age: 46.29 ± 22.04 years) are summarised in Table 2. No difference was seen between male and female donors regarding age (35.60 ± 12.24 vs 52.22 ± 24.56; $$p \leq 0.24$$), BMI (27.60 ± 6.34 vs 33.44 ± 7.51; $$p \leq 0.15$$), diabetes (no diabetic donors included) and tobacco use ($60.00\%$ vs $55.56\%$; $$p \leq 0.99$$). Table 2. \| Age \| Sex \| Ethnicity \| Cook MyoSite catalogue number \| Tissue of origin \| Diabetes \| BMI \| Tobacco use \| Additional information \| Fusion Index \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| 37 \| M \| Caucasian \| 01242-37M \| AR \| 0 \| 38 \| 1 \| Non-IV drug use; Alcohol use \| 0.37 \| \| 18 \| M \| Caucasian \| 01236-18M \| AR \| 0 \| 21 \| 0 \| Active \| 0.73 \| \| 41 \| M \| Caucasian \| 01033-41M \| AR \| 0 \| 25 \| 1 \| Alcohol use \| 0.74 \| \| 32 \| F \| Caucasian \| 01034-32F \| AR \| 0 \| 42 \| 1 \| \| 0.72 \| \| 16 \| F \| Caucasian \| 01055-16F \| AR \| 0 \| 37 \| 0 \| Active \| 0.67 \| \| 36 \| F \| Caucasian \| 01277-36F \| AR \| 0 \| 26 \| 1 \| Alcohol use \| 0.7 \| \| 52 \| F \| African American \| 01035-52F \| AR \| 0 \| 47 \| 1 \| Non-IV drug use \| 0.69 \| \| 51 \| M \| Caucasian \| 01267-51M \| AR \| 0 \| 26 \| 0 \| Non-IV drug use; Hypertension; High blood pressure; High cholesterol \| 0.82 \| \| 31 \| M \| Caucasian \| 01266-31M \| AR \| 0 \| 28 \| 1 \| Non-IV drug use; Alcohol use \| 0.75 \| \| 42 \| F \| Caucasian \| 01269-42F \| AR \| 0 \| 35 \| 1 \| \| 0.65 \| \| 49 \| F \| Caucasian \| P01288-49F \| AR \| 0 \| 30 \| 1 \| Pancreatitis; Non-IV drug use; Alcohol use, GERD \| 0.77 \| \| 70 \| F \| Caucasian \| P01402-70F \| VL \| 0 \| 26 \| 0 \| Hypertension; Osteoporosis; Asthma; Ulcerative colitis \| 0.49 \| \| 81 \| F \| Caucasian \| P01442-81F \| VL \| 0 \| 32 \| 0 \| Hypertension; Heart disease \| 0.68 \| \| 92 \| F \| Caucasian \| P01520-92F \| VL \| 0 \| 26 \| 0 \| Hypertension \| 0.0 \| ## Fusion index of SMDC correlates negatively with donor age The fusion index calculated at day 5 of differentiation at Passage 3 is given for each cell donor in Table 2 with a mean fusion index of 0.63 ± 0.21 among all donors. SMDC from the 92-year-old donor never reached more than $50\%$ of confluence. Therefore, the differentiation medium was not added and the fusion index was consequently considered as 0.00 for this donor. No difference was seen in terms of fusion index between male and female donors (0.69 ± 0.18 vs 0.60 ± 0.24; $$p \leq 0.36$$), suggesting gender did not influence the level of myogenicity observed. Alcohol use, tobacco use and non-IV drug use did not impact on the fusion index (with respectively $$p \leq 0.34$$, $$p \leq 0.28$$ and $$p \leq 0.71$$), indicating that lifestyle factors record for the cohort investigated were not confounding factors for myogenicity. The age of the donors negatively correlated with fusion index (r = −0.578; $$p \leq 0.03$$), indicating stage of life affects the propensity of the cells to form myotubes, but the BMI, a measurement of a person’s leanness or corpulence was not correlated with myogenicity ($r = 0.111$; $$p \leq 0.71$$). The total number of cells and the viability of the cells at the time of recultivation of the cells did not correlate with the fusion index (with respectively $r = 0.384$; $$p \leq 0.22$$ and $r = 0.124$; $$p \leq 0.70$$). ## Fusion index correlates negatively with CD34 expression but not with CD56 and CD90 expression Flow cytometry revealed the mean % expression of CD56, a marker widely used for identification of human myogenic cells, across the different donors was 55.8 ± 16.76, whereas the mean % expression of CD34 was 9.98 ± 13.11 and CD90 was 94.88 ± $8.29\%$ across the different donors (Figure 2). CD34 expression negatively correlated with the fusion index with r = −0.763 and $$p \leq 0.002$$, contrary to CD56 and CD90 expression ($r = 0.131$; $$p \leq 0.67$$ and r = −0.065; $$p \leq 0.83$$, respectively). CD34 expression did not correlate with neither CD56 expression (r = −0.250; $$p \leq 0.41$$) nor CD90 expression ($r = 0.190$; $$p \leq 0.54$$), but CD56 expression did correlate with CD90 expression ($r = 0.610$; $$p \leq 0.03$$). Age did not appear to be a confounding factor for myogenicity. The donor with the highest CD34 expression ($43.70\%$) was the 92-year-old woman. However, no correlation was found between age and CD56, CD34 or CD90 expression (with respectively $$p \leq 0.432$$, $$p \leq 0.123$$ and $$p \leq 0.292$$). Lastly, the majority of CD34+ cells expressed CD56, with a mean of 0.83 ± $0.19\%$ of CD34+CD56+ cells among CD34+ cells but no correlation was found between the percentage of CD34+CD56+ cells and fusion index (r = −0.192; $$p \leq 0.53$$). Figure 2.: Confluence study: (a) Image of one well using the Cytosmart Omni software to automatically calculate the confluence at hour 0, (b) image of one well using the Cytosmart Omni software to automatically calculate the confluence at hour 120, and (c) graphical representation of the evolution of the confluence during time for each cell line (Bar 1 cm). ## Early shape characteristics correlate negatively with the fusion index The results for the correlation between each early shape characteristics of the cells (at 12 and 24 h of imaging during the growing phase) and the fusion index at day 5 of differentiation after Passage 3 are presented in Table 3. At 12 h of imaging, 5 cell shape characteristics were negatively correlated with the fusion index, including total area occupied by cells (r = −0.815, $$p \leq 0.001$$), area shape (r = −0.631; $$p \leq 0.028$$), bounding box area (r = −0.576; $$p \leq 0.049$$), minimum ferret diameter (r = −0.589; $$p \leq 0.044$$) and minor axis length (r = −0.614; $$p \leq 0.034$$), indicating the presence of cells exhibiting these shapes was associated with subsequent poorer myogenicity. Likewise, at 24 h of imaging, 8 cell shape characteristics were negatively correlated with the fusion index, including total area occupied by cells (r = −0.686; $$p \leq 0.007$$), area shape (r = −0.709; $$p \leq 0.005$$), bounding box area (r = −0.563; $$p \leq 0.036$$), compactness (r = −0.534; $$p \leq 0.049$$), equivalent diameter (r = −0.576; $$p \leq 0.031$$), minimum ferret diameter (r = −0.620; $$p \leq 0.018$$), minor axis length (r = −0.590; $$p \leq 0.026$$) and perimeter (r = −0.654; $$p \leq 0.011$$). There was a high correlation between all these characteristics at 24 h of imaging (Supporting Information Table S2). Regarding the confluence study after Passage 3 (Figure 3), no correlation was found between the confluence of the cells during the growing stage and the fusion index at 5 days of differentiation at hour 0 ($r = 0.826$, $$p \leq 0.068$$), 24 h ($r = 0.160$; $$p \leq 0.414$$), 48 h ($r = 0.314$; $$p \leq 0.303$$), 72 h ($r = 0.252$; $r = 0.342$), 96 h ($r = 0.303$; $$p \leq 0.310$$) and 120 h ($r = 0.270$; $$p \leq 0.330$$). Lastly, the distance reached by the cells using the tracking function was not correlated with the fusion index neither at 12 h ($r = 0.070$; $$p \leq 0.812$$) nor at 24 h of imaging ($r = 0.231$; $$p \leq 0.472$$). ## Intracellular CD56 intensity per nuclei does not correlate with the fusion index but a significant correlation exists between the intracellular CD56 intensity and the distance between nuclei CD56 is a reliable marker for myoblasts among cultured cells from skeletal muscle and it was reasoned that its level of expression could reflect the fusion index in cells derived from different donors. However, the mean intracellular CD56 intensity per nuclei did not correlate with the fusion index after 5 days of differentiation ($r = 0.083$; $$p \leq 0.786$$) and the range of CD56 intracellular staining intensity between donors was small (Figure 4). The closest distance between nuclei was negatively correlated to the fusion index, as expected (r = −0.815; $$p \leq 0.001$$). The correlation between the mean intracellular CD56 intensity and the closest distance between nuclei was calculated for cells from each donor (Table 4) and showed a statistically significant but poor correlation for all donors with the exception of the 92-year-old subject. Figure 4.: CD56 intracellular expression: (a) image taken with inverted microscope at 10X magnification with DAPI (left) and DAPI + CD56 antibody (right) validating the correlation between the distance between nuclei and the CD56 intensity (Bar 100 µm), (b) graphical representation of the mean CD56 intensity per nuclei per donor, and (c) graphical representation of the mean closest distance between nuclei for each donor. TABLE_PLACEHOLDER:Table 4. ## β-Gal expression between donors correlates with age and CD34 expression but is negatively correlated with fusion index Detection of β-Gal expression can be used to identify senescent cells in heterogeneous cell populations. The mean percentage of cells expressing β-Gal from different donors was 17.95 ± $22.26\%$ at Passage 3 (Figure 5). The fusion index after 5 days of differentiation was negatively correlated with the percentage of cells expressing β-Gal (Figure 5(c); r = −0.695; $$p \leq 0.006$$). A positive correlation was found between the percentage of cells expressing β-Gal and the age of the donor (Figure 5(d); $r = 0.808$; $$p \leq 0.0005$$). No correlation was found between BMI of donors and the percentage of cells expressing β-Gal (r = −0.114; $$p \leq 0.698$$). The percentage of cells expressing β-Gal was significantly lower in male donors in comparison to female donors (5.34 ± $0.03\%$ vs 24.96 ± $25.38\%$; $$p \leq 0.007$$). No difference was seen when comparing the percentage of cells expressing β-Gal in donors with alcohol use and in those without ($$p \leq 0.282$$), the same for tobacco use ($$p \leq 0.298$$) or for non-IV drug use ($$p \leq 0.240$$). Lastly, a positive correlation was found between the percentage of CD34+ SMDC and the percentage of cells expressing β-Gal among donors ($r = 0.797$; $$p \leq 0.001$$), but not with the percentage of CD56+ (r = −0.143; $$p \leq 0.640$$) or CD90+ ($r = 0.166$; $$p \leq 0.588$$) SMDC. Figure 5.: β-Gal expression. (a) Image of growing SkMDC cells from the 92-year-old donor taken with an optical microscope at 10X magnification. The strong blue colouration shows a high expression of β-Gal by 80.52% of cells (left image) and the same image after use of ImageJ software to identify the blue staining using the Colour threshold function. (b) Image of growing SkMDC cells from the 41-year-old donor taken with an optical microscope at 10X magnification. The strong blue colouration shows expression of β-Gal by 4.24% of cells (left image) and the same image used with ImageJ software to identify the blue colouration using the Colour threshold function (black arrow heads). (Bar 100 µm) (c) Graphical representation of the B-Gal expression in function of the fusion index after 5 days of differentiation for each cell line showing a significant negative correlation (r = −0.695; p = 0.006). (d) Graphical representation of the B-Gal expression in function of the age of the donor showing a significant positive correlation (r = −0.695; p = 0.006). ## Acetylcholinesterase activity does not correlate with fusion index AChE activity was evaluated in cells from 11 donors. The mean AChE activity after 5 days of differentiation among cell lines was 10.12 ± 5.83 mU/mL. AChE activity did not correlate with age (r = −0.243; $$p \leq 0.059$$) or BMI (r = −0.340; $$p \leq 0.096$$) and no difference was found for AChE activity with use of alcohol ($$p \leq 0.429$$), tobacco ($$p \leq 0.92$$) or non-iv drug use ($$p \leq 0.083$$). Although AChE has previously been suggested to correlate with clinical potency of SMDC, no correlation was found between AChE activity and the fusion index after 5 days of differentiation ($r = 0.2859$; $$p \leq 0.394$$). The percentage of CD56+ ($r = 0.180$; $$p \leq 0.596$$), CD34+ ($r = 0.031$; $$p \leq 0.927$$) and CD90+ ($r = 0.495$; $$p \leq 0.122$$) growing SMDC did not correlate with the AChE activity. There was a tendency of a negative correlation with the β-Gal expression with r = −0.572 IC$95\%$ [−0.873 to 0.042] ($$p \leq 0.327$$). ## β-Gal and CD34 expression correlate with surface area and size of cells A supplementary analysis was performed to see if any cell shape characteristics correlate with the other markers studied (β-Gal, CD56, CD34, CD90 and AChE activity) is reported in Table 5. Among all the cell shape characteristics, the area shape ($r = 0.645$; $$p \leq 0.012$$), the compactness ($r = 0.698$; $$p \leq 0.005$$), the form factor (r = −0.548; $$p \leq 0.047$$), the minor axis length ($r = 0.646$; $$p \leq 0.013$$) and the perimeter ($r = 0.687$; $$p \leq 0.007$$) were correlated with β-Gal expression. The same characteristics also correlated with the CD34 (Table 6), but also the major axis length ($r = 0.624$; $$p \leq 0.017$$). No characteristic correlated with the CD56 and the CD90 expression and only the form factor correlated with the AChE activity ($r = 0.651$; $$p \leq 0.030$$; Table 6). ## Fusion index does not differ between freshly thawed cells and cells already established in culture but freshly thawed cells take longer to reach confluence The delay to reach ~$80\%$ confluence during the cell expansion phase (necessary before adding differentiation medium) was greater in the freshly thawed cells than in sub-cultured cells (7.91 ± 1.70days vs 6.35 ± 1.36 days; $$p \leq 0.008$$; Figure 6). SMDC from the 92-year-old donor never reached more than $50\%$ of confluence in proliferation medium, so this donor was excluded from the analysis. The mean fusion index at day 5 of differentiation was 0.67 ± 0.23 for the freshly thawed cells and 0.66 ± 0.23 for subcultured cells ($$p \leq 0.28$$; Figure 6). Figure 6.: Comparison of fusion index at day 5 between defrosted cells and sub-cultured cells. (a) Comparison of the delay to reach 80% of confluence between defrosted cells (P2) and subcultured cells (Passage 3). The delay was higher in defrosted cells than in subcultured cells with p = 0.008. (b) Comparison of the fusion index at 5 days of differentiation between defrosted cells (Passage 2) and subcultured cells (Passage 3). No difference was observed between the 2 groups (p = 0.28). (c) Image taken with inverted microscope at 10X magnification and with blue and green filters to identify DAPI and myosin heavy chain intracellular expression of defrosted cells from the 32-year-old woman donor. The fusion index was 0.72. (d) Image taken with inverted microscope at 10X magnification and with blue and green filters to identify DAPI and myosin heavy chain intracellular expression of subcultured cells of the 32-year-old woman donor. The fusion index was 0.73. (Bar 100 µm). ## Cell shape characteristics differ between freshly thawed cells and sub-cultured cells As the pipeline was efficient in identifying cells that do not overlap, the number of cells identified from frame to frame was calculated during the 3 days of imaging for the freshly thawed cells (Passage 2) and sub-cultured cells (Passage 3). The time to double the initial number of objects by frame was 44.90 ± 17.42 h for sub-cultured cells versus 65.41 ± 17.40 h for freshly thawed cells ($p \leq 0.001$), showing that cells grow and overlap faster in sub-cultured cells. Moreover, no correlation was found between the cell shape characteristics of the freshly thawed cells at both 12 and 24 h of imaging and the fusion index at day 5 of differentiation (Table 3). Another analysis showed that almost all the cell shape characteristics differed between freshly thawed cells and sub-cultured cells at both 12 and 24 h of imaging (Supporting Information S4 Appendix), with the exception of the minor axis length, mean and maximum radius. At 24 h of imaging, the freshly thawed cells seemed to be smaller (in both length and width), more circular and less close to each other than the sub-cultured cells. Twelve of the 19 cell shape characteristics of the freshly thawed cells at 24 h of imaging were correlated with those of the sub-cultured cells at the same time of imaging (Supporting Information Table S3). Lastly, the distance reached by cells was higher in sub-cultured cells than in freshly thawed cells at both 12 and 24 h of imaging (with respectively $p \leq 0.001$ and $p \leq 0.001$; Supporting Information S4 Appendix). AChE activity in SMDC after 5 days of differentiation did not differ between freshly thawed cells and sub-cultured cells (11.36 ± 3.19 mU/mL in the freshly thawed cell group versus 10.12 ± 1.76 mU/mL in sub-cultured cell group ($$p \leq 0.99$$)). Lastly, no difference was seen regarding the mean percentage of cells expressing β-Gal between freshly thawed cells and sub-cultured cells after 5 days of differentiation, with respectively 0.20 ± 0.06 and 0.17 ± 0.06; $$p \leq 0.93.$$ ## Discussion The inability to predict potency of autologous cell-based products before they are implanted into patients hinders development of therapies and may prevent elucidation of the underlying mechanistic effects that can be attributed to clinical outcomes observed. The use of quality attributes for cell-based products that focus solely on the use of cell marker expression rather than biomarkers that reflect the ultimate intended functional properties of the resultant tissue is likely to be a contributing factor in the heterogenous outcomes of clinical studies. The inability to stratify patients into groups that identifies those likely to respond best is costly and can lead to the abandonment of potentially life-changing new treatments. This is particularly relevant for tissue replacement therapies, where the product being delivered comprises of parenchymal cells, such as autologous SMDC in the treatment of FI, that are intended to directly replace or restore definitive function of the tissue. In the present study, we have shown that the SMDC fusion index is negatively correlated to the age of the donor, which is in line with findings previously reported that separated patients among three age groups (20–39 years; 40–59 years and 60–80 years).18 This result is supported by the negative correlation between β-Gal expression and the fusion index at a cellular level, even at low passage. Indeed, the donor with the highest β-Gal expression (more than $80\%$ of cells) was the 92-year-old subject. We found no difference in terms of fusion index between sex, which differs from previous report,18 where young female donors were associated with fast-growing, functional cells. It is possible that our result could be explained by a lack of power, since only 5 of the 14 donors were men in the present study. Furthermore, the donors in our study were not strictly ‘healthy’ volunteers based on the reported abnormal BMI, alcohol, non-IV drug or tobacco use, which might affect the quality of the SMDC via oxidative modifications of proteins.26,27 Although many of these characteristics were correlated with fusion index outcomes, the results should be interpreted with caution. Several of the early cell shape characteristics were found to negatively correlate with the fusion index. These included total area occupied by cells, area shape, bounding box area, compactness, equivalent diameter, minimum ferret diameter, minor axis length and perimeter of SMDC at 24 h after initiating culture. The results indicate that monitoring of cell shape during the early stages of bioprocessing using real-time imaging could be used to predict cellular competency necessary for differentiation and myofibre formation in vivo, which in turn could help with selection of either patients to treat or cell populations more likely to yield better outcomes in cell-based therapy. Confirmation of the ability of cell shape characteristics in vitro to predict clinical efficacy requires further pre-clinical in vivo exploration that will enable testing of functional outcomes, such as cell engraftment and force generation in treated muscle following transplantation of cells. Monitoring of myogenicity markers in vivo alongside to cell shape characteristics would also be interesting to confirm the heterogeneity of cells population. The methodology used in the current study is rapid and non-destructive, relying on the shape of the cells in images acquired using bright field microscopy. To our knowledge, no similar study on human cells have been performed to date, whereas one used an immortalised mouse cell line (C2C12) to predict myotube formation.28 Consistent with our study, results indicated the width, length and perimeter of cells being important parameters during monitoring that could distinguish between different experimental conditions. However, in the current study these characteristics were positively correlated with β-Gal expression. Senescent cells are known to increase in size, which is in line with our results.29 We did not find any correlation between the CD56+ membrane expression of SMDC and the fusion index. This observation also coincided with a lack of correlation between cell shape characteristics and CD56 expression. The mean level of CD56+ expression was 55.8 ± $16.76\%$ in the present study, in line with previous reports,30,31 where the expression was between $50\%$ and $80\%$ of the total cell population. When separating myoblasts on basis of CD56 expression, CD56+ myoblasts were previously shown to better fuse into myotubes in comparison to CD56 negative cells, where only a low number of myotubes were observed.30 Unlike previous studies, the current study did not separate cells according to CD56 expression prior to conducting the fusion index, therefore the mixed population of cells investigated might have contributed to the lack of correlation observed. The brightfield microscopy imaging system used to acquire cell images in the current study was not compatible with cell staining to distinguish myogenic cells in the mixed population. Future studies are planned to incorporate this feature so that shape characteristics in the context of myogenic cell population can be distinguished. Furthermore, CD34 membrane expression correlated negatively with the fusion index in our study. Few studies have assessed CD34 expression in human skeletal myogenesis. One report found that the CD34 expression had no impact on the myogenic potential of human myoblasts in cell populations containing a similar percentage of CD56+ cells.32 In the same study, loss of CD34 expression was associated with loss of adipogenic potential of cells. Another report found that high expression of CD34 in stem cells was associated with stemness properties and that low CD34 expression was more associated with myogenic differentiation.33 It is possible that CD34+ cells are less engaged in myogenic differentiation than the CD34-cells and could stay in an inactivated quiescent state, explaining the inverse correlation with the fusion index.34 Lastly, we found no correlation between the fusion index and the AChE activity in the present study. AChE is a type B carboxylesterase found in skeletal muscle that is primarily active in cholinergic synapses and neuromuscular junctions.35–38 Previously, AChE activity has been reported to be low in proliferating myoblasts but increased following differentiation and fusion of SMDC.39,40 Our findings contrasted the prior studies, where the AChE activity was correlated to the fusion index.16 The reason for this is unclear and requires further investigation. The cells used in the current study and previous study by Thurner et al. are primary cultures derived from different commercial sources, which is likely to give rise to different confounding factors associated with cell donor demographics as well as differences in the way cells have been processed. A limitation of our study could be that the 92-year-old donor was excluded from this analysis as we did not manage to grow enough cells for this experiment. Secondly, the density of cells used for our experiments was lower (48,000 cells per well, 6-well plate) compared with density used in the latter study (120,000 cells per well, 24-well plate).16 Differences in shape characteristics were observed between freshly thawed and sub-cultured cells. We observed that the distance reached by the cells and almost all the cells shape characteristics were different between our two groups. This difference may be due to the effect of freezing the cells but did not impact the differentiation potential once growth was established, since there was no observable difference in the fusion index. Neither was there a difference in AChE activity in SMDC between freshly thawed cells and sub-cultured cells. The impact of subculturing cells beyond P3 was not investigated in the current study but the impact of higher passaging on cell shape descriptors in relation to the onset of cell senescence would be worthy of further exploration. This finding is in accordance with previous reports that compared fresh cells with frozen SMDC.31,35 Indeed, primary myoblast cultures can be effectively established from tissue cryopreserved for various lengths of time.41 ## Conclusion Results from the current study indicate cell shape descriptors applied to images non-destructively acquired during cell culture could be used as markers to predict potency of SMDC in replenishing muscle function prior to implantation. Cell-based therapies are being explored for a number of muscle disorders, including cardiac disease, incontinence, dystrophies and volumetric muscle loss associated with trauma. It is reasonable to predict that similar non-destructive imaging-based approaches could be applied to isolated cells intended to treat these conditions following identification cell-shape descriptors that correlate with established markers of potency. The method is simple and low-cost, enabling it to be incorporated into existing bioprocessing regimes. The findings present a novel approach and further studies conducted in conjunction with clinical investigations exploring SMDC-based therapies could establish whether selection of either patients or cell populations on this basis have a higher probability of yielding better outcomes in cell-based therapies. ## References 1. Dib N, Michler RE, Pagani FD. Safety and feasibility of autologous myoblast transplantation in patients with ischemic cardiomyopathy: four-year follow-up. Circulation (2005) 112 1748-1755. PMID: 16172284 2. Yamanaka Y, Takenaka N, Sakurai H. Human skeletal muscle cells derived from the orbicularis oculi have regenerative capacity for Duchenne muscular dystrophy. Int J Mol Sci (2019) 20 3456. PMID: 31337111 3. Frudinger A, Kölle D, Schwaiger W. Muscle-derived cell injection to treat anal incontinence due to obstetric trauma: pilot study with 1 year follow-up. Gut (2010) 59 55-61. PMID: 19875391 4. Frudinger A, Pfeifer J, Paede J. Autologous skeletal-muscle-derived cell injection for anal incontinence due to obstetric trauma: a 5-year follow-up of an initial study of 10 patients. Colorectal Dis (2015) 17 794-801. PMID: 25773013 5. Romaniszyn M, Rozwadowska N, Malcher A. Implantation of autologous muscle-derived stem cells in treatment of fecal incontinence: results of an experimental pilot study. Tech Coloproctol (2015) 19 685-696. PMID: 26266767 6. Boyer O, Bridoux V, Giverne C. Autologous myoblasts for the treatment of fecal incontinence: results of a Phase 2 randomized placebo-controlled study (MIAS). Ann Surg (2018) 267 443-450. PMID: 28426476 7. Sharma A, Yuan L, Marshall RJ. Systematic review of the prevalence of faecal incontinence. Br J Surg (2016) 103 1589-1597. PMID: 27704537 8. Damon H, Schott AM, Barth X. Clinical characteristics and quality of life in a cohort of 621 patients with faecal incontinence. Int J Colorectal Dis (2008) 23 845-851. PMID: 18506453 9. Abrams P, Andersson KE, Birder L. The Members of the Committees. Fourth International Consultation on Incontinence Recommendations of the International Scientific Committee: Evaluation and treatment of urinary incontinence, pelvic organ prolapse, and fecal incontinence. Neurourol Urodyn (2010) 29 213-240. PMID: 20025020 10. Desprez C, Damon H, Meurette G. Ten-year evaluation of a large retrospective cohort treated by sacral nerve modulation for fecal incontinence: Results of a French Multicenter Study. Ann Surg (2022) 275 735-742. PMID: 32740249 11. Malouf AJ, Norton CS, Engel AF. Long-term results of overlapping anterior anal-sphincter repair for obstetric trauma. Lancet (2000) 355 260-265. PMID: 10675072 12. Xu X, Wilschut KJ, Kouklis G. Human satellite cell transplantation and regeneration from diverse skeletal muscles. Stem Cell Reports (2015) 5 419-434. PMID: 26352798 13. Capkovic KL, Stevenson S, Johnson MC. Neural cell adhesion molecule (NCAM) marks adult myogenic cells committed to differentiation. Exp Cell Res (2008) 314 1553-1565. PMID: 18308302 14. Sacco A, Doyonnas R, Kraft P. Self-renewal and expansion of single transplanted muscle stem cells. Nature (2008) 456 502-506. PMID: 18806774 15. Frudinger A, Marksteiner R, Pfeifer J. Skeletal muscle-derived cell implantation for the treatment of sphincter-related faecal incontinence. Stem Cell Res Ther (2018) 9 233. PMID: 30213273 16. Thurner M, Asim F, Garczarczyk-Asim D. Development of an in vitro potency assay for human skeletal muscle derived cells. PLoS One (2018) 13 17. Riddle ES, Bender EL, Thalacker-Mercer AE. Expansion capacity of human muscle progenitor cells differs by age, sex, and metabolic fuel preference. Am J Physiol Cell Physiol (2018) 315 C643-C652. PMID: 30110562 18. Stölting MN, Hefermehl LJ, Tremp M. The role of donor age and gender in the success of human muscle precursor cell transplantation. J Tissue Eng Regen Med (2017) 11 447-458. PMID: 25052735 19. Uhrig M, Ezquer F, Ezquer M. Improving cell recovery: freezing and thawing optimization of induced pluripotent stem cells. Cells (2022) 11 799. PMID: 35269421 20. Bahsoun S, Coopman K, Akam EC. Quantitative assessment of the impact of cryopreservation on human bone marrow-derived mesenchymal stem cells: up to 24 h post-thaw and beyond. Stem Cell Res Ther (2020) 11 540. PMID: 33317625 21. Baboo J, Kilbride P, Delahaye M. The impact of varying cooling and thawing rates on the quality of cryopreserved human peripheral blood T cells. Sci Rep (2019) 9 3417. PMID: 30833714 22. Grant L, Raman R, Cvetkovic C. Long-term cryopreservation and revival of tissue-engineered skeletal muscle. Tissue Eng Part A (2019) 25 1023-1036. PMID: 30412045 23. Chesnais F, Hue J, Roy E. High-content image analysis to study phenotypic heterogeneity in endothelial cell monolayers. J Cell Sci (2022) 135 24. Carpenter AE, Jones TR, Lamprecht MR. CellProfiler: image analysis software for identifying and quantifying cell phenotypes. Genome Biol (2006) 7 25. Sun Y, Ge Y, Drnevich J. Mammalian target of rapamycin regulates miRNA-1 and follistatin in skeletal myogenesis. J Cell Biol (2010) 189 1157-1169. PMID: 20566686 26. Sumanasekera WK, Dao HT, Shekhovtsova V. The mechanistic role of oxidative stress in cigarette smoke-induced cardiac stem cell dysfunction and prevention by ascorbic acid. Cell Biol Toxicol (2019) 35 111-127. PMID: 30006751 27. Geiger AE, Daughtry MR, Yen CN. Dual effects of obesity on satellite cells and muscle regeneration. Physiol Rep (2020) 8 28. Ishikawa K, Yoshida K, Kanie K. Morphology-based analysis of myoblasts for prediction of myotube Formation. SLAS Discov (2019) 24 47-56. PMID: 30102873 29. Payea MJ, Anerillas C, Tharakan R. Translational control during cellular senescence. Mol Cell Biol (2021) 41 30. Stockholm D, Edom-Vovard F, Coutant S. Bistable cell fate specification as a result of stochastic fluctuations and collective spatial cell behaviour. PLoS One (2010) 5 31. Tabakov VY, Zinov’eva OE, Voskresenskaya ON. Isolation and characterization of human myoblast culture in vitro for technologies of cell and gene therapy of skeletal muscle pathologies. Bull Exp Biol Med (2018) 164 536-542. PMID: 29504093 32. Pisani DF, Dechesne CA, Sacconi S. Isolation of a highly myogenic CD34-negative subset of human skeletal muscle cells free of adipogenic potential. Stem Cells (2010) 28 753-764. PMID: 20135684 33. García-Prat L, Perdiguero E, Alonso-Martín S. FoxO maintains a genuine muscle stem-cell quiescent state until geriatric age. Nat Cell Biol (2020) 22 1307-1318. PMID: 33106654 34. Beauchamp JR, Heslop L, Yu DS. Expression of CD34 and Myf5 defines the majority of quiescent adult skeletal muscle satellite cells. J Cell Biol (2000) 151 1221-1234. PMID: 11121437 35. Fambrough DM, Engel AG, Rosenberry TL. Acetylcholinesterase of human erythrocytes and neuromuscular junctions: homologies revealed by monoclonal antibodies. Proc Natl Acad Sci U S A (1982) 79 1078-1082. PMID: 6175961 36. Krejci E, Martinez-Pena y, Valenzuela I, Ameziane R. Acetylcholinesterase dynamics at the neuromuscular junction of live animals. J Biol Chem (2006) 281 10347-10354. PMID: 16455662 37. Singh M, Kaur M, Kukreja H. Acetylcholinesterase inhibitors as Alzheimer therapy: from nerve toxins to neuroprotection. Eur J Med Chem (2013) 70 165-188. PMID: 24148993 38. Zhang X-J, Greenberg DS. Acetylcholinesterase involvement in apoptosis. Front Mol Neurosci (2012) 5 40. PMID: 22514517 39. Barjot C, Jbilo O, Chatonnet A. Expression of acetylcholinesterase gene during in vitro differentiation of rabbit muscle satellite cells. Neuromuscul Disord (1993) 3 443-446. PMID: 8186690 40. Siow NL, Choi RC, Cheng AW. A cyclic AMP-dependent pathway regulates the expression of acetylcholinesterase during myogenic differentiation of C2C12 cells. J Biol Chem (2002) 277 36129-36136. PMID: 12140295 41. Balci-Hayta B, Bekircan-Kurt CE, Aksu E. Establishment of primary myoblast cell cultures from cryopreserved skeletal muscle biopsies to serve as a tool in related research & development studies. J Neurol Sci (2018) 393 100-104. PMID: 30153568
--- title: Age-related reference intervals for ambulatory electrocardiographic parameters in healthy individuals authors: - Kenichi Hashimoto - Naomi Harada - Motohiro Kimata - Yusuke Kawamura - Naoya Fujita - Akinori Sekizawa - Yosuke Ono - Yasuhiro Obuchi - Tadateru Takayama - Yuji Kasamaki - Yuji Tanaka journal: Frontiers in Cardiovascular Medicine year: 2023 pmcid: PMC10026132 doi: 10.3389/fcvm.2023.1099157 license: CC BY 4.0 --- # Age-related reference intervals for ambulatory electrocardiographic parameters in healthy individuals ## Abstract ### Background The advent of novel monitoring technologies has dramatically increased the use of ambulatory electrocardiography (AECG) devices. However, few studies have conducted detailed large-scale investigations on the incidence of arrhythmias over 24 h, especially ectopy, in healthy individuals over a wide age range. ### Objectives This study aimed to investigate the incidence of arrhythmias detected using AECG and associated factors, in healthy individuals, over a wide age range. ### Methods In this cross-sectional study, we performed AECG on 365 healthy volunteers (median [interquartile range]: 48 [36, 67], 20–89 years, 165 men) under free-running conditions for 24 h. Ultrasonic echocardiography and heart rate variability analysis were performed to explore the factors associated with the incidence of arrhythmias. ### Results The 97.5th percentile of single ventricular ectopy (VE) was 149/day, 254/day, and 1,682/day in the 20–39-, 40–59- and 60–89-year age groups, respectively; that of single supraventricular ectopy (SVE) was 131/day, 232/day, and 1,063/day, respectively. Multivariate analysis revealed that aging was the only independent significant factor influencing the frequency of VE (β = 0.207, $$P \leq 0.001$$). Age (β = 0.642, $P \leq 0.001$), body mass index (BMI) (β = −0.112, $$P \leq 0.009$$), and the root mean square of successive differences in RR intervals (β = 0.097, $$P \leq 0.035$$) were factors significantly associated with SVE frequency. ### Conclusions Age-specific reference intervals of VE and SVE in a large population of healthy participants over a wide age range were generated. VE and SVE increased with age; SVE was influenced by BMI and the aging-induced decrease in parasympathetic tone activity. ## Introduction The development of ambulatory electrocardiography (AECG) by Holter in 1957 enabled 24-h-ECG recording [1]. Since then, AECG has been widely used for detecting arrhythmic events in clinical settings. Recently, the use of AECG devices has dramatically increased, especially with the advent of novel monitoring technologies, such as patch-type, implantable, and smartwatch-type ECG devices (2–4). Thus, it is necessary to establish reference intervals for AECG parameters to guide interpretation and clinical care. It is well-known that the prevalence of arrhythmic events depends on age. However, few studies have focused on the reference values for the prevalence of arrhythmias in each generation (younger, middle-aged, older populations) over a wide age range among healthy individuals. Previous studies on this subject included small sample sizes or were limited to fewer age groups, such as young (20–39 years) (5–9), middle-aged (40–59 years) [7, 10, 11], or older-aged cohorts (over 60 years) (7, 11–16). Moreover, most of these studies were conducted 20–40 years prior. Lifestyle and average longevity have changed over the 21st century, and few studies have investigated the incidence of arrhythmia in a wide age range using AECG. Supraventricular ectopy (SVE) (incidence: $56\%$–$87\%$) is reportedly the most common arrhythmia type in healthy individuals detected using AECG, followed by ventricular ectopy (VE) (incidence: $46\%$–$69\%$) [17, 18]. Previous studies have stated that SVE or VE should not be treated if they are infrequent or not severe in the absence of structural heart disease [19]. However, recent studies have suggested that a higher frequency of ventricular extrasystole was associated with reversible cardiomyopathy [20], inducing a decreased left ventricular ejection fraction, increased chronic heart failure incidence, and a high mortality rate even in individuals without structural heart disease [21]. Moreover, a recent study reported that frequent excessive supraventricular activity was associated with a risk of atrial fibrillation (AF), stroke, and total mortality in apparently healthy individuals [22]. Therefore, establishing reference values of VE and SVE is of paramount importance. Furthermore, the factors influencing the incidence of VE and SVE are not fully understood. This cross-sectional study entailed AECG examination of healthy volunteers whose ages varied widely, from 20 to 89 years. This study aimed to investigate the incidence of bradyarrhythmia and tachyarrhythmia and establish age-related reference values for AECG parameters. Moreover, we explored the factors associated with these AECG parameters, including ultrasonic echocardiography (UCG) and autonomic nervous system activity parameters expressed as heart rate variability (HRV), which can influence the prevalence of ectopy. ## Study population We recruited healthy volunteers between April 2015 and March 2018 for this study. The inclusion criteria were as follows: individuals with no history of cardiovascular disease, respiratory disease, dyslipidemia, diabetes mellitus, chronic kidney disease, psychiatric disease, and autonomic nervous system disorders. Moreover, participants who underwent annual medical examinations within the past year without abnormal findings on chest radiographs and 12-lead ECG were also included. Night-shift workers and current smokers were excluded during the initial stage. A total of 400 participants, without structural heart disease, were initially included in this study. The study procedures included the following (in order): detailed medical history, general physical examination, systolic and diastolic blood pressure measurements, 12-lead standard ECG, ultrasonic echocardiology (UCG), and 24-h AECG. The recording time of AECG was stipulated to be more than 23 h/day. The exclusion criteria were as follows: participants with ST-T abnormalities on baseline 12-lead ECG, second- or third-degree atrio-ventricular (AV) block and left ventricular conduction block on standard 12-lead ECG, low ejection fraction (EF) (<$50\%$) with wall motion abnormalities, significant left atrial dilatation and/or left ventricular dilatation detected with echocardiography, ST-T changes of an ischemic nature during daily activity and/or ambulatory ECG monitoring, long QT interval (>500 ms) on baseline 12-lead ECG, family history of sudden cardiac death, and body mass index (BMI) over 30 kg/m2. Thirty-five participants [48 [36, 47]] were excluded from this study based on the above-mentioned exclusion criteria, while 365 participants were enrolled (Table 1). Most participants were healthy volunteers who were citizens of the Tokyo Metropolitan and Saitama Prefecture area, Japan. All volunteers provided written informed consent before participation. The study protocol conformed to the Declaration of Helsinki and was approved by the Medical Ethics Committee of the National Defense Medical College Hospital (approval no. 4645), Saitama, Japan, and Nihon University School of Medicine, Itabashi Hospital, Tokyo, Japan (approval no. MF 2208-0037). Table 1 \| Unnamed: 0 \| 20–39 years \| 40–59 years \| 60–89 years \| P-value \| Total \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| \| (n = 120) \| (n = 124) \| (n = 121) \| P-value \| (N = 365) \| \| Demographics \| \| \| \| \| \| \| Age \| 31 [25, 36] \| 47 [43, 53] \| 71 [66.8, 75] \| <0.001 \| 48 [36, 67] \| \| Men \| 61 \| 52 \| 53 \| \| 166 \| \| Height (cm) \| 164.4 ± 9.2 \| 162.8 ± 8.7 \| 158.7 ± 9.2 \| <0.001 \| 162.0 ± 9.3 \| \| Body weight (kg) \| 60.1 ± 10.8 \| 62.2 ± 12.0 \| 56.3 ± 9.9 \| <0.001 \| 59.6 ± 11.2 \| \| Body mass index (kg/m2) \| 22.1 ± 4.1 \| 23.3 ± 3.6 \| 22.2 ± 2.5 \| 0.016 \| 22.6 ± 3.5 \| \| Systolic blood pressure (mmHg) \| 115.5 ± 11.4 \| 126.2 ± 17.3 \| 129.4 ± 14.2 \| <0.001 \| 123.8 ± 15.7 \| \| Diastolic blood pressure (mmHg) \| 71.1 ± 10.4 \| 79.4 ± 13.5 \| 76.1 ± 10.9 \| <0.001 \| 75.6 ± 12.2 \| \| AECG \| \| \| \| \| \| \| Total beat/day \| 109,462.9 ± 12,341.5 \| 110,534.9 ± 11,045.2 \| 103,693.1 ± 11,055.1 \| <0.001 \| 107,957.0 ± 1,1845.9 \| \| Maximum heart rate/day \| 144.0 ± 18.0 \| 133.3 ± 14.7 \| 121.1 ± 15.1 \| <0.001 \| 131.0 [120.0, 143.0] \| \| Minimum heart rate/day \| 50.6 ± 6.9 \| 53.9 ± 6.2 \| 53.0 ± 6.0 \| <0.001 \| 52.5 ± 6.5 \| \| Mean heart rate/day \| 78.9 ± 8.7 \| 78.8 ± 8.0 \| 73.8 ± 8.1 \| <0.001 \| 77.2 ± 8.6 \| \| Ventricular ectopy (single)/day \| 1.0 [0, 3.0] \| 2.0 [0, 6.0] \| 4.0 [0, 13] \| <0.001 \| 2.0 [0, 7] \| \| Supra ventricular ectopy (single)/day \| 6.0 [2.0, 14] \| 13.0 [6.0, 30] \| 67.0 [30.0, 189] \| <0.0001 \| 18.0 [5.0, 51] \| \| HRV \| \| \| \| \| \| \| SDNN (ms) \| 154.7 [129.4, 186.3] \| 136.2 [115.8, 158.1] \| 129.9 [109.7, 153.2] \| <0.001 \| 139.5 [117.1, 165.6] \| \| RMSSD (ms) \| 35.4 [27.1, 50.3] \| 25.0 [18.6, 32.5] \| 23.3 [17.1, 31.8] \| <0.001 \| 27.2 [20.6, 36.7] \| \| pNN50 (%) \| 11.2 [5.2, 20.7] \| 4.0 [1.3, 9.0] \| 2.8 [0.8, 7.0] \| <0.001 \| 5.4 [1.9, 11.2] \| \| SDANN (ms) \| 144.7 [113.8, 173.2] \| 125.4 [104.6, 146.2] \| 123.0 [99.8, 142.5] \| <0.001 \| 128.2 [107.4, 153.8] \| \| VLF (ms2) \| 4,026.5[2,764.5, 6,373.8] \| 3,374.6 [2,424.0, 4,324.5] \| 2,996.0[2,155.8, 3,929.6] \| <0.001 \| 3,387.8[2,509.2, 4,741.7] \| \| LF (ms2) \| 972.3 [698.9, 1,513.4] \| 586.2 [395.3, 885.4] \| 321.3 [214.5, 536.6] \| <0.0001 \| 616.6 [339.2, 1,001.9] \| \| LFnu \| 17.1 [13.9, 20.0] \| 14.1 [10.8, 17.0] \| 9.4 [7.3, 12.5] \| <0.0001 \| 13.2 [9.9, 17.5] \| \| HF (ms2) \| 418.3 [256.7, 900.5] \| 224.7 [118.3, 381.7] \| 152.8 [76.4, 259.6] \| <0.0001 \| 251.5 [129.1, 456.4] \| \| HFnu \| 7.6 [5.4, 11.2] \| 4.9 [3.1, 7.6] \| 4.3 [2.6, 6.4] \| <0.001 \| 5.4 [3.4, 8.4] \| \| LF/HF \| 2.2 [1.5, 2.8] \| 2.8 [1.9, 4.1] \| 2.2 [1.3, 3.2] \| <0.001 \| 2.3 [1.6, 3.4] \| \| UCG \| \| \| \| \| \| \| LVDd (mm) \| 46.8 ± 5.7 \| 46.0 ± 6.2 \| 45.9 ± 4.8 \| 0.432 \| 46.4 ± 5.7 \| \| EF (%) \| 67.4 ± 6.8 \| 66.6 ± 6.3 \| 67.1 ± 7.1 \| 0.684 \| 66.7 ± 6.9 \| \| E/e′ (septal) \| 5.8 ± 1.7 \| 6.3 ± 1.8 \| 6.6 ± 2.2 \| <0.001 \| 6.1 ± 1.9 \| \| E/e′ (lateral) \| 4.9 ± 1.3 \| 5.7 ± 1.6 \| 5.9 ± 1.8 \| <0.001 \| 5.4 ± 1.5 \| \| E/A \| 1.4 ± 0.4 \| 1.2 ± 0.4 \| 0.9 ± 0.3 \| <0.001 \| 1.2 ± 0.4 \| ## Study protocol Standard 12-lead ECG was performed, followed by UCG. Thereafter, an AECG recorder (FM180S, Fukuda Denshi Co., Ltd., Tokyo, Japan) was used to record the ECG for 24 h under free-running conditions, followed by analysis with the Holter ECG system (SCM8000, Fukuda Denshi Co., Ltd., Tokyo, Japan). ## Routine AECG data analysis Routine AECG data analysis was performed automatically with manual editing (Table 1). The parameters analyzed included the total number of beats, maximum, minimum, and mean heart rate (HR), and the frequency of VE and SVE per day. Ventricular arrhythmias were defined as follows: ventricular tachycardia (VT), ≥3 consecutive ventricular complexes at a rate >100 bpm; ventricular triplet (V3), more than three ventricular ectopic beats in a row at a rate <100 bpm; and ventricular couplet (V2), two ventricular ectopic beats in a row. Supraventricular arrhythmia was defined as follows: supraventricular tachycardia (SVT), ≥3 consecutive ventricular complexes at a rate >150 bpm; supraventricular triplet (S3), more than three supraventricular ectopic beats in a row at a rate <150 bpm; and supraventricular couplet (S2), two supraventricular ectopic beats in a row. The total number of beats, and the maximum and mean HR were significantly lower in the older generation ($P \leq 0.001$ for all, respectively) (Table 1). Thus, the prevalence of VE and SVE was significantly higher in the older-aged group ($P \leq 0.001$ and $P \leq 0.0001$, respectively) (Table 1). ## Analysis of HRV HRV analysis was also performed to evaluate autonomic nervous system activity using the SCM 8,000 system (Fukuda Denshi Co., Ltd., Tokyo, Japan) (Table 1). The RR interval was calculated for HRV analysis via the corrected maximum entropy method using Akaike's algorithm, as previously reported [23]. The HRV data were subjected to time and frequency domain analyses at 60-min intervals. The definitions of all HRV parameters were based on previous studies [24]. The parameters for time domain analysis, which were evaluated every 5 min over 24 h, included the following: standard deviation of the mean normal RR interval (SDNN), the square root of the mean of the sum of the squares of differences between adjacent normal to normal intervals (RMSSD), proportion of times between adjacent cycles that are different by >50 ms (pNN50), and standard deviation of the averages of NN intervals in all 5-min segments of the entire recording (SDANN). Frequency domain analysis entailed evaluation of the power in the very low-frequency area (VLF), power in low-frequency area (LF), power in the high-frequency area (HF), and power in the low-frequency/power in the high-frequency (LF/HF) ratio every 5 min. The power spectra of frequency domain analysis were defined as follows: total power (TP), approximately <0.4 Hz; power in the very low-frequency range (VLF), 0–0.04 Hz; power in the low-frequency range (LF), 0.04–0.15 Hz; and power in the high-frequency range (HF), 0.15–0.40 Hz. The normalized values (nu) were calculated using the following formula: LF/TP × 100 for LFnu, and HF/TP × 100 for HFnu. All HRV parameters, except for LF/HF, were significantly lower in the older generation ($P \leq 0.001$ for SDNN, RMSSD, pNN 50, SDANN, VLF; $P \leq 0.0001$ for LF, HF, and HFnu) (Table 1). ## Echocardiography recordings Echocardiography was performed using the Xalio (Toshiba Co., Ltd., Tokyo, Japan) system to evaluate left ventricular EF and left ventricular end-diastolic dimension (LVDd). Left ventricular EF was calculated during sinus beats using Simpson's method [25]. LVDd and EF did not differ significantly among the three generations ($$P \leq 0.432$$ and $$P \leq 0.684$$). E/e′ (septal) and E/e′ (lateral) were significantly higher in the older generation ($P \leq 0.001$ for all, respectively) (Table 1). ## Statistical analyses Data are presented as the mean ± standard deviation for normally distributed continuous variables, and as medians (interquartile range: 25–75th percentile) for non-normally distributed variables. Patient characteristics including demographic features, and the AECG, HRV, and UCG parameters were compared using the χ2 test for categorical variables, analysis of variance for continuous and parametric data, and the Kruskal–Wallis test for nonparametric data. Comparisons of frequencies among each hour in bradyarrythmias (sinus pause and AV block) and VE/SVE were performed using the Kruskal–Wallis test. The parameters influencing the ectopy prevalence in each generation were also compared using the Kruskal–Wallis test; post hoc multiple comparisons were performed using the Bonferroni method. Multivariate regression analysis was performed to determine the intensity of the incidence of premature atrial and ventricular complex and theoretical consideration of important factors such as the UCG and HRV indices. We also selected age, sex, and BMI as the explanatory variables for multivariate analysis. Before performing multiple regression for the incidence of VE and SVE, the HRV indices (SDNN, SDANN, RMSSD, PNN50, LFnu, HFnu, and LF/HF) were transformed to natural logarithms, as these parameters showed skewed distributions. Multivariate linear regression was performed after simultaneously controlling for potential confounders, followed by step-wise selection or backward selection. Log SDANN and log LFnu were excluded owing to multicollinearity (variance inflation factor >10) during multivariate regression analysis for both VE and SVE. We set the reference interval for the AECG parameters as the 2.5th–97.5th percentile according to the Clinical and Laboratory Standards Institute guidelines and meta-analysis [26, 27]. Furthermore, the sample size of the reference interval of AECG parameters was set at 120 participants minimum in each generation (20–39, 40–59, and 60–89 years) according to the Clinical and Laboratory Standards Institute guidelines [26]. Statistical analyses were conducted using SPSS version 28 (IBM Corp, Armonk, NY, USA). All tests were two-sided, and P-values <0.05 were considered statistically significant. ## Sinus pauses and conduction abnormalities The incidence of sinus pause >2 s was $4.1\%$, $1.6\%$, and $0.8\%$ in the 20–39-, 40–59-, and 60–89-year age groups, respectively (Table 2). The incidence was higher in the younger-aged group. The incidence of pause >2 s was under $5\%$ in all generations, rendering these findings abnormal. Generation-dependent incidence was observed in the case of second-degree AV block, akin to sinus pause. Additionally, the incidence of second-degree AV block (i.e., abnormal findings) was under $5\%$ for all generations, rendering these findings abnormal. The evaluation of the diurnal variations in the prevalence of sinus pause and second-degree AV block revealed that both were observed mainly at night-time: from 21:00 to 8:00 (Figures 1A,B). Figure 1: Diurnal variation in the median number of bradyarrhythmias. Significant night-time dominance was observed in the diurnal variation of the median number of sinus pauses ($$P \leq 0.002$$) (A). Significant night-time dominance in the diurnal variation of the median number of atrio-ventricular blocks ($$P \leq 0.009$$) (B). Comparisons of frequencies among each hour in sinus pause and AV block were performed using the Kruskal–Wallis test. AV block: atrio-ventricular block. TABLE_PLACEHOLDER:Table 2 ## Percentile of simple VE and SVE number (reference values of ectopy) The principal results of this study are presented in Table 3 and summarized in the Figures 3A,B. The 97.5th percentile of simple VE frequency (reference values of ectopy) was 149, 254, and 1,682/day in the 20–39-, 40–59-, and 60–89-years age groups, respectively. The overall reference value for premature ventricular ectopy for all generations was 366/day. On the other hand, the 95th percentile of the frequency of simple SVE (reference values of extrasystole) was 131, 232, and 1,063/day for the 20–39-, 40–59-, and 60–89-year age groups, respectively. Overall, the reference value of SVE for all generations was 537/day. Table 3 \| Unnamed: 0 \| Percentile \| Percentile.1 \| Percentile.2 \| Percentile.3 \| Percentile.4 \| Percentile.5 \| Percentile.6 \| Percentile.7 \| Percentile.8 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| \| 2.5 \| 5 \| 10 \| 25 \| 50 \| 75 \| 90 \| 95 \| 97.5 \| \| VE \| \| \| \| \| \| \| \| \| \| \| 20–39 years \| 0 \| 0 \| 0 \| 0 \| 1 \| 3 \| 9 \| 59 \| 149 \| \| 40–59 years \| 0 \| 0 \| 0 \| 0 \| 2 \| 6 \| 37 \| 144 \| 254 \| \| 60–89 years \| 0 \| 0 \| 0 \| 1 \| 4 \| 13 \| 171 \| 393 \| 1682 \| \| 20–89 years \| 0 \| 0 \| 0 \| 0 \| 2 \| 7 \| 57 \| 194 \| 366 \| \| \| Percentile \| Percentile \| Percentile \| Percentile \| Percentile \| Percentile \| Percentile \| Percentile \| Percentile \| \| \| 2.5 \| 5 \| 10 \| 25 \| 50 \| 75 \| 90 \| 95 \| 97.5 \| \| SVE \| SVE \| SVE \| SVE \| SVE \| SVE \| SVE \| SVE \| SVE \| SVE \| \| 20–39 years \| 0 \| 0 \| 0 \| 2 \| 6 \| 13 \| 24 \| 52 \| 131 \| \| 40–59 years \| 1 \| 1 \| 2 \| 6 \| 13 \| 29 \| 55 \| 125 \| 232 \| \| 60–89 years \| 4 \| 9 \| 14 \| 27 \| 67 \| 189 \| 453 \| 558 \| 1063 \| \| 20–89 years \| 0 \| 0 \| 2 \| 5 \| 18 \| 50 \| 186 \| 311 \| 537 \| Significant diurnal variation was observed in the mean HR and mean frequency of VE and SVE. The mean frequency of VE was significantly higher during the waking hours (8:00–24:00) than during sleeping hours (23:00–7:00) ($P \leq 0.001$) (Figure 2A). In contrast, the mean frequency distribution of SVE had two peaks at 4:00 and 13:00–15:00 ($P \leq 0.001$) (Figure 2B). Figure 2: Diurnal variation in the median frequency of ventricular and supraventricular ectopy. A significant diurnal variation was observed in the mean frequency of VE (P < 0.001). The daytime prevalence of VE was predominant, which was parallel to the diurnal variation in HR (A). Meanwhile, the mean frequency of SVE was significantly higher at 4:00 and during 13:00–15:00 (P < 0.001). However, the diurnal variation of SVE was not parallel to the diurnal variation of HR (B). Comparisons of frequencies among each hour in VE and SVE were performed using the Kruskal–Wallis test. HR: heart rate, SVE: supraventricular ectopy, VE: ventricular ectopy. ## Complex VE and atrial SVE The findings associated with complex ectopy and tachycardia are described in Table 4. VE Multiform was observed in $\frac{138}{365}$ ($37.8\%$) of the participants (Table 4). VT and V3 were observed in $\frac{6}{365}$ ($1.6\%$) and $\frac{4}{365}$ ($1\%$) of the participants, respectively, whereas R-on-T were not observed in any participant. All types of SVT, S3, and S2 were observed in $\frac{37}{365}$ ($10.1\%$), $\frac{86}{365}$ ($23.5\%$), and $\frac{151}{365}$ ($41.3\%$) participants, respectively. The incidence of complex SVE increased with age progression (Table 4). Table 4 \| VE \| Incidence \| Multiform \| VT \| V3 \| V2 \| R-on-T \| Bigeminy \| Trigeminy \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| 20–39 years (n = 120) \| 68 (56.7%) \| 27 (22.5%) \| 2 (1.7%) \| 0(0%) \| 2 (1.6%) \| 0 (0%) \| 4 (3.3%) \| 2 (1.7%) \| \| 40–59 years (n = 124) \| 88 (71.0%) \| 46 (37.0%) \| 2 (1.6%) \| 0 (0%) \| 6 (4.8%) \| 0 (0%) \| 4 (3.2%) \| 2 (1.6%) \| \| 60–89 years (n = 121) \| 102 (84.3%) \| 65 (53.7%) \| 2 (1.7%) \| 4 (3.3%) \| 16 (13.2%) \| 0 (0%) \| 10 (8.2%) \| 11 (9%) \| \| 20–89 years (N = 365) \| 258 (70.7%) \| 138 (37.8%) \| 6 (1.6%) \| 4 (1%) \| 25 (6.6%) \| 0 (0%) \| 18 (4.9%) \| 15 (4.1%) \| \| SVE \| Incidence \| All types of SVT \| SVT >10 beats \| S3 \| S2 \| \| \| \| \| 20–39 years (n = 120) \| 103 (85.8%) \| 1 (0.8%) \| 0 (0%) \| 6 (5%) \| 19 (15.8%) \| \| \| \| \| 40–59 years (n = 124) \| 122 (98.4%) \| 4 (3.2%) \| 0 (0%) \| 22 (17.7%) \| 42 (33.9%) \| \| \| \| \| 60–89 years (n = 121) \| 121 (100%) \| 32 (26.4%) \| 10 (12%) \| 73 (60.3%) \| 90 (74.3%) \| \| \| \| \| 20–89 years (N = 365) \| 346 (94.7%) \| 37 (10.1%) \| 10 (2.7%) \| 86 (23.5%) \| 151 (41.3%) \| \| \| \| ## Correlation between the incidence of ectopy and UCG and HRV indices Multivariate regression analysis was performed to explore the intensity of factors affecting the incidence of VE and SVE. Log VE was higher in the older generation ($$P \leq 0.014$$) (Figure 3A). Age was an independent factor influencing the VE incidence (β = 0.293, $$P \leq 0.001$$), whose effect was retained in step-wise selection (β = 0.207, $$P \leq 0.001$$) (Table 5). In a sub-analysis, multivariate regression analysis with the backward selection method showed that age tended to be the most influential factor for log VE in all the generations (20–39, 40–59, and 60–89 years) ($$P \leq 0.054$$–0.079) (Supplementary Tables S1–S3). Log SVE was higher in the older generation ($P \leq 0.001$) (Figure 3B). However, BMI was significantly higher in the 40–59-year age group than in the 20–39- and 60–89-year age groups ($$P \leq 0.016$$) (Figure 3C). Hence, log SDNN, log RMSSD, and log HFnu were lower in the older generation ($P \leq 0.001$ for all) (Figure 3D,E,, F). Age, BMI, log SDNN, log RMSSD, and log HFnu were significant factors affecting the SVE incidence (age, β = 0.532, $P \leq 0.001$; BMI, β = −0.099, $$P \leq 0.029$$; log SDNN, β = −0.136, $$P \leq 0.02$$; RMSSD, β = 0.457, $P \leq 0.001$; log HFnu, β = −0.368, $$P \leq 0.001$$). Moreover, these indices were significant factors affecting the SVE incidence, even according to step-wise selection (age, β = 0.642, $P \leq 0.001$; BMI, β = v0.112, $$P \leq 0.009$$; log RMSSD, β = 0.097, $$P \leq 0.035$$) (Table 6). In contrast, a sub-analysis was performed on the most influential factors for log SVE in each generation (20–39, 40–59, and 60–89 years). Multivariate analysis revealed that age, BMI, and RMSSD were significant factors (Supplementary Tables S4–S6), with a similar trend as in the analysis of all generations (Table 6). However, in the 60–89-year age group, BMI was not a significant factor. Figure 3: Changes in the ectopy prevalence and parameters influencing ectopy prevalence in each generation. The log VE/day and log SVE/day were significantly decreased in the older generation ($$P \leq 0.014$$ for log VE; $P \leq 0.001$ for log SVE), (A), (B). On the other hand, body mass index was significantly higher in the 40–59-year age group than in the 20–39- and 60–89-year age groups ($$P \leq 0.016$$) (C). HRV parameters (log SDNN, log RMSSD, and log HFnu), which influenced the prevalence of SVE, were significantly decreased in the older generation ($P \leq 0.001$ for all, respectively) (D–F). The parameters influencing the ectopy prevalence in each generation were compared using the Kruskal–Wallis test; post hoc multiple comparisons were performed using the Bonferroni method. HRV, heart rate variability; SVE, supraventricular ectopy; VE, ventricular ectopy. TABLE_PLACEHOLDER:Table 5 TABLE_PLACEHOLDER:Table 6 ## Discussion In this study, we provided age-specific reference values for AECG parameters, including bradycardia detected using 24-h AECG, in each generation, spread over a wide age range (20–89 years) in a healthy population. Moreover, we provided evidence that the incidence of VE was only related to the increase in age; hence, SVE was influenced by the increase in age and BMI and decrement in RMSSD and HFnu, which are reflective of parasympathetic nervous system activity. This is the first study to demonstrate the relationship between autonomic tone activity, expressed as HRV, and the incidence of VE and SVE over a wide age range. ## Reference intervals of AECG parameters The differences between the “normal” and “abnormal” AECG findings in each generation (20–39, 40–59, and 60–89 years) (Table 7) were determined, based on the assumption that events occurring in less than $2.5\%$ of a healthy population were “abnormal” and those occurring in more than $2.5\%$ of the population were “normal.” The 2.5th–97.5th range is defined as the reference interval in the Clinical and Laboratory Standards Institute guidelines, as well as in many of the articles included in the meta-analysis and the meta-analysis itself [26, 27]. Therefore, in the present study, the 95th percentile distribution was also defined as the reference interval or reference value. The strength of the reference values calculated in our study is the mild skew in age and sex, and the wide age range (20–89 years) of the participants (Table 1). Table 7 \| Unnamed: 0 \| Normal \| Abnormal \| \| --- \| --- \| --- \| \| 20–39 years (n = 120) \| \| \| \| Bradycardia \| Sinus pause <3 s \| Sinus pause >3 s,Second-degree atrio-ventricular block (Mobitz),Third-degree atrio-ventricular block \| \| Ectopy and tachycardia (ventricular) \| VE < 149 \| VE >149, Multiform VE, VT, V3, V2, R on T, Bigeminy, Trigeminy \| \| Ectopy and tachycardia (supraventricular) \| SVE <131, S3, S2 \| SVE >131, any SVT \| \| 40–59 years (n = 124) \| \| \| \| Bradycardia \| Sinus pause <2 s \| Sinus pause >3 s,All second-degree atrio-ventricular block,Third-degree atrio-ventricular block \| \| Ectopy and tachycardia (ventricular) \| VE <232 \| VE >232, Multiform VE, VT, V3, V2, R on T, Bigeminy, Trigeminy \| \| Ectopy and tachycardia (supraventricular) \| SVE <144, S3, S2 \| SVE >144SVT >10 beat \| \| 60–89 years (n = 121) \| \| \| \| Bradycardia \| Sinus pause <2 s \| Sinus pause >3 s,All second-degree atrio-ventricular block, Third-degree atrio-ventricular block \| \| Ectopies and tachycardia (ventricular) \| VE <1,682, V2, V3, Bigeminy, Trigeminy \| VE >1,682, Multiform VE, VT, R on T \| \| Ectopy and tachycardia (supraventricular) \| SVE <1,063, S3, S2 \| SVE >1,063SVT >10 beat \| Moreover, we set stringent criteria for healthy participants in this study, who were defined as individuals with no history of the following: cardiac abnormalities, abnormality on physical examination, 12-lead ECG, chest radiograph, blood investigations, and almost normal UCG findings; previous studies did not establish such strict criteria, especially with respect to blood work and UCG [25]. Therefore, it is possible to designate this result as a precise reference interval. However, this reference interval is not normally distributed. There is a large discrepancy in the 90–97.5 percentile, especially in VE and SVE; therefore, this value should be treated with caution (Table 3). Since strict criteria of reference values, such as those in this study, have not existed in recent years, this information may be very useful not only for physicians but also for healthcare professionals in clinical settings in many situations, such as outpatient clinics and health checkup posts. Moreover, this reference interval could be versatile, because an AECG is performed in not only cardiology, but also various other medical departments and in general medicine. ## Sinus pauses and conduction abnormality We found that the incidence of sinus pause >3 s and second-degree AV block (Mobitz)was less than $2.5\%$ in all generations (Table 2). Therefore, sinus pause >3 s and second-degree AV block (Mobitz) are abnormal findings in healthy individuals. Although the incidence of second-degree AV block (Wenckebach) in 20–39 years was more than $2.5\%$, that in 40–59 and 60–89 years was less than $2.5\%$ each. Therefore, second-degree AV block (Wenckebach) in an abnormal finding in 40–59 and 60–89 years. Moreover, the incidence of bradyarrhythmia was higher in the younger-age group. These findings are consistent with those of a previous meta-analysis [27]. Hingorani et al. reported that the incidence of sinus pause >2 s in 1,273 healthy normal volunteers aged 18–45 and 46–65 years was $4.4\%$ and $0\%$, respectively, whereas the incidence of second-degree AV block was $2.6\%$ and $0.9\%$, respectively [17]. The precise pathogenesis responsible for the higher incidence of bradyarrhythmia in the younger population is unknown. However, we speculated that autonomic nervous system activity, especially parasympathetic dominance, in younger individuals contributes to the susceptibility to bradyarrhythmia. The night-time predominance of sinus pause and diurnal variation in the AV block suggests the involvement of parasympathetic tone in these arrhythmias. Our findings provide valuable evidence, as no study has focused on diurnal variation in bradycardia detected on AECG [27]. ## Reference intervals of VE and SVE We provided reference values for both VE and SVE in each generation (20–39, 40–59, and 60–89 years) over a wide age range in a population (Table 3 and central illustration). Several studies have reported on the frequency of VE and SVE/24 h using AECG in a few age groups in apparently healthy participants (Tables 8–10) (5–16). However, few studies have demonstrated reference values in each generation (20–39, 40–59, and 60–89 years) over a wide age range. Notably, the frequency of VE and SVE/24 h was higher in the older-aged group than in the younger-aged group in all percentile categories (from the 2.5th–97.5th percentiles) (Table 3). Recently, Williams et al. conducted a meta-analysis of 33 studies from 1950 to 2020 concerning reference intervals for AECG parameters and reported that the normal range of VE and SVE was 0–$\frac{500}{24}$ h, 0–1,$\frac{000}{24}$ h, and >1,$\frac{000}{24}$ h, for 20–39, 40–59, and 60–89 years, respectively [27]. These findings are consistent with our data, except for the older generation (60–89 years). However, most studies (28 of 33) incorporated in that meta-analysis were published before 2000. The reference values for the young generation (18–36 years) were published in 1981 (Table 8). Moreover, reference values for the middle-age group are lacking (Table 9). Notably, the latest study to report reference values (age range: 64–80 years) of VE and SVE in a Japanese population was published in 2006, but data for establishing the reference values were collected in 1989 (Table 10) [15]. Thus, the data used, and inferences derived from these studies are extremely old. Therefore, our study's findings significantly contribute to and expand the existing body of evidence. We also investigated complex ectopy and tachycardia using 24-h AECG (Table 4). The incidence of VT, R-on-T, and SVT >10 beats in 20–39 and 40–59 years generation were less than $2.5\%$ in all generations; i.e., these findings are abnormal in healthy individuals. However, the prevalence of bigeminy and trigeminy in the 60–89 years age group was $8.2\%$ and $9\%$, respectively. To the best of our knowledge, this study was the first to conduct such a detailed analysis, rendering these findings novel. ## Correlation between ventricular ectopy and UCG or HRV parameters In the present study, VEs were only correlated with age, whereas SVEs were correlated with BMI, age, log RMSSD, and log HFnu. Aging has the greatest influence on the frequency of VE and SVE. Regarding VE, age was the independent factor that affected the number of VE through all the generations. VE had no relationship with the other factors in Figures 3D–F (log SDNN, log RMSSD, and log HFnu). In the sub-analysis, age tended to be the most influential factor affecting VE, although it was not statistically significant (Supplementary -Tables S1–S3). We speculate that this may have been the case because the sub-analysis was divided based on the generations and therefore did not reach significance. It has been widely reported that the prevalence of VE in the older population was higher than that in the younger population, which was also proven in a meta-analysis [27]. Moreover, Tasaki et al. followed a cohort of healthy individuals for 15 years and found that the incidence of VE and SVE increased significantly after 15 years [15]. Therefore, although the higher incidence of VE and SVE in older individuals is an unquestionable fact, few studies have investigated the mechanism of this phenomenon. The age-related changes in intracellular Ca2+ regulation which play an important role in the development of several types of arrhythmias may explain this phenomenon [28]. Studies have suggested that age-related changes in intracellular Ca2+ regulation may prolong the action potential, especially during tachycardia, inducing electrical instability due to inadequate return of intracellular Ca2+ concentration. VE, which accounts for the high diurnal variation in VE when HR is elevated during the day, supports this hypothesis. ## Correlation between supraventricular ectopy and UCG or HRV parameters Several studies have reported that the frequency of SVEs increases with age, but few have examined the correlation between the frequency of ectopy (SVE) and UCG and HRV parameters simultaneously. SVE was inversely correlated with parasympathetic indices such as log RMSDD and log Hfnu (Figures 3B,E,F), thereby supporting the results of the multivariate analysis in Table 6. In contrast, age, obesity, and RMSSD, a parasympathetic index, were significant factors influencing SVE, as was the case for all age groups (Table 6 and Supplementary Tables S4, S5). However, BMI was not significant in the 60–89-year age group (Supplementary Table S6). The possible causes for this are as follows: BMI was lower in this generation than in the 40–59-year age group (Figure 3C) and the small variation in BMI made it less likely to be statistically significant. The causes for the increase in frequency of SVE with age have not been clarified in humans; however, the following speculations are made regarding the basic experimental study. Age-related changes in ion channels in the atria and ventricles are key to the dynamics of Ca2+ channels. In the animal experimental study, the uptake of Ca2+ into the sarcoplasmic reticulum decreases with age and intracellular Ca2+ increases with age [29]. Increased intracellular Ca2+ causes early posterior depolarization and induces APC and AF [30]. Conversely, it has been reported that aging (degree of frailty) correlates with prolongation of the P wave and PR interval in ECGs of aged mice and that this is caused by elevated levels of interstitial fibrosis and collagen content [31]. The above structural remodeling has been reported to increase the frequency of AF from APCs with aging. It is generally recognized that RMSSD, pNN50, and HFnu are parameters related to parasympathetic nervous system activity [24]. Therefore, there is a possibility that the increment in the frequency of SVE with aging partially results from decreased autonomic nervous system activity due to aging, particularly parasympathetic nervous system activity. Automaticity or triggered activity is thought to be the mechanism underlying SVE occurrence [32]. It is speculated that a decrease in parasympathetic activity can lead to an increase in automaticity [32], which may be responsible for the decrease in parasympathetic activity in middle-aged and older individuals and may increase the frequency of SVEs with aging. It has been widely reported that the incidence of AF increases in middle-aged and older individuals [33]. The incidence of SVE due to aging and the change in the equilibrium of sympathetic/parasympathetic activity may influence the increase in AF in older individuals. Incidentally, fluctuations in heart rate variability, expressed as SDNN, became significantly smaller with age. This result is consistent with that of previous reports and is an age-related change [34]. In this study, multiple regression analysis revealed that BMI was an independent factor influencing SVE prevalence. Naturally, the high prevalence of SVE can induce AF. Obesity is an independent risk factor for increasing the prevalence of AF [35]. Although the pathophysiology of obesity implicating AF is not completely understood, the factors associated with it are as follows: genetic factors; clinical correlations such as hypertension, diabetes mellitus, and sleep apnea syndrome; coronary artery disease; ventricular adaptation; inflammation; oxidative stress; focal adrenergic pathways; and focal adiposity [36]. Among these, epicardial focal adiposity has recently garnered much attention. Recent studies have reported that the increase in epicardial fat caused by obesity leads to the development of adipocyte infiltration into the myocardium, fibrosis, inflammation, oxidative stress, and impaired cardiac muscle activity in the myocardium [37]. These factors can be triggers underlying the development of AF [37]. Our findings show that a higher BMI contributes to the increased incidence of atrial premature complexes and may support recent findings on the role of obesity in AF. ## Limitations There are some limitations to this study. This study was performed for a brief duration of monitoring, i.e., a 24-hr period without any follow-up. We did not evaluate reproducibility between day-to-day values, which should be assessed using novel AECG devices, such as patch ECG, in the future [38]. Moreover, the study population was restricted to individuals of Asian ethnicity; there is a possibility that the reference values of other ethnicities such as European, African, and Hispanic may be different. The minimum sample size required for the reference interval recommended by the Clinical and Laboratory Standards Institute guidelines [26] is met in this paper. However, a larger cohort and several follow-up recordings will be needed to investigate potential future directions of this work. In this study, the age range of 60–89 years was adopted as a single group. However, as shown in the meta-analysis by Williams et al. [ 27], the validity of the healthy value of 80 years of age and older is a controversial area and has not been clarified in previous reports. In order to verify the validity of using 60 years of age as a cutoff, we first compared the items listed as parameters in this study in the age group of 60–70 years and 70–89 years. There were no significant differences in all arrhythmia parameters ($$P \leq 0.204$$–0.916) except R-on-T, V3, and bradyarrhythmia. We then compared the parameters in the 60–75-year and 75–89-year age groups. There were no significant differences in any of the arrhythmia parameters except R-on-T, V3, and bradyarrhythmia ($$P \leq 0.349$$–0.972). These results support the fact that the age category of 60–89 years used in this study is valid. On the other hand, we could not validate R-on-T, V3, and bradyarrhythmia in the 60–89 years age group because the number of patients in all categories of R-on-T, V3, and bradyarrhythmia (Sinus pause and AV block) was less than 2, and statistics were difficult to obtain. ## Conclusions We presented age-specific reference values for AECG parameters derived from 24-h AECG in healthy individuals, over a wide age range (20–89 years). Notably, the reference values of VE and SVE were different in each generation. Moreover, we demonstrated that the incidence of VE was only related to the progression in age; hence, SVE was influenced by age and BMI increases, and RMSSD and HFnu decreases, which represent parasympathetic nervous system activity. This information will be useful for the diagnosis and prevention of diverse cardiac diseases in patients of various age groups in clinical settings. Future studies that account for the daily variance in healthy individuals are warranted to seek the reference interval of AECG. ## 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 Medical Ethics Committee of the National Defense Medical College Hospital (approval no. 4645), Saitama, Japan, and Nihon University School of Medicine, Itabashi Hospital, Tokyo, Japan (approval no. MF 2208-0037). The patients/participants provided their written informed consent to participate in this study. ## Author contributions KH and NH: conception and design of the study, data collection, writing of the manuscript, and formatting and submission of the manuscript. YK and YT: design, conception, and supervision of the study, and revision of the manuscript. MK, YK, NF, AS, YO, YO, and TT: supervision of the study, and revision of 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/fcvm.2023.1099157/full#supplementary-material. ## References 1. Holter NJ. Radioelectrocardiography: a new technique for cardiovascular studies. Ann N Y Acad Sci (1957) 65 913-23. DOI: 10.1111/j.1749-6632.1957.tb36693.x 2. Turakhia MP, Hoang DD, Zimetbaum P, Miller JD, Froelicher VF, Kumar UN. Diagnostic utility of a novel leadless arrhythmia monitoring device. Am J Cardiol (2013) 112 520-4. DOI: 10.1016/j.amjcard.2013.04.017 3. Reiffel JA, Verma A, Kowey PR, Halperin JL, Gersh BJ, Wachter R. Incidence of previously undiagnosed atrial fibrillation using insertable cardiac monitors in a high-risk population: the reveal AF study. JAMA Cardiol (2017) 2 1120-7. DOI: 10.1001/jamacardio.2017.3180 4. Perez MV, Mahaffey KW, Hedlin H, Rumsfeld JS, Garcia A, Ferris T. Large-scale assessment of a smartwatch to identify atrial fibrillation. N Engl J Med (2019) 381 1909-17. DOI: 10.1056/NEJMoa1901183 5. Brodsky M, Wu D, Denes P, Kanakis C, Rosen KM. Arrhythmias documented by 24 h continuous electrocardiographic monitoring in 50 male medical students without apparent heart disease. Am J Cardiol (1977) 39 390-5. DOI: 10.1016/s0002-9149(77)80094-5 6. Doniwa K, Suzuki S, Shimizu M, Sano N, Okajima M. Arrhythmias documented in over 100 healthy young subjects by 24 h holter electrocardiographic monitoring. Heart (1981) 13 330-5 7. Rasmussen V, Jensen G, Schnohr P, Hansen JF. Premature ventricular beats in healthy adult subjects 20 to 79 years of age. Eur Heart J (1985) 6 335-41. DOI: 10.1093/oxfordjournals.eurheartj.a061860 8. Orth-Gomér K, Hogstedt C, Bodin L, Söderholm B. Frequency of extrasystoles in healthy male employees. Br Heart J (1986) 55 259-64. DOI: 10.1136/hrt.55.3.259 9. von Rotz M, Aeschbacher S, Bossard M, Schoen T, Blum S, Schneider S. Risk factors for premature ventricular contractions in young and healthy adults. Heart (2017) 103 702-7. DOI: 10.1136/heartjnl-2016-309632 10. Kostis JB, McCrone K, Moreyra AE, Gotzoyannis S, Aglitz NM, Natarajan N. Premature ventricular complexes in the absence of identifiable heart disease. Circulation (1981) 63 1351-6. DOI: 10.1161/01.cir.63.6.1351 11. Bjerregaard P. Premature beats in healthy subjects 40–79 years of age. Eur Heart J (1982) 3 493-503. DOI: 10.1093/oxfordjournals.eurheartj.a061344 12. Kantelip JP, Sage E, Duchene-Marullaz P. Findings on ambulatory electrocardiographic monitoring in subjects older than 80 years. Am J Cardiol (1986) 57 398-401. DOI: 10.1016/0002-9149(86)90760-5 13. Andersson S, Osterlind PO, Holmboe G, Windblad B. Twenty-four-hour electrocardiography in a healthy elderly population. Gerontology (1988) 34 139-44. DOI: 10.1159/000212943 14. Ribera JM, Cruz AJ, Zamorano JL, Pérez-Casar F. Cardiac rate and hyperkinetic rhythm disorders in healthy elderly subjects: evaluation by ambulatory electrocardiographic monitoring. Gerontology (1989) 35 158-64. DOI: 10.1159/000213015 15. Tasaki H, Serita T, Ueyama C, Kitano K, Seto S, Yano K. Longitudinal age-related changes in 24-hour total heart beats and premature beats and their relationship in healthy elderly subjects. Int Heart J (2006) 47 549-63. DOI: 10.1536/ihj.47.549 16. Heckbert SR, Austin TR, Jensen PN, Floyd JS, Psaty BM, Soliman EZ. Yield and consistency of arrhythmia detection with patch electrocardiographic monitoring: the Multi-Ethnic Study of Atherosclerosis. J Electrocardiol (2018) 51 997-1002. DOI: 10.1016/j.jelectrocard.2018.07.027 17. Hingorani P, Karnad DR, Rohekar P, Kerkar V, Lokhandwala YY, Kothari S. Arrhythmias seen in baseline 24-hour Holter ECG recordings in healthy normal volunteers during phase 1 clinical trials. J Clin Pharmacol (2016) 56 885-93. DOI: 10.1002/jcph.679 18. Stinson JC, Pears JS, Williams AJ, Campbell RW. Use of 24 h ambulatory ECG recordings in the assessment of new chemical entities in healthy volunteers. Br J Clin Pharmacol (1995) 39 651-6. DOI: 10.1111/j.1365-2125.1995.tb05724.x 19. Mann D, Zipes D, Libby P, Bonow R. Baunwald’s heart disease: A textbook of cardiovascular medicine (2015) 712-54 20. Bas HD, Baser K, Hoyt J, Yokokawa M, LaBounty T, Morady F. Effect of circadian variability in frequency of premature ventricular complexes on left ventricular function. Heart Rhythm (2016) 13 98-102. DOI: 10.1016/j.hrthm.2015.07.038 21. Dukes JW, Dewland TA, Vittinghoff E, Mandyam MC, Heckbert SR, Siscovick DS. Ventricular ectopy as a predictor of heart failure and death. J Am Coll Cardiol (2015) 66 101-9. DOI: 10.1016/j.jacc.2015.04.062 22. Binici Z, Intzilakis T, Nielsen OW, Køber L, Sajadieh A. Excessive supraventricular ectopic activity and increased risk of atrial fibrillation and stroke. Circulation (2010) 121 1904-11. DOI: 10.1161/CIRCULATIONAHA.109.874982 23. Hashimoto K, Kasamaki Y, Soma M, Takase B. Diurnal variation of frequency domain T-wave alternans on 24-hour ambulatory electrocardiogram in subjects without heart disease: significant effect of autonomic nervous activity of the heart. Ann Noninvasive Electrocardiol (2019) 24 e12620. DOI: 10.1111/anec.12620 24. Heart rate variability: standards of measurement, physiological interpretation and clinical use. Circulation (1996) 93 1043-65. DOI: 10.1161/01.CIR.93.5.1043 25. Lang RM, Badano LP, Mor-Avi V, Afilalo J, Armstrong A, Ernande L. 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) 16 233-70. DOI: 10.1016/j.echo.2014.10.003 26. 26.Clinical Laboratory, Standards Institute. How to define and determine reference intervals in the clinical laboratory; approved guideline. 2nd ed. CLSI Document c28-A2. Wayne, PA: CLSI (2000). 24 p.. How to define and determine reference intervals in the clinical laboratory; approved guideline (2000) 24 27. Williams CB, Andrade JG, Hawkins NM, Cheung C, Krahn A, Laksman ZW. Establishing reference ranges for ambulatory electrocardiography parameters: meta-analysis. Heart (2020) 106 1732-9. DOI: 10.1136/heartjnl-2020-316925 28. Hatch F, Lancaster MK, Jones SA. Aging is a primary risk factor for cardiac arrhythmias: disruption of intracellular Ca. Expert Rev Cardiovasc Ther (2011) 9 1059-67. DOI: 10.1586/erc.11.112 29. Carbonin PU, Di Gennaro M, Pahor M, Bernabei R, Sgadari A, Gambassi G. Cardiac aging, calcium overload, and arrhythmias. Exp Gerontol (1990) 25 261-8. DOI: 10.1016/0531-5565(90)90061-6 30. Ono N, Hayashi H, Kawase A, Lin SF, Li H, Weiss JN. Spontaneous atrial fibrillation initiated by triggered activity near the pulmonary veins in aged rats subjected to glycolytic inhibition. Am J Physiol Heart Circ Physiol (2007) 292 H639-48. DOI: 10.1152/ajpheart.00445.2006 31. Jansen HJ, Moghtadaei M, Mackasey M, Rafferty SA, Bogachev O, Sapp JL. Atrial structure, function and arrhythmogenesis in aged and frail mice. Sci Rep (2017) 7 44336. DOI: 10.1038/srep44336 32. Bayés de Luna A, Baranchak A. Clinical arrhythmology. Diagnosis, prognosis and treatment of arrhythmias (2017) 85 33. Inoue H, Fujiki A, Origasa H, Ogawa S, Okumura K, Kubota I. Prevalence of atrial fibrillation in the general population of Japan: an analysis based on periodic health examination. Int J Cardiol (2009) 137 102-7. DOI: 10.1016/j.ijcard.2008.06.029 34. Sammito S, Böckelmann I. New reference values of heart rate variability during ordinary daily activity. Heart Rhythm (2017) 14 304-7. DOI: 10.1016/j.hrthm.2016.12.016 35. Wang TJ, Parise H, Levy D, D’Agostino RB, Wolf PA, Vasan RS. Obesity and the risk of new-onset atrial fibrillation. JAMA (2004) 292 2471-7. DOI: 10.1001/jama.292.20.2471 36. Goudis CA, Korantzopoulos P, Ntalas IV, Kallergis EM, Ketikoglou DG. Obesity and atrial fibrillation: a comprehensive review of the pathophysiological mechanisms and links. J Cardiol (2015) 66 361-9. DOI: 10.1016/j.jjcc.2015.04.002 37. Wong CX, Ganesan AN, Selvanayagam JB. Epicardial fat and atrial fibrillation: current evidence, potential mechanisms, clinical implications, and future directions. Eur Heart J (2017) 38 1294-302. DOI: 10.1093/eurheartj/ehw045 38. Hashimoto K, Harada N, Kasamaki Y. Can a patch electrocardiographic device be a leading actor for detecting atrial fibrillation?—diversifying electrocardiographic monitoring devices. Circ J (2022) 86 189-91. DOI: 10.1253/circj.CJ-21-0644
--- title: 'Risk of SARS-CoV-2 infection, severe COVID-19 illness and COVID-19 mortality in people with pre-existing mental disorders: an umbrella review' authors: - Federico Bertolini - Anke B Witteveen - Susanne Young - Pim Cuijpers - Jose Luis Ayuso-Mateos - Corrado Barbui - María Cabello - Camilla Cadorin - Naomi Downes - Daniele Franzoi - Michael Elizabeth Gasior - Brandon Gray - Ann John - Maria Melchior - Mark van Ommeren - Christina Palantza - Marianna Purgato - Judith Van der Waerden - Siyuan Wang - Marit Sijbrandij journal: BMC Psychiatry year: 2023 pmcid: PMC10026202 doi: 10.1186/s12888-023-04641-y license: CC BY 4.0 --- # Risk of SARS-CoV-2 infection, severe COVID-19 illness and COVID-19 mortality in people with pre-existing mental disorders: an umbrella review ## Abstract ### Introduction The COVID-19 pandemic has posed a serious health risk, especially in vulnerable populations. Even before the pandemic, people with mental disorders had worse physical health outcomes compared to the general population. This umbrella review investigated whether having a pre-pandemic mental disorder was associated with worse physical health outcomes due to the COVID-19 pandemic. ### Methods Following a pre-registered protocol available on the Open Science Framework platform, we searched Ovid MEDLINE All, Embase (Ovid), PsycINFO (Ovid), CINAHL, and Web of Science up to the 6th of October 2021 for systematic reviews on the impact of COVID-19 on people with pre-existing mental disorders. The following outcomes were considered: risk of contracting the SARS-CoV-2 infection, risk of severe illness, COVID-19 related mortality risk, risk of long-term physical symptoms after COVID-19. For meta-analyses, we considered adjusted odds ratio (OR) as effect size measure. Screening, data extraction and quality assessment with the AMSTAR 2 tool have been done in parallel and duplicate. ### Results We included five meta-analyses and four narrative reviews. The meta-analyses reported that people with any mental disorder had an increased risk of SARS-CoV-2 infection (OR: 1.71, $95\%$ CI 1.09–2.69), severe illness course (OR from 1.32 to 1.77, $95\%$CI between 1.19–1.46 and 1.29–2.42, respectively) and COVID-19 related mortality (OR from 1.38 to 1.52, $95\%$CI between 1.15–1.65 and 1.20–1.93, respectively) as compared to the general population. People with anxiety disorders had an increased risk of SAR-CoV-2 infection, but not increased mortality. People with mood and schizophrenia spectrum disorders had an increased COVID-19 related mortality but without evidence of increased risk of severe COVID-19 illness. Narrative reviews were consistent with findings from the meta-analyses. ### Discussion and conclusions As compared to the general population, there is strong evidence showing that people with pre-existing mental disorders suffered from worse physical health outcomes due to the COVID-19 pandemic and may therefore be considered a risk group similar to people with underlying physical conditions. Factors likely involved include living accommodations with barriers to social distancing, cardiovascular comorbidities, psychotropic medications and difficulties in accessing high-intensity medical care. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12888-023-04641-y. ## Background The COVID-19 pandemic has affected most of the population worldwide. Disadvantaged population groups are thought to have suffered the most, with possibly among those, people affected by mental disorders. The physical health repercussions have been a particular point of concern, as many known risk factors for severe COVID-19 illness course and COVID-19 related mortality are common in people with mental disorders [1–3]. Firstly, it is unclear whether this population, or specific subgroups, had an increased risk of contracting the SARS-CoV-2 virus than the general population [4, 5]. Shared living accommodations represented contexts in which adhering to social distancing might have been not possible or difficult. Secondly, there has been a focus on the disease course in people with mental disorders. Obesity, diabetes mellitus and cardiovascular disease are common in this population [6, 7]. In addition to that, many psychotropic medications have the potential of a detrimental effect on the respiratory function [8, 9]. Given these premises, it seems highly relevant to assess if and how the COVID-19 pandemic has affected the physical health of people with pre-existing mental disorders. Against this background, an umbrella review was performed aiming to summarize evidence from systematic reviews (SR) and meta-analyses (MAs). This work builds on a wider project of evidence synthesis on the impact of the COVID-19 pandemic on mental health, performed to inform a WHO Scientific Brief [10]. ## Umbrella review design We followed published guidelines on performing umbrella reviews [11–13]. An umbrella review consists of a systematic search and assessment of systematic reviews on a specific research question. This allows for the comparison of results of clinical outcomes and provides a clear picture on broad healthcare areas, possibly revealing whether the evidence base is consistent or contradictory. We registered a protocol to define the research methodology to inform the WHO Scientific brief that is publicly available on the Open Science Framework platform [14]. The present work is based on the ‘Question 3 - Are people living with pre COVID-19 existing mental health disorders at (increased) risk of severe illness and mortality and/or of contracting SARS-CoV-2 compared to any other population?’ of the protocol, with the addition of focusing on estimates that have adjusted for confounding. ## Literature search and study selection An information specialist performed a systematic search on Ovid MEDLINE All, Embase (Ovid), PsycINFO (Ovid), CINAHL, and Web of Science, from December 31, 2019 until October 6, 2021, using a search strategy based on a combination of keywords and text words for reviews, COVID-19, and mental disorders or problems. The search strategy can be found in the supplementary material. This search was designed to inform a wide panel of different research questions on COVID-19 and mental health. We deduplicated search results using Endnote software [15]. Entries were divided into groups, each screened in parallel and independently by two different authors using Rayyan [16]. In case of disagreement, records were conserved. The full text of the remaining articles was assessed against the inclusion and exclusion criteria, again in duplicate and independently. Disagreement was resolved by discussion or with a senior review author. ## Eligibility criteria and data extraction We considered for inclusion reviews that: [1] were systematic, as defined by having a systematic search on at least one bibliographic database, explicitly reporting primary studies selection criteria and providing a list and synthesis of included studies; [2] considered as primary studies either cohort or cross-sectional or case-control studies; [3] have been published in a peer-reviewed journal; [4] compared people with pre-existing mental disorders (i.e., with a diagnosis of mental disorder established before the pandemic onset) with those without a pre-existing mental disorder; and [5] assessed at least one of the following outcomes: (a) risk of contracting SARS-Cov-2 infection, (b) risk of severe COVID-19 illness, (c) risk of long-term physical symptoms after COVID-19 illness, (d) mortality from SARS-CoV-2 infection. For the outcome ‘risk of severe illness’, we considered definitions as provided by original systematic reviewers, or proxy outcomes such as hospitalization. We did not exclude reviews on the basis of language of publication. In the case of reviews considering several pandemics, we planned to include only those that reported separated data for the COVID-19 pandemic. We considered systematic reviews independently of whether a meta-analysis was performed. Two authors independently and in parallel extracted the following data: author, date of publication, timeframe covered by the search, population and comparator, design of included studies, inclusion and exclusion criteria, outcomes, risk of bias assessment strategy, how exposure to SARS-CoV-2 was established, number of included studies and participants, and outcomes of interest with I2 statistic as a measure of heterogeneity. In terms of meta-analysis, we aimed to summarize effect size in terms of adjusted Odds Ratio (aOR), if that was not available, we considered the most adjusted model available. ## Assessment of quality Two independent authors assessed the quality of the included systematic reviews using the AMSTAR 2 tool [17]. For systematic reviews that did not perform a meta-analysis, items 11, 12 and 15 were not considered. Also, small adaptations to some items were performed to make them more suitable to score (see the supplementary material). For an overall rating, we used the proposed scheme by Shea and colleagues that provides judgments of high, moderate, low or critically low confidence [17]. ## Selection and inclusion of systematic reviews The systematic search provided 46,284 record entries, reduced to 31,559 after deduplication. 939 records were selected for full text retrieval by screening titles and abstracts. Of these, nine were finally included after comparing full texts against inclusion and exclusion criteria. Figure 1 reports the details of the study selection process. Fig. 1PRISMA flowchart of studies selection ## Characteristics of included systematic reviews Of the nine included reviews, five performed a meta-analysis on at least one of the outcomes of interest [18–22], while the remaining four were of a narrative nature [23–26]. Table 1 summarises the characteristics of the included reviews. Table 1Characteristics of included systematic reviewsAuthor, yearDesignTimeframe covered by bibliographic searchDesign of included studiesPopulation / comparatorCountriesEffect size estimation methodsNumber of included StudiesOutcomesCeban 2021Systematic review with meta-analysisFrom inception to February 1, 2021Case-control and cohort studiesPeople with a mood disorder / people without mood disordersUSA, South Korea, Spain, Italy, Turkey, UK, IsraelRandom effect model pooling adjusted and crude ORs. Preference given to adjusted ORs when available. Subgroups analyses according to adjustment status21(15 aOR, 6 crude ORs)1- Susceptibility (laboratory testing, ICD-10, EHR, and/or clinical judgment) 2- Hospitalization 3- Occurrence of severe events (includes oxygen therapy) 4- DeathFond 2021Systematic review with meta-analysisFrom inception to February 12, 2021Population-based cohort studies based on medico-administrative health databases or a health care data warehousePeople with any mental disorder (includes substance abuse disorders and ADHD) / people without a mental disorderDenmark, France, Israel, South Korea, Spain, UK, USARandom effects models, separate pooling of aORs and crude ORs211- Mortality2- ICU admissionFornaro 2021Scoping review (no meta-analysis)From inception up to April 24, 2021Cross-sectional, prospective cohort, retrospective cohort studies, case reports/seriesPeople with Bipolar disorder / NAItaly, UK, Australia, USA, the Netherlands, India, China14Identified themes: 1- Impact of COVID-19-related stressors on BD2- Impact of COVID-19 on mental health service utilization among people with BD3- Impact of BD on the risk of acquiring SARS-cov-2 infection4- Engagement in preventative behaviors among people with BDKaraoulanis 2021Systematic review without meta-analysisFrom inception up to 21 March 2021UnreportedPeople with Schizophrenia / people without SchizophreniaUSA, Korea, France, Israel71- Susceptibility (being infected)2- MortalityLemieux 2020Rapid reviewFrom December 2019 up 18 August 2020Opinion pieces by experts (editorials, letters to the editor, position papers and other correspondences), rapid literature reviews, narrative (i.e., Non-systematic) literature reviews, empirical studies (surveys and descriptive/case studies)People with mental illness in secure settings / NAUSA, China, Italy, UK, France, Germany, Ireland, Singapore, Brazil, Canada, Spain, Australia, India, Israel, Poland, New Zealand, South Korea, Scandinavia, Switzerland69Narrative review on strategies, challenges and recommendations for dealing with the COVID-19 outbreak in secure settings for persons with mental illnessLiu 2021Systematic review with meta-analysisFrom inception to January 16, 2021Cohort, case-control, cross-sectional studies and case seriesPeople with mental disorders (includes sleep disorders and ADHD) / people without mental disordersUSA, Italy, Korea, UK, China, Iran, Brazil, Spain, Turkey, Israel, Switzerland, the Netherlands, Denmark, France, Belgium, Sweden, Russia, Peru, Germany, Malaysia, PolandRandom effect models pooling adjusted and crude ORs. Subgroups analyses according to adjustment status for ‘any mental disorder’ population only1491- Susceptibility (positive laboratory results, diagnosis in conjunction with clinical presentation)2- Illness severity (hospitalization, ICU admission, or requirement for other special treatment (e.g., oxygen therapy, mechanical ventilator, extracorporeal membrane oxygenation, and cardiopulmonary resuscitation))3- DeathMurphy 2021Scoping reviewFrom 1 January to 31 December 2020Primary research papers including qualitative, quantitative and mixed methods study designsPeople with mental disorders / NAItaly, USA, China, Canada, Germany, Spain, UK, Ireland, Australia, India, Switzerland, France, Iran, Turkey, Poland, Pakistan, Bosnia and Herzegovina30Narrative review of factors influencing health outcomes in people with pre-existing mental health conditions during the pandemic; impact of the pandemic on the health of people with pre-existing mental health conditions; Strategies or measure to support people with pre-existing mental health conditions during the pandemicToubasi 2021Systematic review with meta-analysisFrom inception to 15 February 2021Cohort, case controlPeople with mental disorders / people without mental disordersUK, Japan, South Korea, France, Scotland, Israel, USARandom effects models, pooling both adjusted and crude ORs. Sensitivity analysis including aOR only16 (aOR: 5)1- Death or Severe illness (ICU admission and ventilation)Vai 2021Systematic review with meta-analysisFrom 1 January 2020 to 5 March 2021Cross-sectional, longitudinal studiesPeople with mental disorders (includes substance use disorders and intellectual disabilities)/ people without mental disordersUK, Japan, South Korea, France, Scotland, Sweden, USARandom effects models pooling crude ORs. Sensitivity analysis polling aORs only.231- Death2- Hospitalization3- ICU admissionaORs: adjusted Odds Ratios, ADHD: Attention-deficit hyperactivity disorder, BP: Bipolar disorder, ICU: intensive care unit, NA: Not applicable, ORs: Odds Ratios, UK: United Kingdom, USA: United States of America, Systematic reviews with meta-analyses (MAs) generally had a search covering up to the first quarter of 2021. They included primary studies with mainly a cohort design, with the notable exceptions of the review by Liu and colleagues which considered case-series too and the review by Fond and colleagues which considered population-based cohort studies. MAs included a mean of 45 studies and a median of 21, the review by Liu and colleagues being an outlier with 149 included studies. Most of the MAs considered people with any mental disorder, possibly providing subgroups for specific disorders groups, while the review by Ceban and colleagues only focused on people with a mood disorder. Included studies covered a wide range of countries but with a general lack of representation from low- and middle-income countries. Infection by SARS-CoV-2 was defined by laboratory testing, ICD-10, electronic health records, or clinical judgment in the review by Ceban and colleagues and by laboratory results and diagnosis in conjunction with clinical presentation in the review by Liu and colleagues. Severe illness was defined as ICU admission, mechanical ventilatory support, oxygen therapy, extracorporeal membrane oxygenation, acute respiratory distress syndrome, and/or cardiopulmonary resuscitation by Ceban and colleagues and as hospitalization, ICU admission, or requirement for other special treatment (including oxygen therapy) by Liu and colleagues. For the review by Vai and colleagues, which does not have a “severe illness” outcome, we considered the outcome ‘hospitalisation’. For the review by Toubasi and colleagues, we considered the pooled mortality and severe illness outcome, as a separate estimate for mortality only was not available. Notably, no review was available to inform on the long-term physical symptoms after SARS-CoV2 infection. Four reviews reported estimates based on pooling adjusted ORs only [18, 19, 21, 22]. The review by Vai and colleagues reported a fully adjusted model only when considering people with any mental disorder, for the diagnostic groups we have then considered the ‘partially adjusted model’, where review authors considered aORs pooled together with crude ORs when an adjusted figure was not available from primary studies. The four narrative reviews varied in their specific design, with two scoping reviews [25, 26], one rapid review [23], and one systematic review without meta-analysis [24]. They considered a wide range of study designs with the review by Lemieux and colleagues considering opinion pieces and other reviews; as for population of interest, they considered people with bipolar disorder [25], people with schizophrenia [24], people with mental illness in secure settings [23], and generally people with mental disorders [26]. ## Quality of included reviews The AMSTAR 2 rated level of quality for all the MAs was “low”, with the exception of the review by Liu and colleagues with “high” and the review by Fond and colleagues with “critically low”. The review by Liu and colleagues did not have any weaknesses in critical items, while all other MAs did not report the list of excluded studies with reasons for exclusion; the review by Fond and colleagues also did not account for the impact of risk of bias in primary studies on the results. The level of quality for the narrative reviews was “critically low” with the exception of the review by Fornaro and colleagues (“low”), mainly because of a lack of risk of bias assessment and protocol registration. See the supplementary material for detailed AMSTAR 2 evaluation of the included reviews. ## Risk of SARS-CoV-2 infection Two MAs informed on the association between Sars-CoV-2 infection and having a pre-existing mental disorder compared to not having a pre-existing mental disorder (Fig. 2) [18, 19]. Fig. 2Risk of contracting SARS-CoV-2 infection. CI: confidence interval; K: number of included studies; OR: Odds ratio; n: total number of included participants; NR: not reported; : partially adjusted model For people with any mental disorder, Liu and colleagues found a statistically significant positive association (OR: 1.71, $95\%$ CI 1.09–2.69). For people with anxiety disorders, Liu 2021 and colleagues in a partially adjusted model found a statistically significant positive association although this effect size relies on two studies only (OR 1.63, $95\%$CI 1.44–1.85). For people with mood disorders, Liu 2021 and colleagues in a partially adjusted model found a statistically significant positive association (OR: 2.02, $95\%$CI 1.08–3.76), while Ceban and colleagues did not find a statistically significant association (OR: 1.50, $95\%$CI 0.75–2.99). For people with schizophrenia spectrum disorders, Liu and colleagues, in a partially adjusted model, did not find a statistically significant association (OR: 1.72, $95\%$CI 0.62–4.77). The level of statistical heterogeneity was been generally very high, with most I2 statistics over $95\%$, with the exception of the estimate for people with mood disorders by Liu and colleagues ($0\%$). I2 was not reported in Ceban et al., 2021. ## Risk of severe illness Three MAs informed on the association between a severe course of COVID-19 and having a pre-existing mental disorder compared to not having a pre-existing mental disorder (Fig. 3) [18–20]. Fig. 3Risk of severe COVID-19 illness. CI: confidence interval; K: number of included studies; OR: Odds ratio; n: total number of included participants; NR: not reported; : partially adjusted model For people with any mental disorder, both the review by Liu and colleagues and Vai and colleagues found a statistically significant positive association (OR: 1.32, $95\%$CI 1.19–1.46 and OR: 1.77, $95\%$CI 1.29–2.42, respectively). No study informed on people with anxiety disorders. For people with a mood disorder, Ceban and colleagues found no association (OR: 0.99, $95\%$CI: 0.80–1.24), Liu and colleagues in a partially adjusted model found a statistically significant positive association (OR: 1.34, $95\%$CI 1.08–1.67) while Vai and colleagues did not find a statistically significant association (OR 1.27, $95\%$CI 0.64–2.50). For people with a schizophrenia spectrum disorder, both the reviews by Liu and colleagues and Vai and colleagues did not find a statistically significant association (OR: 1.22, $95\%$CI 0.70–2.13 and OR: 1.38, $95\%$CI 0.61–2.94, respectively). The level of statistical heterogeneity was moderate to very high, with an I2 statistic between 65 and $100\%$, but low for the estimate for people with mood disorders by Vai and colleagues ($23\%$). I2 was not reported in Ceban et al., 2021. ## COVID-19 related mortality Four MAs informed on the association between mortality and having a pre-existing mental disorder compared to not having a pre-existing mental disorder (Fig. 4) [19–22]. Fig. 4COVID-19 related mortality risk. CI: confidence interval; K: number of included studies; OR: Odds ratio; n: total number of included participants; NR: not reported; : partially adjusted model; : considers mortality and severe illness together For people with any mental disorder all four MAs found a statistically significant positive association, ranging from an OR of 1.38 ($95\%$CI: 1.15–1.65) for the review by Fond and colleagues to 1.52 ($95\%$CI: 1.20–1.93) for the review by Toubasi and colleagues (which however considered in this outcome severe illness cases as well). For people with anxiety disorders both the reviews by Liu and colleagues, in a partially adjusted model, and by Vai and colleagues did not find a statistically significant association (OR: 1.16, $95\%$CI 0.75–1.79 and OR: 1.01, $95\%$CI 0.77–1.32, respectively). For people with mood disorders, all three informing MAs found a statistically significant positive association with ORs ranging from 1.36 ($95\%$CI: 1.15–1.79, Liu and colleagues, partially adjusted model) to 1.57 ($95\%$CI: 1.26–1.95 Vai and colleagues). For people with schizophrenia spectrum disorders both the reviews by Liu and colleagues, in a partially adjusted model, and by Vai and colleagues found a positive association (OR 2.28, $95\%$CI 1.40–3.73, and OR 1.68, $95\%$CI 1.29–2.18, respectively). The level of statistical heterogeneity has been generally moderate to considerable, with an I2 statistic between $60\%$ and $81.4\%$, but for the review by Vai and colleagues for the estimate for people with any mental disorder ($39\%$), and low for the estimate for people with anxiety disorders ($0\%$) and mood disorders ($22\%$). The I2 has not been reported in Ceban et al., 2021. ## Narrative reviews The narrative reviews corroborated meta-analytic findings by indicating that patients with serious pre-COVID-19 mental disorders show adverse health outcomes related to COVID-19 infection in terms of higher severity and mortality. Fornaro and colleagues [25] performed a scoping review on clinical and public health themes for people with bipolar disorder. They identified four major themes from the 14 included papers, among which was the impact of having bipolar disorder on the risk of contracting the SARS-CoV-2 infection. For this theme, one study reported an increased risk of infection contraction for people with bipolar disorder [27]. This study was considered by the MAs previously reported. Karaoulanis and Christodoulou performed a systematic review without meta-analysis on infection rates and mortality in people with schizophrenia spectrum disorders. The included studies suggest an increased infection and mortality risk; these studies have been included in the MAs previously reported. Lemieux and colleagues performed a rapid review on the management of COVID-19 for people with mental illness in secure settings. They considered a wide range of publications including opinion pieces and other reviews and report greater morbidity and mortality. Murphy and colleagues conducted a scoping review on the impact of COVID-19 and related restrictions on people with pre-existing mental disorders. They reported an increased infection risk in this population. ## Discussion To our knowledge this is the first umbrella review aimed at summarizing evidence on the impact of the COVID-19 pandemic on health outcomes in people with pre-existing mental disorders. Compared to the individual systematic reviews previously published, this umbrella review contextualizes the single piece of evidence, and provides an overview for different mental disorders, while systematic reviews have so far focused on specific diagnostic groups or considered only some of the outcomes of interest on the physical health repercussions of people with mental disorders. Two reviews considered several diagnostic groups and outcomes, however one employed only partially adjusted models within diagnostic groups [19], and the other could include a considerably smaller number of studies [20]. Overall, we found consistent results across the various reviews. Of interest, we found no previous reviews exploring the long-term effects of SARS-CoV-2 infection. Having any mental disorder was found to be associated with a higher likelihood of contracting the SARS-CoV-2 infection, a more severe COVID-19 illness, and higher mortality. We found different risks for different disorder groups. People with pre-existing anxiety disorders had an increased risk of contracting the SARS-CoV-2 infection; for people with mood disorders there was conflicting evidence of increased risk for severe COVID-19 course, and evidence of increased mortality; people with schizophrenia spectrum disorders had an increased risk of mortality, but there was no clear evidence of increased severe illness course for people with schizophrenia spectrum disorders. Notably, we found no review assessing the association between long-term physical symptoms after COVID-19 and having a pre-existing mental disorder. Having any mental disorder has been associated with an increased risk of contracting the SARS-CoV-2 infection. Looking at the data for the specific diagnostic groups, however, we observe that the association is confirmed for people with an anxiety disorder (and in an estimate based on only two studies) only, while for people with mood disorders the evidence is not consistent across reviews, and for people with a schizophrenia spectrum disorder the association is not statistically significant with a very wide confidence interval, making it hard to draw clear conclusions. Many people with a mental disorder live in shared households, nursing homes, therapeutic communities, or are inpatients. It has been noted that such settings pose challenges in putting into practice infection control measures [28]. In addition to that, people in an acute phase of a mental disorder might find it difficult to understand the need for and adhere to behavioural means of social distancing [29]. Still, these considerations regard mostly people with serious mental illness such as bipolar disorder or schizophrenia, adding difficulty to the interpretation of these results. Having a mental disorder was associated with having a more severe illness course and increased mortality. This could be partially explained by the increased risk of contracting the SARS-CoV-2 infection, but as the three disorder groups showed differential patterns, we argue that additional factors may have influenced this outcome. In particular, anxiety disorders were not associated with increased mortality despite an increased infection risk, and indeed the results are compatible with a random distribution in terms of effect sizes and confidence intervals. Mood disorders and schizophrenia spectrum disorders have been associated with increased mortality from COVID-19 without conclusive evidence of increased infection risk. Several factors might come into play in determining such a negative outcome. People with severe mental illness, such as bipolar disorder and schizophrenia, more frequently present with high BMI, diabetes mellitus, generally limited exercise tolerance, and are more likely to also smoke and have substance abuse disorders [6, 7, 30]. All of these are known risk factors for severe COVID-19 and for COVID-19 related mortality [2, 3, 31]. Although the use of adjusted odds ratios should have mitigated the impact of these risk factors in the estimate of effect sizes there was high heterogeneity in terms of adjusted factors used by primary studies. Therefore, it is possible that some factors, such as BMI, might not have been appropriately accounted for [32]. Many people with these disorders would have received an antipsychotic (and potentially benzodiazepines), with a possible negative impact on respiratory function [8]. However, there is limited evidence to support these hypotheses and future research is needed to fill this gap in knowledge. Moreover, we should take into consideration that socio-economic factors and stigma might have influenced the access to medical care of these persons [33]. Many primary studies were conducted during the initial phases of the pandemic when medical resources were scarce compared to needs, access to intensive care units was limited and subject to stringent triage, and no vaccine was available yet. The finding that for people with schizophrenia spectrum disorders there was no increased risk for a more severe disease course, but higher risk for COVID-19 related mortality, is somewhat puzzling. The reviews have used slightly different definitions of “severe illness”, but all considered it as an operational composite outcome where many different events qualified, including oxygen therapy. Oxygen therapy has been a widespread need in COVID-19 patients, as arterial hypoxemia is a major feature of the disease [34]. It is possible that the definition of severe illness was therefore excessively sensitive and did not allow for the identification of differences between people with and without a pre-existing mental disorder. Another possible explanation is that people with schizophrenia might have been disproportionately affected by sudden death events. We know that cardiac sudden death events have been shown to be associated with COVID-19 [35] and that people with schizophrenia have a higher frequency of cardiovascular disease which may predispose to such events [36]. However, there is no direct evidence, and the use of adjusted estimates should have compensated for the increased cardiovascular burden. Overall, the mismatch between the risk of severe disease and mortality reinforces the need to better investigate factors associated with the increased mortality risk. The findings of this umbrella review should be put into the context of some limitations. For the various diagnostic groups, the number of included primary studies has been limited, especially for anxiety disorders. Mood disorders include both depression and bipolar disorder, two disorders with different pharmacological approaches and neuro-inflammatory profiles. Additionally, information on the disease status of participants (remission or relapse) has generally not been considered. Low- and middle-income countries have been scarcely represented in the primary studies included by the reviews. The publication timeframe covered by the meta-analyses span to the first quarter of 2021; as such, these findings depict the first year of the pandemic. The* general landscape has since changed, thanks to health care systems adaptations, improved COVID-19 treatments and importantly, the introduction of vaccines. There is, however, conflicting evidence regarding vaccine uptake rates in people with mental disorders, possibly due to regional differences [37, 38]. The reviews considered different study designs for inclusion; notably, Liu and colleagues considered case series. The review by Vai and colleagues informed on many diagnostic groups, but in only partially adjusted models. There has been considerable heterogeneity across all outcomes, which the meta-analyses struggled to explain. Heterogeneity likely reflects methodological and qualitative differences among the primary studies. Moreover, this high heterogeneity and the generally wide confidence intervals limit the accuracy of the estimates. In light of these findings, there are relevant questions for future research. The new Omicron variant of the SARS-CoV-2 virus spreads more easily but usually causes less severe illness [39]. Assessing if this holds true for people with a pre-exiting mental disorder as a risk factor would be valuable. In parallel, there is still limited evidence on vaccine hesitancy and uptake rate in this population. Moreover, the topic of long-term physical symptoms after COVID-19 in people with mental disorders remains scarcely investigated. Addressing these three points would allow for more effective health care planning and possibly targeted intervention to address vaccine hesitancy. ## Conclusion The COVID-19 pandemic has affected people with pre-existing mental disorders more severely than people without in terms of physical health. People with pre-existing mental disorders, and especially those with mood or schizophrenia spectrum disorders, should have been considered at risk of severe course and increased mortality from COVID-19, similar to other identified risk groups such as patients with somatic health conditions. ## Electronic supplementary material Below is the link to the electronic supplementary material. Supplementary Material 1 ## References 1. Booth A, Reed AB, Ponzo S, Yassaee A, Aral M, Plans D, Labrique A, Mohan D. Population risk factors for severe disease and mortality in COVID-19: a global systematic review and meta-analysis. PLoS ONE (2021.0) 16 e0247461. DOI: 10.1371/journal.pone.0247461 2. Singh AK, Gillies CL, Singh R, Singh A, Chudasama Y, Coles B, Seidu S, Zaccardi F, Davies MJ, Khunti K. Prevalence of co-morbidities and their association with mortality in patients with COVID-19: a systematic review and meta-analysis. Diabetes Obes Metab (2020.0) 22 1915-24. DOI: 10.1111/dom.14124 3. Ssentongo P, Ssentongo AE, Heilbrunn ES, Ba DM, Chinchilli VM. Association of cardiovascular disease and 10 other pre-existing comorbidities with COVID-19 mortality: a systematic review and meta-analysis. PLoS ONE (2020.0) 15 e0238215. DOI: 10.1371/journal.pone.0238215 4. Lee SW, Yang JM, Moon SY, Yoo IK, Ha EK, Kim SY, Park UM, Choi S, Lee SH, Ahn YM. Association between mental illness and COVID-19 susceptibility and clinical outcomes in South Korea: a nationwide cohort study. Lancet Psychiatry (2020.0) 7 1025-31. DOI: 10.1016/S2215-0366(20)30421-1 5. Taquet M, Luciano S, Geddes JR, Harrison PJ. Bidirectional associations between COVID-19 and psychiatric disorder: retrospective cohort studies of 62 354 COVID-19 cases in the USA. Lancet Psychiatry (2021.0) 8 130-40. DOI: 10.1016/S2215-0366(20)30462-4 6. Dickerson FB, Brown CH, Kreyenbuhl JA, Fang L, Goldberg RW, Wohlheiter K, Dixon LB. Obesity among individuals with serious mental illness. Acta Psychiatr Scand (2006.0) 113 306-13. DOI: 10.1111/j.1600-0447.2005.00637.x 7. Suvisaari JM, Saarni SI, Perala J, Suvisaari JV, Harkanen T, Lonnqvist J, Reunanen A. Metabolic syndrome among persons with schizophrenia and other psychotic disorders in a general population survey. J Clin Psychiatry (2007.0) 68 1045-55. DOI: 10.4088/JCP.v68n0711 8. Wang MT, Tsai CL, Lin CW, Yeh CB, Wang YH, Lin HL. Association between Antipsychotic Agents and Risk of Acute Respiratory failure in patients with Chronic Obstructive Pulmonary Disease. JAMA Psychiatry (2017.0) 74 252-60. DOI: 10.1001/jamapsychiatry.2016.3793 9. Brandt J, Leong C. Benzodiazepines and Z-Drugs: an updated review of major adverse outcomes reported on in epidemiologic research. Drugs R D (2017.0) 17 493-507. DOI: 10.1007/s40268-017-0207-7 10. 10.Mental Health. and COVID-19: Early evidence of the pandemic’s impact: Scientific Brief [https://www.who.int/publications/i/item/WHO-2019-nCoV-Sci_Brief-Mental_health-2022.1] 11. Ioannidis JP. Integration of evidence from multiple meta-analyses: a primer on umbrella reviews, treatment networks and multiple treatments meta-analyses. CMAJ (2009.0) 181 488-93. DOI: 10.1503/cmaj.081086 12. Solmi M, Correll CU, Carvalho AF, Ioannidis JPA. The role of meta-analyses and umbrella reviews in assessing the harms of psychotropic medications: beyond qualitative synthesis. Epidemiol Psychiatr Sci (2018.0) 27 537-42. DOI: 10.1017/S204579601800032X 13. Fusar-Poli P, Radua J. Ten simple rules for conducting umbrella reviews. Evid Based Ment Health (2018.0) 21 95-100. DOI: 10.1136/ebmental-2018-300014 14. 14.COVID-19 and Mental Health. : An umbrella review of systematic reviews with or without meta-analyses [osf.io/jf4z2] 15. 15.The EndNote Team. : EndNote. In., vol. EndNote 20. Philadelphia, PA: Clarivate; 2013. 16. Ouzzani M, Hammady H, Fedorowicz Z, Elmagarmid A. Rayyan-a web and mobile app for systematic reviews. Syst Rev (2016.0) 5 210. DOI: 10.1186/s13643-016-0384-4 17. Shea BJ, Reeves BC, Wells G, Thuku M, Hamel C, Moran J, Moher D, Tugwell P, Welch V, Kristjansson E. AMSTAR 2: a critical appraisal tool for systematic reviews that include randomised or non-randomised studies of healthcare interventions, or both. BMJ (2017.0) 358 j4008. DOI: 10.1136/bmj.j4008 18. Ceban F, Nogo D, Carvalho IP, Lee Y, Nasri F, Xiong J, Lui LMW, Subramaniapillai M, Gill H, Liu RN. Association between Mood Disorders and Risk of COVID-19 infection, hospitalization, and death: a systematic review and Meta-analysis. JAMA Psychiatry (2021.0) 78 1079-91. DOI: 10.1001/jamapsychiatry.2021.1818 19. Liu L, Ni SY, Yan W, Lu QD, Zhao YM, Xu YY, Mei H, Shi L, Yuan K, Han Y. Mental and neurological disorders and risk of COVID-19 susceptibility, illness severity and mortality: a systematic review, meta-analysis and call for action. EClinicalMedicine (2021.0) 40 101111. DOI: 10.1016/j.eclinm.2021.101111 20. Vai B, Mazza MG, Delli Colli C, Foiselle M, Allen B, Benedetti F, Borsini A, Casanova Dias M, Tamouza R, Leboyer M. Mental disorders and risk of COVID-19-related mortality, hospitalisation, and intensive care unit admission: a systematic review and meta-analysis. Lancet Psychiatry (2021.0) 8 797-812. DOI: 10.1016/S2215-0366(21)00232-7 21. Toubasi AA, AbuAnzeh RB, Tawileh HBA, Aldebei RH, Alryalat SAS. A meta-analysis: the mortality and severity of COVID-19 among patients with mental disorders. Psychiatry Res (2021.0) 299 113856. DOI: 10.1016/j.psychres.2021.113856 22. Fond G, Nemani K, Etchecopar-Etchart D, Loundou A, Goff DC, Lee SW, Lancon C, Auquier P, Baumstarck K, Llorca PM. Association between Mental Health Disorders and Mortality among patients with COVID-19 in 7 countries: a systematic review and Meta-analysis. JAMA Psychiatry (2021.0) 78 1208-17. DOI: 10.1001/jamapsychiatry.2021.2274 23. Lemieux AJ, Dumais Michaud A-A, Damasse J, Morin-Major J-K, Nguyen TN, Lesage A, Crocker AG. Management of COVID-19 for persons with Mental Illness in Secure units: a Rapid International Review to inform practice in Québec. Victims & Offenders (2020.0) 15 1337-60. DOI: 10.1080/15564886.2020.1827111 24. Karaoulanis SE, Christodoulou NG. Do patients with schizophrenia have higher infection and mortality rates due to COVID-19? A systematic review. Psychiatriki (2021.0) 32 219-23. PMID: 34390554 25. Fornaro M, De Prisco M, Billeci M, Ermini E, Young AH, Lafer B, Soares JC, Vieta E, Quevedo J, de Bartolomeis A. Implications of the COVID-19 pandemic for people with bipolar disorders: a scoping review. J Affect Disord (2021.0) 295 740-51. DOI: 10.1016/j.jad.2021.08.091 26. Murphy L, Markey K, C OD, Moloney M, Doody O. The impact of the COVID-19 pandemic and its related restrictions on people with pre-existent mental health conditions: a scoping review. Arch Psychiatr Nurs (2021.0) 35 375-94. DOI: 10.1016/j.apnu.2021.05.002 27. Wang Q, Xu R, Volkow ND. Increased risk of COVID-19 infection and mortality in people with mental disorders: analysis from electronic health records in the United States. World Psychiatry (2021.0) 20 124-30. DOI: 10.1002/wps.20806 28. de Girolamo G, Cerveri G, Clerici M, Monzani E, Spinogatti F, Starace F, Tura G, Vita A. Mental Health in the Coronavirus Disease 2019 emergency-the italian response. JAMA Psychiatry (2020.0) 77 974-6. DOI: 10.1001/jamapsychiatry.2020.1276 29. Muruganandam P, Neelamegam S, Menon V, Alexander J, Chaturvedi SK. COVID-19 and severe Mental illness: impact on patients and its relation with their awareness about COVID-19. Psychiatry Res (2020.0) 291 113265. DOI: 10.1016/j.psychres.2020.113265 30. Hunt GE, Large MM, Cleary M, Lai HMX, Saunders JB. Prevalence of comorbid substance use in schizophrenia spectrum disorders in community and clinical settings, 1990–2017: systematic review and meta-analysis. Drug Alcohol Depend (2018.0) 191 234-58. DOI: 10.1016/j.drugalcdep.2018.07.011 31. Wang QQ, Kaelber DC, Xu R, Volkow ND. COVID-19 risk and outcomes in patients with substance use disorders: analyses from electronic health records in the United States. Mol Psychiatry (2021.0) 26 30-9. DOI: 10.1038/s41380-020-00880-7 32. Luykx JJ, Lin BD. Are psychiatric disorders risk factors for COVID-19 susceptibility and severity? A two-sample, bidirectional, univariable, and multivariable mendelian randomization study. Transl Psychiatry (2021.0) 11 210. DOI: 10.1038/s41398-021-01325-7 33. Fond G, Pauly V, Leone M, Llorca PM, Orleans V, Loundou A, Lancon C, Auquier P, Baumstarck K, Boyer L. Disparities in Intensive Care Unit Admission and Mortality among patients with Schizophrenia and COVID-19: a National Cohort Study. Schizophr Bull (2021.0) 47 624-34. DOI: 10.1093/schbul/sbaa158 34. Tobin MJ. Basing Respiratory Management of COVID-19 on physiological principles. Am J Respir Crit Care Med (2020.0) 201 1319-20. DOI: 10.1164/rccm.202004-1076ED 35. Tan Z, Huang S, Mei K, Liu M, Ma J, Jiang Y, Zhu W, Yu P, Liu X. The prevalence and Associated Death of Ventricular Arrhythmia and Sudden Cardiac death in hospitalized patients with COVID-19: a systematic review and Meta-analysis. Front Cardiovasc Med (2021.0) 8 795750. DOI: 10.3389/fcvm.2021.795750 36. Leucht S, Burkard T, Henderson J, Maj M, Sartorius N. Physical illness and schizophrenia: a review of the literature. Acta Psychiatr Scand (2007.0) 116 317-33. DOI: 10.1111/j.1600-0447.2007.01095.x 37. Mazereel V, Vanbrabant T, Desplenter F, Detraux J, De Picker L, Thys E, Popelier K, De Hert M. COVID-19 vaccination rates in a Cohort study of patients with Mental Illness in Residential and Community Care. Front Psychiatry (2021.0) 12 805528. DOI: 10.3389/fpsyt.2021.805528 38. Nilsson SF, Laursen TM, Osler M, Hjorthoj C, Benros ME, Ethelberg S, Molbak K, Nordentoft M. Vaccination against SARS-CoV-2 infection among vulnerable and marginalised population groups in Denmark: a nationwide population-based study. Lancet Reg Health Eur (2022.0) 16 100355. DOI: 10.1016/j.lanepe.2022.100355 39. Madhi SA, Kwatra G, Myers JE, Jassat W, Dhar N, Mukendi CK, Nana AJ, Blumberg L, Welch R, Ngorima-Mabhena N. Population Immunity and Covid-19 severity with Omicron variant in South Africa. N Engl J Med (2022.0) 386 1314-26. DOI: 10.1056/NEJMoa2119658
--- title: COVID-19 severity is related to poor executive function in people with post-COVID conditions authors: - Mar Ariza - Neus Cano - Bàrbara Segura - Ana Adan - Núria Bargalló - Xavier Caldú - Anna Campabadal - Maria Angeles Jurado - Maria Mataró - Roser Pueyo - Roser Sala-Llonch - Cristian Barrué - Javier Bejar - Claudio Ulises Cortés - Jose A. Bernia - Jose A. Bernia - Vanesa Arauzo - Marta Balague-Marmaña - Berta Valles-Pauls - Jesús Caballero - Anna Carnes-Vendrell - Gerard Piñol-Ripoll - Ester Gonzalez-Aguado - Carme Tayó-Juli - Eva Forcadell-Ferreres - Silvia Reverte-Vilarroya - Susanna Forné - Jordina Muñoz-Padros - Anna Bartes-Plan - Jose A. Muñoz-Moreno - Anna Prats-Paris - Inmaculada Rico - Nuria Sabé - Laura Casas - Marta Almeria - Maria José Ciudad - Anna Ferré - Manuela Lozano - Tamar Garzon - Marta Cullell - Sonia Vega - Sílvia Alsina - Maria J. Maldonado-Belmonte - Susana Vazquez-Rivera - Sandra Navarro - Eva Baillès - Maite Garolera - Carme Junqué journal: Journal of Neurology year: 2023 pmcid: PMC10026205 doi: 10.1007/s00415-023-11587-4 license: CC BY 4.0 --- # COVID-19 severity is related to poor executive function in people with post-COVID conditions ## Abstract Patients with post-coronavirus disease 2019 (COVID-19) conditions typically experience cognitive problems. Some studies have linked COVID-19 severity with long-term cognitive damage, while others did not observe such associations. This discrepancy can be attributed to methodological and sample variations. We aimed to clarify the relationship between COVID-19 severity and long-term cognitive outcomes and determine whether the initial symptomatology can predict long-term cognitive problems. Cognitive evaluations were performed on 109 healthy controls and 319 post-COVID individuals categorized into three groups according to the WHO clinical progression scale: severe-critical ($$n = 77$$), moderate-hospitalized ($$n = 73$$), and outpatients ($$n = 169$$). Principal component analysis was used to identify factors associated with symptoms in the acute-phase and cognitive domains. Analyses of variance and regression linear models were used to study intergroup differences and the relationship between initial symptomatology and long-term cognitive problems. The severe-critical group performed significantly worse than the control group in general cognition (Montreal Cognitive Assessment), executive function (Digit symbol, Trail Making Test B, phonetic fluency), and social cognition (Reading the Mind in the Eyes test). Five components of symptoms emerged from the principal component analysis: the “Neurologic/Pain/Dermatologic” “Digestive/Headache”, “Respiratory/Fever/Fatigue/Psychiatric” and “Smell/ Taste” components were predictors of Montreal Cognitive Assessment scores; the “Neurologic/Pain/Dermatologic” component predicted attention and working memory; the “Neurologic/Pain/Dermatologic” and “Respiratory/Fever/Fatigue/Psychiatric” components predicted verbal memory, and the “Respiratory/Fever/Fatigue/Psychiatric,” “Neurologic/Pain/Dermatologic,” and “Digestive/Headache” components predicted executive function. Patients with severe COVID-19 exhibited persistent deficits in executive function. Several initial symptoms were predictors of long-term sequelae, indicating the role of systemic inflammation and neuroinflammation in the acute-phase symptoms of COVID-19.” Study Registration: www.ClinicalTrials.gov, identifier NCT05307549 and NCT05307575. ### Supplementary Information The online version supplementary material available at 10.1007/s00415-023-11587-4. ## Introduction The post-coronavirus disease 2019 (COVID-19) condition (PCC) manifests 3 months after the onset of the disease, and presents with symptoms that remain for at least 2 months and cannot be explained by other diseases [1]. PCC is characterized by a wide variety of fixed or fluctuating symptoms, including cognitive complaints. While $60\%$-$80\%$ of patients with PCC report experiencing brain fog, memory, loss of attentional focus, and language disturbances [2–4], objective evaluations of people with PCC have shown impairments in attention, processing speed, memory, and executive functions [5–7]. The severity of COVID-19 and post-COVID cognitive impairment assessed through systematic neuropsychological assessments was first shown to be related in hospitalized patients with acute disease [8]. Intensive care unit (ICU) stay has been linked to reduced executive function, and the need for oxygen therapy has been linked to reduced performance in several cognitive measures 10–40 days after hospital discharge. Over the medium-long term, the general severity of acute illness has been related to residual cognitive deficits [9], treatment required for respiratory symptoms has been related to worse global cognitive performance [10], respiratory distress to lower processing speed [11], and hypoxemia to impaired long-term memory and visuospatial learning at five months but not at the one-year evolution [12]. Additional evidence has been obtained from studies comparing hospitalized and non-hospitalized patients. In comparison with non-hospitalized patients, hospitalized individuals are more likely to have impairments in attention, executive functioning, category fluency, and verbal memory [13] or slower processing speed [5]. Post-ICU patients showed a lower cognitive composite score than non-ICU patients. However, among non-ICU patients, the cognitive composite score did not differ between those who were hospitalized and those who were not [14]. In a similar study performed with a healthy control (HC) group, patients with severe PCC showed lower processing speed than those with mild-moderate PCC and healthy control participants [15]. In a Finnish study, both ICU and hospitalized patients underperformed patients treated at home in the total cognitive score at 6 months post-COVID. Moreover, ICU participants underperformed hospitalized patients and HCs in the attention domain [16]. However, in multiple investigations using samples from 18 to 478 hospitalized and non-hospitalized participants with acute illnesses, the severity of COVID-19 was not associated with cognitive impairments at 3–4 months [17–19]. According to a recent meta-analysis, patients admitted to the hospital during the acute infection were less likely to report post-COVID cognitive symptoms than outpatients three months (or more) after the disease [20]. Another aspect that requires consideration is the predictive value of acute symptomatology for long-term cognitive impairment. Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) can infect several human cell types, as seen by COVID-19's vast array of symptoms. Typical signs and symptoms include fever, fatigue, gastrointestinal issues, cough, sore throat, shortness of breath, myalgia, headaches, dizziness, and changes in smell and taste [21–23]. The etiology of cognitive dysfunction may originate from the pathophysiology of acute illness [24]. However, it is currently unknown whether the effects of COVID-19 on the brain are caused by virus invasion in the brain, oxygen deprivation of the brain, or the body's excessive inflammatory response in seriously affected individuals [25]. Acute symptoms, even if they are not neurological manifestations, could contribute to the understanding of post-COVID cognitive problems. In a split study, Guo et al. found that initial illness-related symptoms explained part of the variation in post-COVID subjective cognitive symptoms [3]. They then demonstrated how some aspects of neuropsychological performance can also be explained by acute sickness symptoms [26]. Despite the number of studies in the field, the relationship between cognitive outcomes and the severity of COVID-19 is still not completely clear, probably because the underlying mechanisms of the cognitive deficits identified are mostly unknown. This study aimed to clarify the relationship between the severity of COVID-19 and long-term cognitive outcomes in a large sample of participants, including a control group. Our second objective was to determine if the initial symptomatology can predict long-term cognitive impairment. Since COVID-19 symptoms are highly diverse and heterogeneous, we aimed to use principal component analysis to identify phenotypes or clinical symptoms that frequently coexist. ## Participants The NAUTILUS Project is a cross-sectional observational study of post-COVID-19 cognitive consequences based on multimodal data. We used the available clinical and neuropsychological data of the project in the present study. The sample consisted of 428 participants, including 319 participants with PCC and 109 HC individuals who were evaluated at the Neuropsychology and COVID Units of 16 Hospitals in Catalonia, Madrid, and Andorra, coordinated by the Consorci Sanitari de Terrassa (Terrassa, Barcelona, Spain). The inclusion criteria for the PCC group were as follows: (a) confirmed diagnosis of COVID-19 according to WHO criteria with signs and symptoms of the disease during the acute phase; (b) at least 12 weeks after infection; and (c) age over 18 years. Exclusion criteria were as follows: (a) established diagnosis of psychiatric, neurological, neurodevelopmental disorder, or systemic pathologies known to cause cognitive deficits before the episode of COVID-19, and (b) motor or sensory alterations that impeded neuropsychological examination. The HCs did not have COVID-19 (no positive test or compatible symptoms) and were selected after applying the same exclusion criteria as in the PCC group. All participants were native Spanish speakers. ## Procedure The overall procedure consisted of two sessions. In the first session, various questionnaires were administered to collect information about demographic factors and behaviors related to the participants’ health and medical history. Participants with PCC were questioned about their COVID-19 experience and the symptoms they were experiencing at the time of evaluation. For a list of typical acute COVID-19 acute symptoms, presence/absence and the number of days were recorded. We developed a scale from 0 to 4, in which 0 indicated the absence of the symptom and 4 indicated a long-lasting symptom. Next, participants rated the severity of their COVID-19 experience on a visual analog scale of 1–10. Later, they were asked about the symptoms they were currently experiencing (post-COVID symptoms) and whether these were minor, major, or different from those experienced in the acute phase. Finally, we asked them to report any other symptoms they had been experiencing and had not been covered in the interview. In the second session, each participant underwent a cognitive assessment with a comprehensive neuropsychological battery. We used the Montreal Cognitive Assessment (MoCA) for general cognitive screening [27, 28]. The WAIS-IV Digit Span subtest was used to measure verbal attention (digit span forward) and working memory (digit span backward) [29]. To assess verbal memory, we used the Spanish version of Rey's Auditory Verbal Learning Test (RAVLT) [30, 31]. Visual scanning, tracking, and motor speed were assessed by the Digit Symbol Coding Test (WAIS-III) [29]. Parts A and B of the Trail Making Test (TMT) were administered to measure visual scanning, motor speed and attention, and mental flexibility [32]. A difference score (B-A) that removed the speed element from the test evaluation was calculated [33]. The Controlled Oral Word Association Test (COWAT) [34, 35] was used to evaluate verbal fluency and language. The number of words beginning with the letters P, M, and R recalled in 1 min was recorded. Semantic fluency was evaluated using the category “animals” [36]. The number of correct animals reported in 1 min was counted. The interference score of the Stroop test was calculated as a measure of cognitive inhibitory control [37]. The Boston Naming Test (BNT) was used to evaluate language [38]. Social cognition was assessed with the Reading the Mind in the Eye Test (RMET) [39]. The Word Accentuation Test (TAP) was used to estimate the premorbid intelligence quotient (IQ) [40]. In addition to cognitive measures, we used the Chalder Fatigue Scale (CFQ) [41] to assess fatigue, the Generalized Anxiety Disorder 7-item scale (GAD-7) [42, 43] to assess anxiety, and the Patient Health Questionnaire-9 (PHQ-9) to assess depressive symptoms [44, 45]. The quality of life was evaluated by the WHOQOL-BREFF [46]. Trained neuropsychologists performed all evaluations. The recruitment was conducted between June 2021 and June 2022. The study was conducted with the approval of the Drug Research Ethics Committee (CEIm) of Consorci Sanitari de Terrassa (CEIm code: 02–20-107–070) and the Ethics Committee of the University of Barcelona (IRB00003099). All participants provided written informed consent. ## Statistical analyses Descriptive statistics were obtained for all variables of the study. Group differences in demographics were examined by conducting an analysis of variance (ANOVA). The Chi-square test was performed to compare binarized measures between the groups. One-way analysis of covariance (ANCOVA) with Bonferroni-adjusted post-hoc comparisons was performed to determine group differences in cognitive functioning. Graphical representations and descriptive statistics were used to study the assumptions. The effect size was calculated using the value partial eta squared (ήp2). To investigate if the cognitive symptoms of PCC were predicted by the acute-phase symptoms, principal component analysis (PCA) was performed first on 21 auto-reported acute-phase symptoms and on Z-scores of 15 neuropsychological variables to define the cognitive domains, followed by linear regressions (stepwise) with the acute symptom components as predictors and the neuropsychological components as dependent variables. Analyses were performed using IBM SPSS Statistics 27.0 (IBM Corporation, Armonk, NY, USA) and R Statistical Software (version 4.2.0; The R Foundation for Statistical Computing Platform). The critical level for statistical significance was set at α = 0.05. ## Sample demographics The 319 participants with PCC were classified into three groups according to the WHO clinical progression scale [47]: severe-intensive care unit (ICU-PCC) ($$n = 77$$), hospitalized (H-PCC) ($$n = 73$$), and mild (M-PCC) ($$n = 169$$) (Table 1). The participants’ sociodemographic characteristics and comorbidities are shown in Table 2. The M-PCC and the HC groups were equivalent in age and sex, had a higher proportion of women, and were younger than the ICU-PCC and the H-PCC groups. The three PCC groups showed no differences in formal education and estimated IQ. However, the education level and estimated IQ in the HC group were higher than those in all three PCC groups. Thus, age, sex, educational level, and estimated IQ were covariates in comparing cognitive results among the four groups. On average, all PCC participants had shown a positive test 320 days before their neuropsychological evaluation (SD = 156.66 days), and the ICU-PCC group had fewer days of evolution since the start of COVID-19 than the other two groups. Premorbid high blood pressure and obesity were more prevalent among ICU participants than the other PCC and HC groups. Table 1Clinical characteristics of the PCC groups based on the WHO clinical progression scaleWHO clinical progression scale scoreN (%)ICU-PCC6–977 ($24\%$) IMV38 ($49.4\%$) NIV or HFNC39 ($50.6\%$)H-PCC4–573 ($23\%$) NIV or HFNC25 ($34.2\%$) Mask or nasal prongs37 ($50.7\%$) No O2 treatment11 ($15.1\%$)M-PCC2–3169 ($53\%$) Disturbance of ADL139 ($82.3\%$) No disturbance in ADL30 ($17.7\%$)PCC post-COVID condition, ICU intensive care unit, H hospitalized, M mild, IMV invasive mechanical ventilation, NIV non-invasive ventilation, HFNC high-flow nasal cannula, ADL activities of daily livingTable 2Sociodemographic characteristics and comorbidities of the PCC severity and HC groupsICU-PCCn = 77H-PCCn = 73M-PCCn = 169HCn = 109Mean(SD)Mean(SD)Mean(SD)Mean(SD)FpPost-hocAge51.91(8.32)52.69(7.39)46.21(9.23)46.10(9.31)15.710.0001ICU > HCICU > MH > HCH > MEducation (years)13.14(3.19)13.34(3.50)14.26(3.28)15.57(2.93)11.070.0001HC > ICUHC > HHC > MEstimated IQ100.75(7.82)101.46(8.25)101.85 (7.73)104.79 (6.58)5.430.001HC > ICUHC > HHC > MTime of evolution269.75 (104.10)303.51 (131.49)350.56 (178.87)7.8420.0001M > ICUN (%)N (%)N (%)N (%)χ2pSex (% female)34($44.2\%$)35($50\%$)130($77\%$)84($77.1\%$)41.980.0001ComorbiditiesHeart disease4($5.2\%$)3($4.1\%$)4($2.4\%$)3($2.8\%$)Respiratory disease11($14.3\%$)10($13.7\%$)19($11.2\%$)5($4.6\%$)10.000.124Chronic kidney disease1($1.3\%$)1($1.4\%$)1($0.6\%$)0High blood pressure22($28.6\%$)13($17.8\%$)12($7.1\%$)5($4.6\%$)33.610.0001Dyslipidemia16($20.8\%$)13($17.8\%$)17($10.1\%$)11($10.1\%$)9.400.152Diabetes mellitus5($6.5\%$)7($9.6\%$)1($0.6\%$)3($2.8\%$)Obesity41($53.2\%$)26($35.6\%$)32($18.9\%$)16($14.7\%$)46.300.0001Chronic liver disease4($5.2\%$)5($6.8\%$)1($0.6\%$)0Tobacco smoking3($3.9\%$)2($2.7\%$)17($10.1\%$)27($24.8\%$)PCC = post-COVID condition, ICU intensive care unit, H hospitalized, M mild, HC healthy control, IQ intelligence quotientIntelligence estimated by means of Word Accentuation Test*Time of evolution is the days since first positive test ## Differences in cognitive performance Table 3 shows the fatigue, depression, anxiety, and quality of life scores for each PCC severity and HC group. The CFQ, PHQ-9, GAD-7, and WHOQOL-BREF scores were significantly different among groups. Post-hoc analysis showed that the CFQ, PHQ-9, and GAD-7 scores were higher in the PCC than in the HC group. Individuals in the M-PCC group had higher fatigue and depression levels than those in the H-PCC group. The quality of life assessed by the WHOQOL-BREFF was better in the HC group than in the PCC groups. We used fatigue, depression, and anxiety as covariates in the cognitive analysis. However, we also analyzed the data without these mood and fatigue variables (Supplementary Table 1).Table 3Intergroup differences in fatigue, mood, and quality of life measures adjusted for age, sex, educational level, and estimated IQICU-PCC($$n = 77$$)H-PCC($$n = 73$$)M-PCC($$n = 169$$)HC($$n = 109$$)Madj (SE)Madj (SE)Madj (SE)Madj (SE)Fpη2Post-hocBonferronipCFQ score5.88(0.49)5.23 (0.50)6.68 (0.32)1.80 (0.41)31.2070.00010.190ICU > HCH > HCM > HCM > H0.00010.00010.00010.017PHQ-9 score8.64 (0.70)7.35 (0.72)9.88 (0.45)3.35 (0.58)27.420.00010.172ICU > HCH > HCM > HCM > H0.00010.00010.00010.004GAD-7 score7.61(0.60)6.15(0.61)6.49(0.39)3.34(0.50)11.7730.00010.082ICU > HCH > HCM > HC0.00010.00010.0001WHOQOL-BREF score58.14 (12.38)58.23 (13.82)56.59 (13.19)67.20 (9.86)16.6500.00010.110ICU < HCH < HCM < HC0.00010.00010.0001PCC post-COVID condition, ICU intensive care unit, H = hospitalized, M mild, HC healthy control, CFQ Chandler Fatigue Scale, PHQ-9 Patient Health Questionnaire-9, GAD-7 Generalized Anxiety Disorder 7-item scale, WHOQOL-BREF World Health Organization Quality of Life Scale (General quality of life)Adjusted by age, sex, educational level, and estimated IQη2 effect size is as follows: η2 = 0.009, small; η2 = 0.059, medium; η2 = 0.139, large The groups showed statistically significant differences in MoCA, Digit symbol, TMT-B, TMT-B-A, phonetic fluency, and the RMET scores after controlling for age, sex, educational level, estimated IQ, fatigue, depression, and anxiety test scores. The ICU-PCC group performed worse in the MoCA, Digit symbol, TMT B, TMT-B-A, phonetic fluency, and RMET assessments than the HC group and obtained poorer results than the M-PCC group in the TMT-B and TMT-B-A assessments. The H-PCC group showed worse performance in the Digit symbol assessments than the HC group (Table 4 and Fig. 1).Table 4Adjusted means of the neuropsychological variables in the PCC severity and HC groupsICU-PCC($$n = 77$$)H-PCC($$n = 73$$)M-PCC($$n = 169$$)HC($$n = 109$$)Madj (SE)Madj(SE)Madj(SE)Madj(SE)Fpηp2Post-hocBonferronipMoCA score25.91(0.30)26.08(0.31)26.21(0.20)27.13(0.27)3.6060.0140.027ICU < HC0.021RAVLT total score45.28(0.98)43.35(1.00)44.82(0.66)46.87(0.88)2.3660.0710.018RAVLT immediate recall score8.76(0.32)8.92(0.33)9.17(0.22)9.09(0.29)0.3810.7670.003RAVLT delayed recall score8.63(0.36)8.74(0.36)9.07(0.24)9.23(0.32)0.5810.6280.004RAVLT recognition score12.41(0.29)12.06(0.29)12.21(0.19)12.55(0.26)0.6650.5740.005Digit spanforward score5.38(0.14)5.57(0.14)5.71(0.10)5.56(0.12)2.1090.0990.016Digit spanbackward score4.33(0.14)4.34(0.14)4.51(0.10)4.58(0.13)0.7730.5090.006Digit symbol score63.81(1.98)62.17(2.00)67.44(1.34)71.22(1.78)4.1760.0060.031ICU < HCH < HC0.0470.006TMT-A (time) score41.39(2.49)37.09(2.52)35.34(1.68)35.39(2.23)1.4400.2310.011TMT-B (time) score103.92(6.32)86.98(6.44)77.73(4.24)77.47(5.62)4.2680.0060.032ICU > HCICU > M0.0170.005TMT-B-A(time) score63.78(4.80)50.53(4.89)42.23(3.23)42.33(4.27)4.9720.0020.037ICU > HCICU > M0.0090.002Stroop word score94.20(2.56)93.33(2.56)94.16(1.71)96.66(2.27)0.3530.7870.003Stroop color score65.34(1.61)65.22(1.61)65.31(1.07)67.23(1.43)1.1020.3480.008Stroop interference score39.16(1.22)37.38(1.22)39.46(0.81)42.01(1.10)2.6060.0510.020Phonetic fluency (PMR) score40.14(1.38)42.60(1.40)42.79(0.93)45.73(1.24)2.8160.0390.021ICU < HC0.024Semantic fluency (animals) score20.73(0.62)20.58(0.63)21.41(0.42)22.71(0.59)2.4690.0620.019BNT score51.46(0.59)51.69(0.60)52.68(0.40)52.76(0.53)1.4100.2390.011RMET score21.37(0.45)22.37(0.45)22.65(0.30)23.64(0.40)4.4480.0040.033ICU < HC0.002PCC post-COVID condition, ICU intensive care unit, H hospitalized, M mild, HC healthy control, MoCA Montreal Cognitive Assessment, RAVLT Rey’s auditory verbal Learning Test, TMT Trail Making Test, BNT Boston Naming Test, RMET Reading the Mind in the Eyes TestAdjusted by age, sex, educational level, estimated IQ, Chalder Fatigue Scale (CFQ) score, Generalized Anxiety Disorder 7-item scale (GAD-7) score, and Patient Health Questionnaire-9 (PHQ-9) scoreηp2 effect size is as follows: ηp2 = 0.009, small; ηp2 =.0.059, medium; ηp2 =.0.139, largeFig. 1Cognitive profiles of the post-COVID condition severity groups and healthy controls. Healthy controls (HCs) are presented in green, ICU-PCC in blue, H-PCC in yellow, and M-PCC in red. Data are presented as means of Z-scores adjusted by age, sex, educational level, estimated IQ, fatigue, depression, and anxiety test scores. Lower Z-scores indicate poorer performance, except for TMT (time), where lower Z-scores indicate better performance. Statistically significant differences between groups are marked with an asterisk Table 5 shows the frequency of acute-phase symptoms for each severity group and all the PCC participants. ICU stay was associated with greater limb weakness and the presentation of delirium and psychotic symptoms. Hospitalization was associated with fever. A higher proportion of PCC participants at home had headache, muscle and joint pain, changes in smell and taste, nasal congestion, and sore throat. The three groups did not show differences in the perception of COVID-19 severity measured with the visual analog scale (ICU: mean = 7.91, SD = 2.22; H: mean = 7.86, SD = 1.65; M: mean = 7.05, SD = 2.41).Table 5Reported signs and symptoms in the acute infection period in the PCC severity groupsICU-PCCn = 77H-PCCn = 73M-PCCn = 169Totaln = 319N (%)N (%)N (%)χ2pN (%)Tiredness65 ($84.4\%$)68($93.2\%$)153($92.7\%$)4.9710.083286 ($90.8\%$)Fever66($85.7\%$)68($93.2\%$)121($72.9\%$)15.010.0001254 ($80.6\%$)Headache44($57.1\%$)50($68.5\%$)144($86.7\%$)27.1740.0001237 ($75.2\%$)Muscle and joint pain48($62.3\%$)46($63\%$)137($82.5\%$)15.8200.0001229 ($72.7\%$)Breathing issues55($71.4\%$)51($69.9\%$)97($58.8\%$)4.8790.087204 ($64.8\%$)Cough41($53.2\%$)48($65.8\%$)112($67.5\%$)4.7860.091200 ($63.5\%$)Loss of appetite40($51.9\%$)46($63\%$)101($60.8\%$)2.3010.316187(59.4)Loss of smell25($32.5\%$)30($41.1\%$)114($67.5\%$)31.3600.0001169 ($54.2\%$)Loss of taste27($35.1\%$)30($41.1\%$)106($62.7\%$)19.9820.0001163 ($52.2\%$)Shaking chills35($45.5\%$)37($50.7\%$)100($60.2\%$)5.1740.075171 ($54.3\%$)Limb weakness52($67.5\%$)34($46.6\%$)72($43.6\%$)12.4800.002158 ($50.3\%$)Paresthesia25($32.5\%$)27($37\%$)65(38.5)1.0090.604117($36.7\%$)Dizziness27($35.1\%$)32($43.8\%$)86($51.8\%$)7.2440.124145($46\%$)Nasal congestion28($36.4\%$)28($38.4\%$)86($51.1\%$)7.0000.030141 ($40.4\%$)Chest pain30($39\%$)29($39.7\%$)84($50.6\%$)4.0480.132142 ($45.1\%$)Sore throat22($28.6\%$)20($27.4\%$)93($56\%$)25.3100.0001134 ($42.5\%$)Diarrhea24($31.2\%$)28($38.4\%$)76($45.8\%$)4.8440.089128 ($40.1\%$)Nausea20($26\%$)25($34.2\%$)52($31.3\%$)1.2710.53097($30.8\%$)Conjunctival congestion11($14.3\%$)15($20.5\%$)40($24.1\%$)3.0710.21566($21.0\%$)Skin rash/Discoloration of fingers or toes9($11.7\%$)9($12.3\%$)33($16.1\%$)3.6260.16351($16.2\%$)Tachycardia6($7.8\%$)7($9.6\%$)17($10.1\%$)0.3230.85130($9.4\%$)Seizures1($1.3\%$)1($1.4\%$)02($0.6\%$)Stroke02($2.7\%$)1($0.6\%$)2($0.6\%$)Menstrual cycle issues1($12.5\%$)08($12.7\%$)9 ($11.5\%$)Depression37($48.1\%$)39($53.4\%$)79($47.6\%$)0.7310.694155 ($49.2\%$)Anxiety27($35.1\%$)32($43.8\%$)77($46.4\%$)2.7740.250136($43.2\%$)Psychotic symptoms24($31.2\%$)8($11\%$)4($2.4\%$)43.1160.000136($11.4\%$)Delirium30($39\%$)1($1.4\%$)032($10\%$)Obsessive–compulsive symptoms5($6.5\%$)4($5.5\%$)15($9\%$)24($7.6\%$)PCC post-COVID condition, ICU intensive care unit, H hospitalized, M mild, HC healthy control% women under 45 years ($$n = 78$$) ## Effect of acute symptoms on long-term cognition PCA with initial symptoms was performed with a varimax orthogonal rotation to facilitate interpretability. The Kaiser–Meyer–Olkin (KMO) value (0.834) and Bartlett's test of sphericity (χ2[210] = 1571.92; $p \leq 0.000$) indicated that the data were likely factorizable. PCA revealed five components with eigenvalues more significant than one, which explained $24.92\%$, $8.17\%$, $6.56\%$, $5.71\%$, and $5.11\%$ of the total variance and were classified as “Digestive/Headache” (nausea, loss of appetite, dizziness, diarrhea, shaking chills, and headache), “Respiratory/Fever/Fatigue/Psychiatric” (depressive symptoms, anxious symptoms, psychotic symptoms, breathing issues, fever, and fatigue), “Neurologic/Pain/Dermatologic” (paresthesia, skin problems, limb weakness, and muscle and joint pain), “Smell/Taste” (smell and taste symptoms), and “Cold” (nasal and conjunctival congestion and cough), respectively. The factor scores were computed through the regression method. The rotated (varimax) component loadings for the initial symptoms are shown in Table 6.Table 6Factor and loading in PCA of symptomsComponentsSymptom12345Nausea0.7000.314Loss of appetite0.602Dizziness0.599Diarrhea0.566Shaking chills0.491Headache0.445Depressive symptoms0.608Anxiety symptoms0.605Psychotic symptoms0.597Shortness of breath0.4890.440Fever0.484Fatigue0.436Skins symptoms0.699Paresthesia0.699Limb weakness0.507Muscle and joint pain0.3770.463Smell alterations0.906Taste alterations0.896Nasal congestion0.720Conjunctival congestion0.4630.638Cough0.3760.524Component 1: Digestive/Headache: nausea, loss of appetite, dizziness, diarrhea, shaking chills, and headacheComponent 2: Respiratory/Fever/Fatigue/Psychiatric: breathing issues, fever, depressive symptoms, anxious symptoms, psychotic symptoms, and fatigueComponent 3: Neurologic/Pain/Dermatologic: skins problems, limb weakness, paresthesia, and muscle and joint painComponent: Smell/Taste: smell alterations, taste alterationsComponent: Cold: nasal congestion, conjunctival congestion, coughBold indicates elements that charge above 0.5; the numbers that are not in bold are those that are loaded above 0.3 The scores for the Digestive/Headache, Respiratory/Fever/Fatigue/Psychiatric, and Smell/Taste components were significantly different among the severity groups. Post-hoc analysis showed that the Digestive/Headache score was higher in the M-PCC group than in the ICU-PCC group; the Respiratory/Fever/Fatigue/Psychiatric score was higher in the ICU-PCC and H-PCC groups than in the M-PCC group, and the Smell/Taste score was higher in the M-PCC than in the ICU-PCC and H-PCC groups (Fig. 2 and Supplementary Table 2).Fig. 2Violin plot for symptom factors across of PCC severity groups. Violin plots show the distribution for each symptom factor. Statistically significant differences were noted between PCC severity groups in Digestive/Headache, Respiratory/Fever/Fatigue/Psychiatric and the Smell/Taste score PCA with neuropsychological variables was performed with a direct oblimin rotation to facilitate interpretability. We excluded the scores obtained with the MoCA (a screening tool covering several cognitive domains) and the RMET (social cognition domain). All assumptions were met: overall KMO = 0.910 and Bartlett's test (χ2[105] = 3878.99, $$p \leq 0.0001$$). PCA revealed four components as the best factorial solution, which explained $72.71\%$ of the total variance ($45.14\%$, $12.86\%$, $8.14\%$, and $6.57\%$). We classified the four components as the following cognitive domains: executive function (TMT, Symbol Digit, Stroop task), verbal memory (RAVLT), attention and working memory (WM) (Digits span forward and backward), and language (Phonetic fluency, Semantic fluency, BNT). The regression approach was used to calculate the factor scores. Component loadings of the rotated solution are presented in Table 7. Figure 3 shows the profile of the cognitive domains for the PCC severity and HC groups corrected for age, sex, educational level, time of evolution, fatigue, and depression test scores. Table 7Factor and loading in PCA of neuropsychological variablesComponent1234Troop words (Z score)0.865Stroop colors (Z score)0.828TMT-A (Z score)− 0.786Stroop word-colors (Z score)0.719TMT-B (Z score)− 0.708Digit symbol (Z score)0.640RAVLT delayed recall (Z score)0.954RAVLT immediate recall (Z score)0.950RAVLT learning (Z score)0.865RAVLT recognition (Z score)0.826Digit span backward (Z score)0.896Digit span forward (Z score)0.876BNT (Z score)0.891Semantic fluency (animals) (Z score)0.692Phonetic fluency (PMR) (Z score)0.656Component 1: executive functions; Component 2: verbal memory; Component 3: attention and working memory (WM); Component 4: languageFig. 3Cognitive domain profiles for the post-COVID conditions severity groups. ICU-PCC in blue, H-PCC in yellow, and M-PCC in red. Data are presented as means of Z-scores (adjusted by age, sex, educational level, time of evolution, fatigue, and depression test scores) and deviation error bars. Lower Z-scores indicate poorer performance. Statistically significant differences were noted between PCC severity groups (marked with an asterisk) Linear regressions (stepwise) with the five acute symptom components as predictors and the neuropsychological components as dependent variables were performed. In addition to the four cognitive components, MoCA and RMET scores were used as dependent variables in multiple linear regression. The linear regression models were adjusted for potential confounders (age, sex, years of education, time of evolution, premorbid high blood pressure and obesity). As seen in Table 8, the “Neurologic/Pain/Dermatologic”, “Digestive/Headache”, “Respiratory/Fever/Fatigue/Psychiatric” and “Smell/ Taste” components added statistical significance to the prediction of MoCA scores. Executive function was predicted by the “Respiratory/Fever/Fatigue/Psychiatric,” “Neurologic/Pain/Dermatologic,” and “Digestive/Headache” components. The “Neurologic/Pain/Dermatologic” and “Respiratory/Fever/Fatigue/Psychiatric” components added statistical significance to the prediction of verbal memory scores, and the attention and WM component was predicted by the “Neurologic/Pain/Dermatologic” component. The language and social cognition components were not explained by any acute-phase symptom component but by the variables for demographic characteristics and premorbid conditions. Table 8Multiple linear regression models testing the association between acute symptoms and cognitive performanceMoCAFpR2adjPredictorsBetatp21.727 < 0.0010.282Constant− 7.9430.0001Years of education0.3877.4470.0001Neurologic/Pain/Dermatologic factor− 0.228− 4.4260.0001Digestive/ Headache factor− 0.154− 2.9920.003Respiratory/Fever/Fatigue/Psychiatric factor− 0.139− 2.7150.007Smell/ Taste− 0.132− 2. 5770.010Executive Function component (TMT, Stroop, Digit symbol)FpR2adjPredictorsBetatp17.006 < 0.0010.242Constant− 0.2790.780Years of education0.2274.0170.0001Respiratory/Fever/Fatigue/Psychiatric factor− 0.233− 4.3460.0001Neurologic/ Pain/ Dermatologic factor− 0.215− 4.0070.0001Digestive/ Headache factor− 0.171− 3.1820.002Age− 0.161− 2.8810.004Verbal memory component (RAVLT)FpR2adjPredictorsBetatp17.798 < 0.0010.233Constant0.4850.628Age− 0.246− 4.4760.0001Years of education0.1973.6390.0001Neurologic/ Pain/ Dermatologic factor− 0.188− 3.6590.0001Respiratory/Fever/Fatigue/Psychiatric factor− 0.152− 2.9430.004Sex (female)0.1302.4530.015Language component (Phonetic and semantic fluency, BNT)FpR2adjPredictorsBetatp28.508 < 0.0010.240Constant− 6.1590.0001Years of education0.2328.7800.0001Sex (male)− 0.131− 3.0640.002Age0.1692.9910.003Attention and WM component (Digit span forward, Digit span backward)FpR2adjPredictorsBetatp7.713 < 0.0010.124(Constant)0.4280.669Years of education0.1792.9840.003Neurologic/ Pain/ Dermatologic factor− 0.184− 3.2190.001Age− 0.157− 2.5850.010Sex (male)− 0.134− 2.3020.022Social cognition (RMET)FpR2adjPredictorsBetatp20.738 < 0.0010.119(Constant)− 5.8660.0001Years of education0.3305.9600.0001Obesity− 0.117− 2.1080.036MoCA Montreal Cognitive Assessment, TMT Trail Making Test, RAVLT Rey’s auditory verbal Learning Test, WM working memory, RMET Reading the Mind in the Eyes TestNeurologic/Pain/Dermatologic component: skin problems, limb weakness, paresthesia, and muscle and joint painRespiratory/Fever/Fatigue/Psychiatric component: breathing issues, fever, depressive symptoms, anxious symptoms, psychotic symptoms, and fatigueDigestive/Headache component: nausea, loss of appetite, dizziness, diarrhea, shaking chills, and headacheSmell/Taste component: smell and taste alterations ## Discussion The primary objective of the present study was to elucidate the link between COVID-19 severity and long-term cognitive outcomes. Previous studies have shown inconsistent results: some have reported a relationship [5, 8–16], while others did not identify any severity variable explaining cognitive performance [17–20]. Comparisons of these studies are challenging because their conclusions were drawn using various designs and methodologies. Moreover, only a few studies were specifically designed to examine this association [16, 18]. Some studies did not categorize patients according to the severity of their acute illness [8, 9, 11, 12, 19], or if they did, this categorization was only partially done or did not include a control group [5, 10, 13–15, 18]. Other studies only correlated the results of selected cognitive tests with severity assessments [10–12, 17]. Only one previous study compared groups according to the acute care environment and employed an HC group [16]. The neuropsychological performance profile obtained in our study with 428 participants showed a gradation in the expected direction: ICU-PCC < H-PCC < M-PCC < HC. After controlling for the variables that differed between groups, we found significant differences for the six neuropsychological tests. Post-hoc group comparisons showed that the significant differences arose mainly from the contrast between the HC and ICU-PCC groups. These tests measured global cognition (MoCA), executive functions-mental processing speed (Digit symbol, TMT-B, Phonetic Fluency), and social cognition (RMET). Additionally, the TMT-B test distinguished between ICU-PCC and M-PCC participants. Our findings partially agreed with those of a study with 213 participants and a similar design to ours [16]. In that study, the severity of COVID-19 was related to deterioration in an overall cognitive score and the attention domain. Some of the tests used to define the attentional domain in that study were also used in our study (Digit symbol, Stroop), while one test that was not used in the present study (Continuous Performance Test) was more sensitive to attention. Although depression and post-traumatic stress disorder were controlled in their overall score analysis, they were not controlled in the attention analysis. The authors of that study reported a relationship between executive function impairment and severity, but this relationship was observed only in men. In our sample, this relationship appeared regardless of sex. Our results referring to the relationship between executive function impairment and the severity of COVID-19 also agree with those of another study [13]. However, that study did not distinguish between hospitalized and ICU participants. The hospitalized patients in our sample did not differ from the outpatients in any test. In contrast, the ICU patients differed from the outpatients in two measures. Although the neuropsychological profile indicates impairment in the executive domain, tests grouped under executive functions can also be considered to involve processing speed. Several previous studies have related slowness with illness severity [5, 11, 15]. Our results support this relationship. Long-term slower mental speed processing has been linked to hypoxemia in individuals with acute respiratory distress syndrome (ARDS) [48]. Silent hypoxemia is a common feature in SARS-CoV-2 infections [49]. This trait caused delays in patient treatment, particularly during the first wave of the pandemic, which worsened the patients’ prognosis [50]. The integrity of white matter across the brain is related to processing speed and, more generally, to intellectual ability [51, 52]. White matter intensities have been shown to be associated with nocturnal hypoxemia [53] and hypoxic-ischemic brain injury in COVID-19-related ARDS [54]. Consistent with these findings, effects on the white matter have been reported to occur a year after COVID-19, specifically in the corona radiata, corpus callosum, and superior longitudinal fasciculus, particularly in post-ICU individuals [55]. COVID-19-induced white matter injury may be mediated by hypoxia as well as indirect viral invasion [56, 57], the systemic inflammatory response [58], or coagulopathy [59]. COVID-19 severity was not related to memory in our research, even though this relationship has been reported previously [12–14]. This result was unexpected due to the poor memory performance in the entire sample of PCC individuals in comparison with the HC group in our previous study [7]. The high prevalence of depression and anxiety symptoms and fatigue in our groups may explain this finding. In our previous study, fatigue, depression, and anxiety symptoms explained part of the memory performance variance in our PCC groups. Here, when we analyzed the data without controlling for emotional variables and fatigue, the H-PCC and M-PCC participants' learning was inferior to that in the HC group. In addition, the M-PCC group demonstrated poorer long-term memory and recognition than the HC group. Numerous studies have found a link between depression and memory problems in post-COVID individuals [60–62]. The causal connection between depression and memory impairment is, however, uncertain. In contrast to the findings reported in other studies [10, 63], we did not find differences in cognitive impairment between M-PCC and HC participants. The previous studies performed cognitive assessments of participants 3–6 months after the positive COVID-19 test. In contrast, cognitive assessments for the M-PCC group in the present study were performed an average of eleven months from the acute infection, when most participants may have recovered, at least in part. Most post-COVID symptoms decrease between 3 and 12 months [64], and this change has also been reported in the cognitive symptoms [12]. One study showed no differences between patients with mild- moderate COVID-19 and HCs 4 months post-infection. However, the groups in that study showed remarkable differences in anxiety, depression, and stress [62]. On the other hand, one study evaluating mild COVID-19 individuals at 11 months found several impaired cognitive measures relative to HC. Nevertheless, these authors did not assess whether their participants had fatigue or mood disturbances [65]. Different pathophysiological pathways for brain damage are probably implicated in mild, hospitalized, and critical cases of COVID-19. Despite the possibility of shared pathophysiological mechanisms, assumptions can be made for each group of patients. Mild cases may be caused directly by the virus (olfactory channel of entry) [66, 67]. The degree of systemic inflammation and level of hypoxemia are presumably higher in moderate-COVID-19 individuals [68]. In addition to more severe hypoxemia, systemic inflammation, and organ failure, brain injury may result from ICU therapies, including bed rest, life support equipment, and drugs in critical patients [69]. As a second aim, we investigated the relationship between acute symptoms and long-term cognitive outcomes. We identified five acute symptom components and found correlations between some of these components and long-term cognitive performance. “ Neurologic/Pain/Dermatologic,” “Digestive/Headache, Respiratory/Fever/Fatigue/Psychiatric,” and “Smell/Taste” predicted $28\%$ of the variance in global cognition. The “Neurologic/Pain/Dermatologic” component also explained $12\%$ of the variance in attention and WM, and the “Neurologic/Pain/Dermatologic” and “Respiratory/Fever/Fatigue/Psychiatric” components together explained $23\%$ of variance in verbal memory. Finally, $24\%$ of the variance in executive function was accounted for the “Neurologic/Pain/Dermatologic,” “Respiratory/Fever/Fatigue/Psychiatric,” and “Digestive/Headache” components. These three components included the symptoms limb weakness, paresthesia, muscle and joint pain, respiratory issues, fever, depression, anxiety, psychotic symptoms, fatigue, dizziness, and headache. These results may provide insights into the mechanisms underlying cognitive changes. The fact that the initial symptoms explain some of the variations in long-term cognition suggests that the brain regions responsible for these cognitive tasks were affected, and some of this impairment may have occurred during the acute phase of the illness. The Neurological/Pain/Dermatological, Respiratory/Fever/Fatigue/Psychiatric and Digestive/Headache components included symptoms that develop during systemic inflammation (pain, fatigue, fever, limb weakness, and paresthesia) and neuroinflammation (headache, dizziness, limb weakness, paresthesia, and mood alterations), although these components cannot be explained in terms of inflammation. We speculate that long-term cognitive impairment could have been caused by sustained systemic or neurological inflammation. Infections result in systemic inflammation and are associated with activation of microglial cells and the appearance of cognitive deficits. Neuroinflammation is caused by activation of microglial cells and the overexpression of proinflammatory cytokines, both of which are induced by the peripheral immune system [70]. In the initial phase of the study, critical patients showed impairment in global cognition, executive function, and social cognition. The variance of these cognitive areas is partially explained here by acute symptom variables. In addition to the inflammation mechanisms underlying the Neurologic/Pain/Dermatologic and Respiratory/Fever/Fatigue/Psychiatric factors, the added Digestive/Headache factor provides an alternative pathophysiological mechanism to explain executive function impairment. The hypothalamus regulates symptoms such as nausea and loss of appetite. SARS-CoV-2 has been suggested to use the nervus terminalis rather than the olfactory nerve as a direct pathway to infect the brain from the nasal cavity [71]. Bypassing the olfactory bulb, nerve terminal neurons project straight to locations in the brain, including the hypothalamus. Infection of the hypothalamus can produce these symptoms and allow the infection to spread to the medial prefrontal lobe [72], contributing to the pathophysiology of executive dysfunction. Although the results of the verbal memory, attention, and working memory tests did not differ significantly between groups in the initial phase of the study, models predicting the early symptomatology were identified for the components corresponding to these tests. However, language impairments were not predicted by any symptom factor. Instead, these impairments were predicted by demographic variables. Since emotion recognition is associated with the orbitofrontal cortex and temporal regions, we anticipated that the route of entry of the virus through the olfactory system could cause damage to these structures. However, patients with milder disease were more likely to experience impairments in smell and taste, whereas the ICU group, with the most severe condition, demonstrated the poorest social cognition. In this model, obesity, a chronic inflammatory condition [73] linked to the severity of COVID-19 [74], served as an explanatory variable. In addition to the risk posed by chronic inflammation, severely obese patients show considerable management issues in the ICU, particularly for the respiratory level [75]. Therefore, the impairment of the brain structures responsible for recognizing emotions should be attributed to indirect mechanisms, such as hematogenous pathways of virus entry to the central nervous system or systemic inflammatory mechanisms, and not to the direct action of the virus. We cannot rule out the possibility that sedated and intubated participants’ self-reported baseline symptoms were not as accurate as those with less severe COVID-19. Thus, we may have lacked complete and reliable data regarding symptoms such as anosmia/ageusia in severely ill patients. The limitations and strengths of the study require consideration while interpreting the findings. A major limitation refers to the collection of initial symptoms, which were self-reported through a questionnaire in the first session with the patient; the questionnaire itself was based on the symptoms most frequently reported in the literature. Thus, the presence of initial symptoms was recorded and scored retrospectively, which may have introduced recall bias. Moreover, we collected data for the presence and duration but not the intensity of each symptom. We did not use objective severity measures such as hypoxemia, days of sedation or weaning, or blood inflammatory levels, and the analysis was based solely on the reported symptoms. Since these factors may better explain the cognitive deficit, these variables will be examined in depth in future studies to understand the pathogenesis of cognitive dysfunction in PCC individuals. On the other hand, our sample size was reasonably large and represented the full spectrum of COVID-19 severity. Although the control group was not optimal because we had to control for some variables statistically, it was tested simultaneously with the COVID-19 participants, with the HCs experiencing the same pandemic circumstances. Unlike other studies, our participants were selected on the basis of inclusion criteria that precluded the presence of neurological, psychiatric, or systemic illnesses before COVID-19, conditions that could have influenced the cognitive findings. In addition, the cognitive examination was carried out in person with an extensive neuropsychological battery commonly used in the clinical context, which validated its applicability. In conclusion, the results of this study showed evident long-term impairments in patients with severe COVID-19 requiring ICU admission, although hospitalization per se did not involve long-term neuropsychological sequelae. Global cognition, executive function, and social cognition were the domains most affected by the severity of COVID-19. For the initial symptomatology, the factors Neurologic/Pain/Dermatologic, Respiratory/Fever/Fatigue/Psychiatric, and Digestive/Headache explained part of the variance of global cognition, attention and working memory, verbal memory and executive function. ## Supplementary Information Below is the link to the electronic supplementary material. Supplementary material 1 (DOCX 386 KB) ## References 1. Soriano JB, Murthy S, Marshall JC. A clinical case definition of post-COVID-19 condition by a Delphi consensus. Lancet Infect Dis* (2022.0) 22 e102-e107. DOI: 10.1016/S1473-3099(21)00703-9/ATTACHMENT/EF4FD06B-88FA-4A0C-B837-DCFEE13E82D7/MMC1.PDF 2. 2.Davis HE, Assaf GS, McCorkell L, et al (2021) Characterizing long COVID in an international cohort: 7 months of symptoms and their impact. EClinicalMedicine 38:101019. 10.1016/J.ECLINM.2021.101019/ATTACHMENT/499C606A-AE36-49F5-87DD-09E3B87369C9/MMC1.DOCX 3. 3.Guo P, Benito Ballesteros A, Yeung SP, et al (2022) COVCOG 1: factors predicting physical, neurological and cognitive symptoms in long COVID in a community sample. a first publication from the COVID and cognition study. Front Aging Neurosci 14:. 10.3389/FNAGI.2022.804922/FULL 4. 4.Ziauddeen N, Gurdasani D, O’Hara ME, et al (2022) Characteristics and impact of Long Covid: Findings from an online survey. PLoS One 17:e0264331. 10.1371/JOURNAL.PONE.0264331 5. 5.García‐Sánchez C, Calabria M, Grunden N, et al (2022) Neuropsychological deficits in patients with cognitive complaints after COVID‐19. Brain Behav e2508. 10.1002/brb3.2508 6. Delgado-Alonso C, Valles-Salgado M, Delgado-Álvarez A. Cognitive dysfunction associated with COVID-19: A comprehensive neuropsychological study. J Psychiatr Res (2022.0) 150 40-46. DOI: 10.1016/j.jpsychires.2022.03.033 7. 7.Ariza M, Cano N, Segura B, et al (2022) Neuropsychological impairment in post-COVID condition individuals with and without cognitive complaints. Front Aging Neurosci 1–12. 10.3389/fnagi.2022.1029842 8. 8.Almeria M, Cejudo JC, Sotoca J, et al (2020) Cognitive profile following COVID-19 infection: Clinical predictors leading to neuropsychological impairment. Brain Behav Immun Health 9:100163. 10.1016/J.BBIH.2020.100163 9. 9.Hampshire A, Chat A, Mphil M, et al (2022) Multivariate profile and acute-phase correlates of cognitive deficits in a COVID-19 hospitalised cohort. EClinical Medicine 47:1–10. 10.1016/j.eclinm.2022.101417 10. 10.Hampshire A, Trender W, Chamberlain SR, et al (2021) Cognitive deficits in people who have recovered from COVID-19. EClinicalMedicine 39:101044. 10.1016/j.eclinm.2021.101044 11. Cecchetti G, Agosta F, Canu E. Cognitive, EEG, and MRI features of COVID-19 survivors: a 10-month study. J Neurol (2022.0) 269 3400-3412. DOI: 10.1007/s00415-022-11047-5 12. Ferrucci R, Dini M, Rosci C. One-year cognitive follow-up of COVID-19 hospitalized patients. Eur J Neurol (2022.0) 29 2006. DOI: 10.1111/ENE.15324 13. Becker JH, Lin JJ, Doernberg M. Assessment of cognitive function in patients after COVID-19 infection. JAMA Netw Open (2021.0) 4 e2130645-e2130645. DOI: 10.1001/JAMANETWORKOPEN.2021.30645 14. Vannorsdall TD, Brigham E, Fawzy A. Cognitive dysfunction, psychiatric distress, and functional decline after COVID-19. J Acad Consult Liaison Psychiatry (2022.0) 63 133-143. DOI: 10.1016/J.JACLP.2021.10.006 15. Santoyo-Mora M, Villaseñor-Mora C, Cardona-Torres LM. COVID-19 Long-term effects: is there an impact on the simple reaction time and alternative-forced choice on recovered patients?. Brain Sci (2022.0) 12 1258. DOI: 10.3390/BRAINSCI12091258 16. Ollila H, Pihlaja R, Koskinen S. Long-term cognitive functioning is impaired in ICU-treated COVID-19 patients: a comprehensive controlled neuropsychological study. Crit Care (2022.0) 26 1-11. DOI: 10.1186/s13054-022-04092-z 17. Woo MS, Malsy J, Pöttgen J. Frequent neurocognitive deficits after recovery from mild COVID-19. Brain Commun (2020.0). DOI: 10.1093/braincomms/fcaa205 18. Wild CJ, Norton L, Menon DK. Disentangling the cognitive, physical, and mental health sequelae of COVID-19. Cell Rep Med (2022.0). DOI: 10.1016/J.XCRM.2022.100750 19. Miskowiak KW, Johnsen S, Sattler SM. Cognitive impairments four months after COVID-19 hospital discharge: Pattern, severity and association with illness variables. Eur Neuropsychopharmacol (2021.0) 46 39. DOI: 10.1016/J.EURONEURO.2021.03.019 20. 20.Premraj L, Kannapadi N v., Briggs J, et al (2022) Mid and long-term neurological and neuropsychiatric manifestations of post-COVID-19 syndrome: a meta-analysis. J Neurol Sci 434:120162. 10.1016/J.JNS.2022.120162 21. Singhavi H, Pai A, Mair M. SARS-Cov2: a meta-analysis of symptom distribution by continent in 7310 adult COVID-19 infected patients. Virusdisease (2021.0) 32 400-409. DOI: 10.1007/S13337-021-00699-Y/FIGURES/5 22. 22.Lechien JR, Chiesa-Estomba CM, de Siati DR, et al (2020) Olfactory and gustatory dysfunctions as a clinical presentation of mild-to-moderate forms of the coronavirus disease (COVID-19): a multicenter European study. European Archives of Oto-Rhino-Laryngology 23. 23.Mao L, Jin H, Wang M, et al (2020) Neurologic manifestations of hospitalized patients with coronavirus disease 2019 in Wuhan, China. JAMA Neurol. 10.1001/jamaneurol.2020.1127 24. 24.Proal AD, VanElzakker MB (2021) Long COVID or post-acute sequelae of COVID-19 (PASC): an overview of biological factors that may contribute to persistent symptoms. Front Microbiol 12:. 10.3389/FMICB.2021.698169 25. 25.Zhou Z, Kang H, Li S, Zhao X (2020) Understanding the neurotropic characteristics of SARS-CoV-2: from neurological manifestations of COVID-19 to potential neurotropic mechanisms. J Neurol 1. 10.1007/s00415-020-09929-7 26. 26.Guo P, Benito Ballesteros A, Yeung SP, et al (2022) COVCOG 2: cognitive and memory deficits in long COVID: a second publication from the COVID and Cognition Study. Front Aging Neurosci 14:. 10.3389/FNAGI.2022.804937/FULL 27. Nasreddine ZS, Phillips NA, Bédirian V. The Montreal Cognitive Assessment, MoCA: a brief screening tool for mild cognitive impairment. J Am Geriatr Soc (2005.0) 53 695-699. DOI: 10.1111/j.1532-5415.2005.53221.x 28. 28.Ojeda N, del Pino R, Ibarretxe-Bilbao N, et al (2016) [Montreal Cognitive Assessment Test: normalization and standardization for Spanish population]. Rev Neurol 63:488–496. 10.33588/rn.6311.2016241 29. 29.Wechsler D (2001) Wais III. Escala de inteligencia de wechsler para adultos. Manual de aplicación. TEA Ediciones. Departamento I+D., Barcelona 30. 30.Schmidt M (1996) Rey Auditory and Verbal Learning Test: A handbook. LosAngeles, CA 31. 31.Alviarez-Schulze V, Cattaneo G, Pachón-García C, et al (2022) Validation and normative data of the spanish version of the rey auditory verbal learning test and associated long-term forgetting measures in middle-aged adults. Front Aging Neurosci 14:. 10.3389/fnagi.2022.809019 32. Reitan RMM. Validity of the trail making test as an indicator of organic brain damage. Perceptual and Motos Skills (1958.0) 8 271-276. DOI: 10.2466/PMS.8.7.271-276 33. 33.Lezak, M. D., Howieson, D. B., Bigler, E. D., & Tranel D (2012) Neuropsychological assessment, 5th ed 34. Benton AL, Hamsher K. Multilingual Aphasia Examination (1989.0) 35. Peña-Casanova J, Quiñones-Úbeda S, Gramunt-Fombuena N. Spanish multicenter normative studies (NEURONORMA Project): norms for verbal fluency tests. Arch Clin Neuropsychol (2009.0) 24 395-411. DOI: 10.1093/ARCLIN/ACP042 36. Ardila A, Ostrosky-Solís F, Bernal B. Cognitive testing toward the future: the example of semantic verbal fluency (ANIMALS). Int J Psychol (2006.0) 41 324-332. DOI: 10.1080/00207590500345542 37. 37.Golden CJ (2005) Test de colores y palabras (Stroop). Madrid 38. Allegri RF, Mangone CA, Villavicencio AF. Spanish boston naming test norms. Clin Neuropsychol (1997.0) 11 416-420. DOI: 10.1080/13854049708400471 39. Fernández-Abascal EG, Cabello R, Fernández-Berrocal P, Baron-Cohen S. Test-retest reliability of the “Reading the Mind in the Eyes” test: A one-year follow-up study. Mol Autism (2013.0) 4 1-6. DOI: 10.1186/2040-2392-4-33/TABLES/2 40. Gomar JJ, Ortiz-Gil J, McKenna PJ. Validation of the Word Accentuation Test (TAP) as a means of estimating premorbid IQ in Spanish speakers. Schizophr Res (2011.0) 128 175-176. DOI: 10.1016/j.schres.2010.11.016 41. Jackson C. The Chalder Fatigue Scale (CFQ 11). Occup Med (Chic Ill) (2015.0) 65 86-86. DOI: 10.1093/OCCMED/KQU168 42. García-Campayo J, Zamorano E, Ruiz MA. Cultural adaptation into Spanish of the generalized anxiety disorder-7 (GAD-7) scale as a screening tool. Health Qual Life Outcomes (2010.0) 8 8. DOI: 10.1186/1477-7525-8-8 43. Spitzer RL, Kroenke K, Williams JBW, Löwe B. A Brief Measure for Assessing Generalized Anxiety Disorder: The GAD-7. Arch Intern Med (2006.0) 166 1092-1097. DOI: 10.1001/ARCHINTE.166.10.1092 44. Kroenke K, Spitzer RL, Williams JBW. The PHQ-9: validity of a brief depression severity measure. J Gen Intern Med (2001.0) 16 606. DOI: 10.1046/J.1525-1497.2001.016009606.X 45. Diez-Quevedo C, Rangil T, Sanchez-Planell L. Validation and utility of the patient health questionnaire in diagnosing mental disorders in 1003 general hospital spanish inpatients. Psychosom Med (2001.0) 63 679-686. DOI: 10.1097/00006842-200107000-00021 46. 46.WHOQOL-BREF (1996) WHOQOL-BREF : introduction, administration, scoring and generic version of the assessment : field trial version, December. World Health Organization 1–16 47. Marshall JC, Murthy S, Diaz J. A minimal common outcome measure set for COVID-19 clinical research. Lancet Infect Dis (2020.0) 20 e192-e197. DOI: 10.1016/S1473-3099(20)30483-7 48. 48.Hopkins RO, Weaver LK, Pope D, et al (1999) Neuropsychological Sequelae and Impaired Health Status in Survivors of Severe Acute Respiratory Distress Syndrome 49. Serrano R, Corbella X, Rello J. Management of hypoxemia in SARS-CoV-2 infection: lessons learned from one year of experience, with a special focus on silent hypoxemia. J Intensive Med (2021.0) 1 26-30. DOI: 10.1016/J.JOINTM.2021.02.001 50. Dhont S, Derom E, van Braeckel E. The pathophysiology of “happy” hypoxemia in COVID-19. Respir Res (2020.0) 21 1-9. DOI: 10.1186/s12931-020-01462-5 51. Brugulat-Serrat A, Salvadó G, Operto G. White matter hyperintensities mediate gray matter volume and processing speed relationship in cognitively unimpaired participants. Hum Brain Mapp (2020.0) 41 1309. DOI: 10.1002/HBM.24877 52. 52.Penke L, Mañiega SM, Bastin ME, et al (2012) Brain-wide white matter tract integrity is associated with information processing speed and general intelligence. Molecular Psychiatry 2012 17:10 17:955–955. 10.1038/MP.2012.127 53. Patel SK, Hanly PJ, Smith EE. Nocturnal hypoxemia is associated with white matter hyperintensities in patients with a minor stroke or transient ischemic attack. J Clin Sleep Med (2015.0) 11 1417. DOI: 10.5664/JCSM.5278 54. 54.Radnis C, Qiu S, Jhaveri M, et al (2020) Radiographic and clinical neurologic manifestations of COVID-19 related hypoxemia. J Neurol Sci 418:117119. 10.1016/J.JNS.2020.117119 55. Huang S, Zhou Z, Yang D. Persistent white matter changes in recovered COVID-19 patients at the 1-year follow-up. Brain (2022.0) 145 1830-1838. DOI: 10.1093/BRAIN/AWAB435 56. Paniz-Mondolfi A, Bryce C, Grimes Z. Central nervous system involvement by severe acute respiratory syndrome coronavirus -2 (SARS-CoV-2). J Med Virol (2020.0). DOI: 10.1002/jmv.25915 57. 57.Sankowski R, Mader S, Valdés-Ferrer SI (2015) Systemic Inflammation and the Brain: Novel Roles of Genetic, Molecular, and Environmental Cues as Drivers of Neurodegeneration. Front Cell Neurosci 9:. 10.3389/FNCEL.2015.00028 58. 58.Liddelow SA, Guttenplan KA, Clarke LE, et al (2017) Neurotoxic reactive astrocytes are induced by activated microglia. Nature 2017 541:7638 541:481–487. 10.1038/NATURE21029 59. 59.Wang Z, Yang Y, Liang X, et al (2020) COVID-19 Associated Ischemic Stroke and Hemorrhagic Stroke: Incidence, Potential Pathological Mechanism, and Management. Front Neurol 11:571996. 10.3389/FNEUR.2020.571996 60. Whiteside DM, Basso MR, Naini SM. Outcomes in post-acute sequelae of COVID-19 (PASC) at 6 months post-infection Part 1: Cognitive functioning. Clinical Neuropsychologist (2022.0) 36 806-828. DOI: 10.1080/13854046.2022.2030412 61. Ferrucci R, Dini M, Groppo E. Long-lasting cognitive abnormalities after COVID-19. Brain Sci (2021.0) 11 1-11. DOI: 10.3390/brainsci11020235 62. Mattioli F, Stampatori C, Righetti F. Neurological and cognitive sequelae of Covid-19: a four month follow-up. J Neurol (2021.0) 268 4422. DOI: 10.1007/S00415-021-10579-6 63. 63.Henneghan AM, Lewis KA, Gill E, Kesler SR (2022) Cognitive Impairment in Non-critical, Mild-to-Moderate COVID-19 Survivors. Front Psychol 13:770459. 10.3389/fpsyg.2022.770459 64. 64.Lorent N, Weygaerde Y vande, Claeys E, et al (2022) Prospective longitudinal evaluation of hospitalised COVID-19 survivors 3 and 12 months after discharge. ERJ Open Res 8:. 10.1183/23120541.00004-2022 65. 65.Stavem K, Einvik G, Tholin B, et al (2022) Cognitive function in non-hospitalized patients 8–13 months after acute COVID-19 infection: a cohort study in Norway. PLoS One 17:e0273352. 10.1371/JOURNAL.PONE.0273352 66. 66.Crunfli F, Carregari VC, Veras FP, et al (2022) Morphological, cellular, and molecular basis of brain infection in COVID-19 patients. Proc Natl Acad Sci U S A 119:. 10.1073/pnas.2200960119 67. 67.Douaud G, Lee S, Alfaro-Almagro F, et al (2022) SARS-CoV-2 is associated with changes in brain structure in UK Biobank. Nature 2022 604:7907 604:697–707. 10.1038/s41586-022-04569-5 68. Boldrini M, Canoll PD, Klein RS. How COVID-19 Affects the Brain. JAMA Psychiat (2021.0) 78 682-683. DOI: 10.1001/JAMAPSYCHIATRY.2021.0500 69. Colbenson GA, Johnson A, Wilson ME. Post-intensive care syndrome: impact, prevention, and management. Breathe (2019.0) 15 98-101. DOI: 10.1183/20734735.0013-2019 70. Walker KA, Ficek BN, Westbrook R. Understanding the role of systemic inflammation in Alzheimer’s disease. ACS Chem Neurosci (2019.0) 10 3340-3342. DOI: 10.1021/ACSCHEMNEURO.9B00333/ASSET/IMAGES/LARGE/CN-2019-00333X_0001.JPEG 71. Butowt R, von Bartheld CS. The route of SARS-CoV-2 to brain infection: have we been barking up the wrong tree?. Mol Neurodegener (2022.0) 17 1-4. DOI: 10.1186/S13024-022-00529-9/FIGURES/1 72. 72.Reppucci CJ, Petrovich GD (2015) Organization of connections between the amygdala, medial prefrontal cortex, and lateral hypothalamus: a single and double retrograde tracing study in rats. Brain Structure and Function 2015 221:6 221:2937–2962. 10.1007/S00429-015-1081-0 73. Chiappetta S, Sharma AM, Bottino V, Stier C. COVID-19 and the role of chronic inflammation in patients with obesity. Int J Obes (2020.0) 44 1790-1792. DOI: 10.1038/s41366-020-0597-4 74. 74.Chu Y, Yang J, Shi J, et al (2020) Obesity is associated with increased severity of disease in COVID-19 pneumonia: a systematic review and meta-analysis. Eur J Med Res 25:. 10.1186/S40001-020-00464-9 75. Selim BJ, Ramar K, Surani S. Obe Intensive Care Unit (2016.0) 44 146-156. DOI: 10.1080/21548331.2016.1179558
--- title: Seven bacterial response-related genes are biomarkers for colon cancer authors: - Zuming Xiong - Wenxin Li - Xiangrong Luo - Yirong Lin - Wei Huang - Sen Zhang journal: BMC Bioinformatics year: 2023 pmcid: PMC10026208 doi: 10.1186/s12859-023-05204-4 license: CC BY 4.0 --- # Seven bacterial response-related genes are biomarkers for colon cancer ## Abstract ### Background Colon cancer (CC) is a common tumor that causes significant harm to human health. Bacteria play a vital role in cancer biology, particularly the biology of CC. Genes related to bacterial response were seldom used to construct prognosis models. We constructed a bacterial response-related risk model based on three Molecular Signatures Database gene sets to explore new markers for predicting CC prognosis. ### Methods The Cancer Genome Atlas (TCGA) colon adenocarcinoma samples were used as the training set, and Gene Expression Omnibus (GEO) databases were used as the test set. Differentially expressed bacterial response-related genes were identified for prognostic gene selection. Univariate Cox regression analysis, least absolute shrinkage and selection operator-penalized Cox regression analysis, and multivariate Cox regression analysis were performed to construct a prognostic risk model. The individual diagnostic effects of genes in the prognostic model were also evaluated. Moreover, differentially expressed long noncoding RNAs (lncRNAs) were identified. Finally, the expression of these genes was validated using quantitative polymerase chain reaction (qPCR) in cell lines and tissues. ### Results A prognostic signature was constructed based on seven bacterial response genes: LGALS4, RORC, DDIT3, NSUN5, RBCK1, RGL2, and SERPINE1. Patients were assigned a risk score based on the prognostic model, and patients in the TCGA cohort with a high risk score had a poorer prognosis than those with a low risk score; a similar finding was observed in the GEO cohort. These seven prognostic model genes were also independent diagnostic factors. Finally, qPCR validated the differential expression of the seven model genes and two coexpressed lncRNAs (C6orf223 and SLC12A9-AS1) in 27 pairs of CC and normal tissues. Differential expression of LGALS4 and NSUN5 was also verified in cell lines (FHC, COLO320DM, SW480). ### Conclusions We created a seven-gene bacterial response‐related gene signature that can accurately predict the outcomes of patients with CC. This model can provide valuable insights for personalized treatment. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12859-023-05204-4. ## Background Colon cancer (CC) is a major human cancer accounting for approximately $10\%$ of all cancer cases [1]. In addition, colorectal cancer (CRC) is a leading cause of cancer and cancer-related deaths worldwide, with a multifactorial etiology that likely includes pro-carcinogenic bacteria [2]. Microbial dysbiosis is a hallmark of CRC and contributes to inflammation, tumor growth, and therapeutic response [3]. The composition of the intestinal microbiota is associated with both tumor development and anticancer immunity. Microbiota-specific T cells have a significant impact on anti-CRC immunity. The introduction of immunogenic intestinal bacteria can promote T follicular helper-associated antitumor immunity in the colon, suggesting therapeutic approaches for the treatment of CRC [4]. Several researchers preliminarily studied the predictive value of bacterial/microbiome-related genes in sepsis, anti-cancer drugs and cancers [5–7]. Metabolism, immune or cell death-related genes have been applied to construct prognostic models for CC or CRC [8–12], but bacterial response-related genes are rarely used as prognostic model to predict prognosis. The aim of this study was to develop a novel bacterial response prognostic risk score model to provide new insights into the diagnosis, evaluation, and treatment of CC. ## Bacterial response-related differentially expressed genes (DEGs) in CC samples We examined DEGs between normal colon and CC tissue using The Cancer Genome Atlas (TCGA) database. A total of 276 statistically significant differentially regulated genes were identified (Fig. 1A, Additional file 2: Table S1). Of these, 124 genes were increased and 152 were decreased in CC samples. Figure 1B shows the top 50 upregulated and downregulated bacterial response-related DEGs. 276 DEGs mianly enriched in defense response to bacterium and other Gene Ontology (GO) terms (Fig. 1C). They also enriched in many Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways,including Coronavirus disease—COVID-19, Tuberculosis, Pathogenic Escherichia coli infection, Staphylococcus aureus infection, Salmonella infection and so on(Fig. 1D).Fig. 1A Volcano map of bacterial response-related genes (red: upregulated genes, green: downregulated genes). B Differential expression heatmap between CC and normal colon tissues in the TCGA database. C GO analysis of 276 DEGs. D KEGG pathways analysis of 276 DEGs ## Prognostic risk score model construction in the TCGA cohort TCGA cohort samples were used as the training set to construct a prognostic model. Univariate Cox regression analysis was applied to the 276 bacterial response-related DEGs. Seventeen genes were related to prognosis (all $p \leq 0.05$) (Fig. 2A). The somatic mutation profiles of the 17 genes were collected. Mutations in these bacterial response-related genes were detected in 89 of 447 CC samples (frequency: $19.91\%$, Fig. 2B). GRIK2 had the highest mutation frequency; no mutations were identified in IFNE, WFDC10B, or HAMP in any of the CC samples. Further analyses demonstrated that most of these 17 genes had mutational co-occurrence relationships (Fig. 2C).Fig. 2Prognostic risk score model construction. A Forest plot of 17 bacterial response-related DEGs associated with prognosis. B The mutation frequency of the 17 bacterial response-related DEGs in 447 CC patients from TCGA. C Mutational co-occurrence (green) and exclusion (purple) analyses of the 17 bacterial response-related DEGs. D LASSO coefficients of the seven bacterial response-related DEGs. E Construction of the prognostic risk score model. F PCA of all bacterial response-related genes in the training set. G PCA of the bacterial response risk score to separate CC samples from normal samples in the training set; red indicates patients with a high risk score, and blue indicates patients with a low risk score. H and I OS of training and test sets in the two risk score groups. J Progression-free survival of the training set in the two risk score groups. K Relapse-free survival of the test set in the two risk score groups. L ROC curves of the predictive efficiency in the training set. M and N Forest plots of univariate and multivariate Cox regression analyses in the training set Expression requirements and least absolute shrinkage and selection operator (LASSO) Cox regression analyses were utilized together to decrease the number of genes incorporated into the model. Finally, seven genes (LGALS4, RORC, DDIT3, NSUN5, RBCK1, RGL2, and SERPINE1) were included in the prognostic risk score model (Figs. 2D and E). The following formula was used to calculate the risk score of each sample: risk score = (− 0.0849135799481108) × LGALS4 + (− 0.172643229881538) × RORC + (0.134531384115045) × DDIT3 + (0.144199163862461) × NSUN5 + (0.280831773407297) × RBCK1 + (0.484753562928209) × RGL2 + (0.193242156693903) × SERPINE1. The median risk score in the training set was set as the cutoff value. Based on each patient’s calculated risk score, 446 patients each were allocated to the low and high risk score groups. Principal-component analysis (PCA) results showed that the risk score had good discrimination in CC (Figs. 2F and G). Patients in the high risk score group had a poorer overall survival (OS) and progression-free survival (PFS) than those in the low risk score group (Figs. 2H and J). The 1-, 3-, and 5-year time-dependent receiver operating characteristic (ROC) curves confirmed that the prognostic risk score model had good predictive performance (area under the ROC curves [AUCs] of 0.712, 0.669, and 0.703, respectively; Fig. 2L). Figures 2M and N indicate that the risk score can serve as an independent risk indicator. We next validated this prognostic model in the test set (GSE39582). These patients were also divided into low and high risk score groups using the cutoff value from the training set, OS and relapse-free survival (RFS) prognosis were assessed (Fig. 2I and K). Patients in the high risk score group had worse prognosis than those in the low risk score group, demonstrating that the prognostic risk score model had good OS PFS and RFS predictive ability in CC across multiple datasets. The 1-, 3-, and 5-year time-dependent ROC curves, univariate and multivariate Cox regression analyses were also assesed in the test set (Additional file 1: Fig. S1A–C). ## Construction of a nomogram for predicting survival Figure 3A shows the OS predictive nomogram combining age, sex, pathological stage, and the prognostic risk score model. The 1-, 3-, and 5-year calibration curves confirmed the ability of the nomogram to predict OS in CC patients (Fig. 3B). The AUC of the nomogram was 0.771, indicating that it had better prognostic performance than age (AUC = 0.628), pathological stage (AUC = 0.675), or the prognostic risk score model (AUC = 0.702; Fig. 3C). In addition, the nomogram model was an independent prognostic factor for OS (Figs. 3D and E).Fig. 3The predictive value of the nomogram. A Nomogram predicting the OS of CC patients in TCGA. B Calibration plot of the nomogram. The nomogram-predicted survival is displayed on the x-axis, and the actual survival is displayed on the y-axis. C ROC curves for the risk score and clinicopathological characteristics. D, E Univariate and multivariate Cox regression analyses of factors associated with OS ## Association between the risk score and clinical characteristics The distribution of risk scores based on age, sex, pathological stage, and American Joint Committee on Cancer (AJCC) Tumor Node Metastasis (TNM) classification of malignant tumors stage [8] of the corresponding samples was analyzed. The risk score had no significant associations with age or sex (Fig. 4A, B). However, risk scores were consistently correlated with four clinical characteristics: T (tumor invasion), N (lymphoid metastasis), and M (distal metastasis) stages, as well as advanced pathological stage (all $p \leq 0.05$, Fig. 4C–F).Fig. 4A − F The association of the risk score with clinicopathological features, including age, sex, and T, N, M, and TNM stages ## Response to chemotherapy The pRRophetic R package was used to evaluate differences in chemosensitivity between the two risk groups in the training set. Figure 5A–E shows the calculated half-maximal inhibitory concentrations (IC50s) of several traditional anticancer drugs, including dasatinib, obatoclax mesylate, pazopanib, shikonin, and talazoparib, in the two risk score groups. These five drugs had lower IC50s in the high risk score group, which is suggestive of better efficacy. Fig. 5A–E Differential chemotherapeutic response based on the IC50s in the high and low risk score groups. IC50s of five chemotherapeutic agents (dasatinib, obatoclax mesylate, pazopanib, shikonin, and talazoparib) ## Gene set variation analysis (GSVA) and TP53 mutation Differences in biological behaviors between the high and low risk score groups were examined using GSVA enrichment. Basal cell carcinoma was enriched in the high risk score group. However, most metabolic pathways, including butanoate metabolism, fatty acid metabolism, sphingolipid metabolism, and starch and sucrose metabolism, were enriched in the low risk score group (Fig. 6A). Furthermore, CC patients with TP53 mutations had higher risk scores (Fig. 6B).Fig. 6A The heatmap of GSVA enrichment in the low and high risk score groups. B Differences in risk scores in TP53 wild-type and mutant samples ## Immune-related features in the low and high risk score groups The high risk score group was rich in M0 macrophages. However, the low risk score group was rich in CD4 memory resting T cells (Fig. 7A). Moreover, HLA was abundant in the high risk score group, suggesting that patients with a high risk score and immune suppression may benefit from immunotherapy (Fig. 7B).Fig. 7A Differences in infiltration of immune cells between the low and high risk score groups. B Differences in known immunity-related functions between the low and high risk score groups ($p \leq 0.05$) ## GO and KEGG pathway analyses and protein–protein interaction (PPI) networks of DEGs in the low and high risk score groups GO and KEGG pathway analyses of 60 DEGs were performed using the “GOplot” R package to better explore the function of these DEGs (Additional file 3: Table S2). The results of the GO analysis indicated that DEGs participated in the positive regulation of granulocyte chemotaxis, negative regulation of endopeptidase activity, negative regulation of peptidase activity, positive regulation of leukocyte chemotaxis, fatty acid transport, negative regulation of proteolysis, regulation of leukocyte chemotaxis, and regulation of granulocyte chemotaxis (Fig. 8A).Fig. 8A and B GO and KEGG enrichment analyses of 60 DEGs in the low and high risk score groups. C PPI network processed by Cytoscape. Red: DEGs with high expression in the high risk score group; green: DEGs with high expression in the low risk score group. D Top 10 hub genes in cytoHubba analysis (red and yellow) The KEGG pathway analysis indicated that DEGs were enriched in gastric cancer, malaria, phagosome, cell adhesion molecules, Wnt signaling pathway, PPAR signaling pathway, and ECM-receptor interaction (Fig. 8B). The DEGs in the high and low risk score groups were analyzed using the STRING online database. The PPI network shown in Additional file 1: figure S2 was plotted using DEGs. These DEGs in Fig. 8 C and D may have protein–protein interactions. Figure 8C shows DEG interactions, where red indicates genes increased in the high risk score group and green indicates genes increased in the low risk score group. Figure 8D shows the 10 hub genes selected in the network, including CLCA1, FABP4, KRT14, REG4, RETNLB, S100A7, SFRP2, SPINK4, WIF1, and WNT10A. REG4 and S100A7 were upregulated in CC samples compared to normal colon samples, whereas CLCA1, FABP4, KRT14, RETNLB, SFRP2, and WNT10A were downregulated in CC samples (Additional file 1: Fig. S3).K-M analysis indicated that REG4, S100A7, CLCA1, FABP4, RETNLB, SFRP2, and WNT10A expressions were significantly related to CC patient prognosis (Additional file 1: Fig. S4). ## Expression, diagnostic value, and prognostic value of the seven prognostic model genes We further characterized the seven prognostic model genes in the TCGA and GEO datasets. DDIT3, NSUN5, RBCK1, RGL2, and SERPINE1 were increased in CC samples compared to normal samples in the TCGA cohort, whereas LGALS4 and RORC were decreased in CC samples (Fig. 9A). The same trend was observed in the GSE44076 dataset (Fig. 9B). In the TCGA cohort and GSE44076 dataset, the expression of the seven prognostic model genes could distinguish cancer tissue from normal tissue (Figs. 9C, D). LGALS4 and NSUN5 had the best accuracy in differentiating CC and normal tissues (AUCs: TCGA, LGALS4, 0.968 and NSUN5, 0.947; GSE44076, LGALS4, 0.974 and NSUN5, 0.991). Moreover, each of the seven prognostic model genes could be individually used to stratify OS (Fig. 9E–K).Fig. 9A Expression of the seven prognostic model genes in normal and cancer tissue samples in the TCGA database. B Expression of the seven prognostic model genes in GSE44076 (98 CC tissues and 148 normal colon tissues). C and D Diagnostic ROC curves for the expression of each prognostic gene in the TCGA and GSE44076 datasets between CC tissues and normal colon tissues. E–K Kaplan–Meier curves of OS for the seven prognostic model genes in TCGA. ( $p \leq 0.001$) ## Identification of coexpressed prognostic long noncoding RNAs (lncRNAs) A total of 1832 significantly differentially expressed lncRNAs were identified in the TCGA cohort; 1594 lncRNAs were upregulated in CC samples and 238 were downregulated in CC samples (Additional file 4: Table S3). A total of 999 lncRNAs were significantly associated with OS, of which 90 were coexpressed with the seven prognostic model genes (Fig. 10A), including C6orf223 and SLC12A9-AS1 (Figs. 10B–D).Fig. 10A Ninety lncRNAs were coexpressed with the seven prognostic model genes. Gray lines indicate negative correlation; red lines indicate positive correlation. B Expression of the coexpressed lncRNAs C6orf223 and SLC12A9-AS1 in TCGA. C and D Kaplan–Meier curves of OS based on C6orf223 and SLC12A9-AS1 expression in TCGA. (* $p \leq 0.001$) ## qPCR validation in cell lines and colon tissues The expression of the seven prognostic genes and two prognostic lncRNAs was verified in 27 CC and matched normal tissues. NSUN5, RGL2, SERPINE1, C6orf223, and SLC12A9-AS1 had higher expression in CC tissues than in normal tissues, whereas LGALS4 and RORC had lower expression in CC tissues (all $p \leq 0.05$; Fig. 11A). This was consistent with the results of the bioinformatics analysis. NSUN5 expression was higher in COLO320DM and SW480 than in FHC (Fig. 11C), and LGALS4 expression was lower in COLO320DM and SW480 (Fig. 11B).Fig. 11The bioinformatics results were validated using quantitative PCR in CC tissues and cell lines. A Relative expression of the seven prognostic model genes and their two coexpressed lncRNAs in CC and normal colon tissues. B, C Relative expression of LGALS4 and NSUN5 in three cell lines (FHC, COLO320DM, and SW480). (* $p \leq 0.05$, $p \leq 0.001$, $p \leq 0.001$) ## Discussion The interactions and relationships between CC and bacteria are complex. Chronic and low-grade inflammation associated with persistent bacterial infections have been linked to the development of colon tumors [13]. Each tumor type may have a specific microbiome. Intratumoral bacteria are mostly intracellular and present in both cancer and immune cells. Manipulation of the tumor microbiome may also affect tumor immunity and response to immune therapy. Therefore, better understanding bacterial response effects may pave the way for novel treatment options for cancer patients [14]. Previous studies found that antimicrobial genes had strong correlation with sepsis and may predict sepsis [5]; bacterial infection related genes may predict safety and efficacy of Immunotherapy [6]. Bacterial infection is one kind of common infection inducing immune. Immune—related genes have been applied to construct prognostic models for CC [12]. Our seven-gene bacterial response‐related prognostic model can accurately predict the OS of CC patients. We used data from the TCGA database to create a prognostic risk score model to predict the OS of CC patients. Our risk model included five genes expressed at high levels in cancer tissues and two genes expressed at low levels in cancer tissues. Patients with CC in the high risk score group had poorer OS than those in the low risk score group. The same result was obtained using a dataset from the GEO database, demonstrating that the prognostic risk score model can predict patient survival. A high risk score was also associated with more advanced T, N, M, and TNM stages. Further, these seven genes could also independently predict the prognosis and diagnosis of CC. By comparing immune status and response to therapy between the low and high risk score groups, we also showed that the prognostic risk score model can also distinguish immune and therapeutic differences in CC. Patients with a high risk score were also more likely to have mutated TP53. Owing to the significant differences between the low and high risk score groups, DEGs in the two groups were further studied. Seven hub genes, REG4, S100A7, CLCA1, FABP4, RETNLB, SFRP2, and WNT10A, were all significantly associated with CC patient prognosis and differentially expressed in normal and tumor tissues. Previous studies had also demonstrated that these prognostic model genes are closely related to diseases. LGALS4 is associated with multiple cancer types and is a possible prognostic and diagnostic marker of colon adenocarcinoma (COAD) [15]. This is consistent with our findings. LGALS4 is expressed at significantly higher levels in adjacent normal tissues than in CC tissues [16], and it has been suggested to function as a tumor suppressor in CRC [17]. RORC is a critical transcription factor for the generation of pro-inflammatory cytokines, which are closely related to the pathogenesis of autoimmune diseases [18]. RORC is also closely related to inflammatory reaction and colitis [19, 20]. DDIT3 is a key stress response regulator activated in diverse settings, including DNA damage, ER stress, hypoxia, and nutrient deprivation [21, 22]. DDIT3 also plays an important role in the integrated stress response [23]. The summary actions of DDIT3 increase the survival of cancer cells when exogenous glutamine is limited [21]. NSUN5 is a conserved RNA methyltransferase that belongs to the Nop2/SUN domain family. The loss of NSUN5 decreases growth, cell size, proliferation, and bulk protein translation [24]. NSUN5 is overexpressed in CRC, and NSUN5 increases CRC proliferation and progression mostly through cell cycle regulation. Knockdown of NSUN5 inhibits tumor growth in vivo and in vitro [25]. High expression of RBCK1 is associated with poorer overall outcomes in patients with CRC. Moreover, RBCK1 suppression markedly reduces stemness in CRC [26]. RGL2 is significantly upregulated in primary tumors compared to normal tissues and serves as a poor prognostic marker in patients with CRC. Cell-based and animal experiments have further demonstrated that RGL2 acts as a driver to promote the metastatic progression of CRC, most likely by preventing the degradation of β-catenin and KRAS [27]. SERPINE1 is associated with cell migration and cell death. Increased SERPINE1 expression induces the expression of MMP1 and increases cell motility. The administration of agents that inhibit SERPINE1 should also be considered to reduce the risk of cancer cell metastasis [28]. In a previous study, C6orf223 (also known as LINC03040) was confirmed to be continuously upregulated in CC tissues compared to normal colon tissues [29]. Our study also confirmed that C6orf223 was significantly upregulated in CC tissues and predicted poor prognosis for patients with CC. In addition, we demonstrated that SLC12A9-AS1 was highly expressed in CC tissues by qPCR. However, the potential mechanism by which C6orf223 and SLC12A9-AS1 affect the prognosis of CC patients remains unclear. Our study has some limitations. First, our results were mainly based on bioinformatics analysis, and further in vivo and in vitro experiments are needed. Second, some factors that may affect CC prognosis were not examined, including weight, diet, and family history [30]. Finally, there was no significant difference in DDIT3 or RBCK1 expression between the two risk score groups, which may be due to a bias in the sample distribution. ## Conclusions Our seven-gene signature based on bacterial response-related genes demonstrated favorable predictive ability in both the training and test sets. The seven genes were all independent prognostic and diagnostic factors. Therefore, we recommend this signature to evaluate the prognostic risk of patients with CC. ## Datasets COAD patient genomic data were downloaded from the TCGA database (https://portal.gdc.cancer.gov/). Gene expression data from 41 normal and 473 CC cases were selected for the analyses. The corresponding clinical data and somatic mutation data were obtained from the TCGA database. The external dataset for the prognostic test set of 562 CC patients was derived from the GEO database (https://www.ncbi.nlm.nih.gov/geo/) (GSE39582). Data from 98 CC tissues and 148 normal colon tissues for the diagnostic test set was obtained from GSE44076. The expression data was standardized to mRNA and lncRNA transcript fragments per kilobase million using Perl software. We searched the keywords “bacteria & bacterium” in the Molecular Signatures Database (MSigDB) (https://www.gsea-msigdb.org/gsea/msigdb/human/search.jsp) [31–34]. Three GSEA gene sets (GOBP_DEFENSE_RESPONSE_TO_BACT- ERIUM; GOBP_T_HELPER_17_TYPE_IMMUNE_RESPONSE; GSE20151_CTRL_VS_FUSOBACT_NUCLEATUM_NEUTROPHIL_DN) and 611 bacterial response- related genes were identified (Additional file 5: Table S4). ## Prognostic risk score model construction and verification Bacterial response-related genes that were differentially expressed between normal and cancer tissue samples were identified using the “limma” R package (\|logFC\|> 0.585 and false discovery rate [FDR] < 0.05). The TCGA cohort samples were used as the training set, and GSE39582 samples were used as the test set. The differentially expressed bacterial response-related genes were combined with the corresponding prognosis results. These DEGs wrer conducted GO and KEGG enrichment analyses ($p \leq 0.05$) [35–38]. Intersecting differentially expressed bacterial response-related genes in the training and test sets were identified. Prognostic* genes were identified from the intersecting genes using univariate Cox regression analysis of the training set, and genes with $p \leq 0.05$ were selected. The mutations and correlations of the prognostic genes of the training set CC samples were analyzed using the Maftools package in R. Seven genes were chosen for LASSO Cox regression, which requires that the average gene expression of cancer and normal is greater than 1, and the average expression of high-risk genes is higher in the cancer group, whereas the average expression of low-risk genes is higher in the normal group. Then, based on these seven genes, prognostic risk score model was constructed to predict OS. The following equation was performed to calculate the risk score of each sample:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\text{Risk}}\;{\text{score}} = \sum\nolimits_{1}^{i} {({\text{Coefi }} \times {\text{ExpGenei}})}$$\end{document}Riskscore=∑1i(Coefi×ExpGenei) Finally, the median risk score was used to divide patients into the low and high risk score groups. The OS and PFS differences in low- and high-risk score groups were analyzed. The predictive accuracy of the prognostic risk score mode was estimated using the “survivalROC” package in R. Finally, the same procedures were performed in the test set. ## PCA comparison before and after construction of the prognostic risk score model The PCA difference in the two risk score groups was identified using the “limma” R package. PCA was performed and displayed using the ggplot2 package before and after construction of the prognostic model. ## Nomogram construction for predicting OS The “rms” R package was used to construct the OS predictive nomogram in CC patients, including age, sex, pathological stage, and the prognostic risk score model. Time-dependent calibration curves were drawn and AUCs were calculated to predict the accuracy of the nomogram. Univariate and multivariate Cox regression analyses were analyzed. ## Relationship between risk scores and clinical characteristics The CMScaller package in R was used to divide all the samples into consensus molecular subtypes according to their features in the training set. The Limma package in R was further utilized to screen the relationship between risk scores and sex, age, pathological stage, and AJCC TNM stage ($p \leq 0.05$). ## GSVA The “GSVA” R package was used to perform GSVA to compare differences in biological processes between gene profiles in the low and high risk score groups (FDR < 0.05). GSVA is a non-parametric and unsupervised method to distinguish pathway variations or biological processes through an expression matrix sample [39]. The reference gene set came from the “c2.cp.kegg.v7.4. symbols” gene set in MSigDB (https://www.gsea-msigdb.org/gsea/msigdb). ## Characteristic differences between the low and high risk score groups The “pRRophetic” package in R was used to evaluate the chemosensitivity dasatinib, obatoclax mesylate, pazopanib, shikonin, and talazoparib in the low and high risk groups in the training set ($p \leq 0.001$). The IC50 indicates the ability of a substance to inhibit certain biological or biochemical functions [8]. The immune-related infiltration of each training set sample was performed through ssGSEA. A previously verified gene set was utilized to evaluate the immune-related characteristics in the tumor microenvironment, including different immune-related activities and human immune cell subtypes, such as B cells, CD4 + T cells, and macrophages [8, 40, 41]. The ssGSEA algorithm was used to calculate enrichment scores and analyze the relationship between the risk score and immune-related characteristics. ## GO and KEGG analyses and PPI network construction The “limma” R package was used to compare the RNA-seq data profiles of the low and high risk score groups. DEGs in the low and high risk score groups were identified (adjusted $p \leq 0.05$), and the “clusterProfiler” R package was used to conduct GO and KEGG enrichment analyses of DEGs ($p \leq 0.05$). The STRING online database (version 11.5; https://string-db.org/) was used to analyze the DEGs to create a PPI network with median confidence (interaction score > 0.40). Cytoscape software (version 3.9.1) was used to plot the PPI network. Hub genes from all DEGs were searched using cytoHubba (Cytoscape plug-in, version 0.1) and topological algorithms. ## Expression of the seven prognostic model genes and lncRNAs The “limma” R package was used to analyze differential expression of the seven prognostic model genes in the TCGA and GSE44076 datasets. LncRNAs were also analyzed using R in normal colon and CC tissue samples. LncRNAs with \|logFC\|> 1 and FDR < 0.05 were considered statistically significant. ## Identification of lncRNAs associated with the seven prognostic model genes Kaplan–Meier analysis was used to batch filter the prognostic lncRNAs. R was used to further evaluate the association between the expression of lncRNAs and the seven prognostic model mRNAs in CC samples. The association was determined using Pearson’s correlation coefficient analysis (lncRNA mean > 1, $p \leq 0.001$, \|correlation coefficient\|> 0.2). Finally, the co-expression network data were analyzed and plotted using Cytoscape software. ## qPCR LGALS4, RORC, DDIT3, NSUN5, RBCK1, RGL2, and SERPINE1 and their coexpression with lncNRAs C6orf223 and SLC12A9-AS1 were verified in Twenty-seven pairs of matched adjacent normal and CC tissue samples. Informed consent was obtained from all the participants. Cell lines FHC, COLO320DM, SW480 were used for verifing LGALS4 and NSUN5 expressions. All aspects of this study were approved by the Ethics Committee of the First Affiliated Hospital of Guangxi Medical University (Aproval Number: 2022-E415-01, detailed experimental methods and qPCR primers are in Additional file 6). ## Statistical analysis Differences between the two groups were compared using the Wilcoxon test. One-way analysis of variance, Welch one-way ANOVA or K-W test was used to compare three or more groups. All statistical analyses were performed using R, and $p \leq 0.05$ was considered statistically significant. ## Supplementary Information Additional file 1: Figure S1. A ROC curves of the predictive efficiency in the test set. B and C Forest plots of univariate and multivariate Cox regression analyses in the test set. Fig. S2. PPI network of DEGs in the high and low risk score groups. Fig. S3. Expression of REG4， S100A7， CLCA1, FABP4, KRT14, RETNLB, SFRP2, and WNT10A in TCGA CC and normal colon samples. Fig. S4. A–G OS of REG4, S100A7, CLCA1, FABP4, RETNLB, SFRP2, and WNT10A in TCGA.Additional file 2: Table S1. 276 differentially expressed genes between normal and colon cancer samples in TCGA.Additional file 3: Table S2. Differentially expressed genes in the low and high risk score groups. Additional file 4: Table S3. Differentially expressed lncRNAs of TCGA CC samples. Additional file 5: Table S4. Total 611 bacterial response-related genes from 3 GSEA gene sets. Additional file 6. qPCR methods and the primers (Sangon) used for 9 genes qPCR.Additional file 7. Data or codes for figure 4–7. ## References 1. Jia SN, Han YB, Yang R, Yang ZC. Chemokines in colon cancer progression. Semin Cancer Biol (2022.0) 86 400-407. DOI: 10.1016/j.semcancer.2022.02.007 2. Drewes JL, Chen J, Markham NO, Knippel RJ, Domingue JC, Tam AJ, Chan JL, Kim L, McMann M, Stevens C. Human colon cancer-derived clostridioides difficile strains drive colonic tumorigenesis in mice. Cancer Discov (2022.0) 12 1873-1885. DOI: 10.1158/2159-8290.CD-21-1273 3. Bell HN, Rebernick RJ, Goyert J, Singhal R, Kuljanin M, Kerk SA, Huang W, Das NK, Andren A, Solanki S. Reuterin in the healthy gut microbiome suppresses colorectal cancer growth through altering redox balance. Cancer Cell (2022.0) 40 185-200. DOI: 10.1016/j.ccell.2021.12.001 4. Overacre-Delgoffe AE, Bumgarner HJ, Cillo AR, Burr AHP, Tometich JT, Bhattacharjee A, Bruno TC, Vignali DAA, Hand TW. Microbiota-specific T follicular helper cells drive tertiary lymphoid structures and anti-tumor immunity against colorectal cancer. Immunity (2021.0) 54 2812-2824. DOI: 10.1016/j.immuni.2021.11.003 5. Lindig S, Quickert S, Vodovotz Y, Wanner GA, Bauer M. Age-independent co-expression of antimicrobial gene clusters in the blood of septic patients. Int J Antimicrob Agents (2013.0) 42 S2-7. DOI: 10.1016/j.ijantimicag.2013.04.012 6. Meijerink M, van den Broek T, Dulos R, Neergaard Jacobsen L, Staudt Kvistgaard A, Garthoff J, Knippels L, Knipping K, Houben G, Verschuren L. The impact of immune interventions: a systems biology strategy for predicting adverse and beneficial immune effects. Front Immunol (2019.0) 10 231. DOI: 10.3389/fimmu.2019.00231 7. 7.Bao R, Hernandez K, Huang L, Luke JJ: ACE2 and TMPRSS2 expression by clinical, HLA, immune, and microbial correlates across 34 human cancers and matched normal tissues: implications for SARS-CoV-2 COVID-19. J Immunother Cancer 2020;8(2). 8. Ding C, Shan Z, Li M, Chen H, Li X, Jin Z. Characterization of the fatty acid metabolism in colorectal cancer to guide clinical therapy. Mol Ther Oncolytics (2021.0) 20 532-544. DOI: 10.1016/j.omto.2021.02.010 9. Liu Z, Liu Z, Zhou X, Lu Y, Yao Y, Wang W, Lu S, Wang B, Li F, Fu W. A glycolysis-related two-gene risk model that can effectively predict the prognosis of patients with rectal cancer. Hum Genom (2022.0) 16 5. DOI: 10.1186/s40246-022-00377-0 10. Ren Y, He S, Feng S, Yang W. A prognostic model for colon adenocarcinoma patients based on ten amino acid metabolism related genes. Front Public Health (2022.0) 10 916364. DOI: 10.3389/fpubh.2022.916364 11. Zhu J, Kong W, Xie Z. Expression and prognostic characteristics of ferroptosis-related genes in colon cancer. Int J Mol Sci (2021.0) 22 5652. DOI: 10.3390/ijms22115652 12. Zhou R, Gao Z, Ju Y. Novel six-gene prognostic signature based on colon adenocarcinoma immune-related genes. BMC Bioinform (2022.0) 23 418. DOI: 10.1186/s12859-022-04909-2 13. Liu Y, Fu K, Wier EM, Lei Y, Hodgson A, Xu D, Xia X, Zheng D, Ding H, Sears CL. Bacterial genotoxin accelerates transient infection-driven murine colon tumorigenesis. Cancer Discov (2022.0) 12 236-249. DOI: 10.1158/2159-8290.CD-21-0912 14. Nejman D, Livyatan I, Fuks G, Gavert N, Zwang Y, Geller LT, Rotter-Maskowitz A, Weiser R, Mallel G, Gigi E. The human tumor microbiome is composed of tumor type-specific intracellular bacteria. Science (2020.0) 368 973-980. DOI: 10.1126/science.aay9189 15. Acharjee A, Agarwal P, Nash K, Bano S, Rahman T, Gkoutos GV. Immune infiltration and prognostic and diagnostic use of LGALS4 in colon adenocarcinoma and bladder urothelial carcinoma. Am J Transl Res (2021.0) 13 11353-11363. PMID: 34786063 16. Jiang C, Liu Y, Wen S, Xu C, Gu L. In silico development and clinical validation of novel 8 gene signature based on lipid metabolism related genes in colon adenocarcinoma. Pharmacol Res (2021.0) 169 105644. DOI: 10.1016/j.phrs.2021.105644 17. Michalak M, Golde V, Helm D, Kaltner H, Gebert J, Kopitz J. Combining recombinase-mediated cassette exchange strategy with quantitative proteomic and phosphoproteomic analyses to inspect intracellular functions of the tumor suppressor galectin-4 in colorectal cancer cells. Int J Mol Sci (2022.0) 23 6414. DOI: 10.3390/ijms23126414 18. 18.Pandya VB, Kumar S, Sachchidanand, Sharma R, Desai RC. Combating autoimmune diseases with retinoic acid receptor-related orphan receptor-gamma (RORgamma or RORc) inhibitors: hits and misses. J Med Chem 2018;61(24):10976–10995 19. Alexander M, Ang QY, Nayak RR, Bustion AE, Sandy M, Zhang B, Upadhyay V, Pollard KS, Lynch SV, Turnbaugh PJ. Human gut bacterial metabolism drives Th17 activation and colitis. Cell Host Microbe (2022.0) 30 17-30. DOI: 10.1016/j.chom.2021.11.001 20. Lopez-Posadas R, Fastancz P, Martinez-Sanchez LDC, Panteleev-Ivlev J, Thonn V, Kisseleva T, Becker LS, Schulz-Kuhnt A, Zundler S, Wirtz S. Inhibiting PGGT1B disrupts function of RHOA, resulting in T-cell expression of integrin alpha4beta7 and development of colitis in mice. Gastroenterology (2019.0) 157 1293-1309. DOI: 10.1053/j.gastro.2019.07.007 21. Li M, Thorne RF, Shi R, Zhang XD, Li J, Li J, Zhang Q, Wu M, Liu L. DDIT3 directs a dual mechanism to balance glycolysis and oxidative phosphorylation during glutamine deprivation. Adv Sci (2021.0) 8 e2003732. DOI: 10.1002/advs.202003732 22. Jauhiainen A, Thomsen C, Strombom L, Grundevik P, Andersson C, Danielsson A, Andersson MK, Nerman O, Rorkvist L, Stahlberg A. Distinct cytoplasmic and nuclear functions of the stress induced protein DDIT3/CHOP/GADD153. PLoS ONE (2012.0) 7 e33208. DOI: 10.1371/journal.pone.0033208 23. Fessler E, Eckl EM, Schmitt S, Mancilla IA, Meyer-Bender MF, Hanf M, Philippou-Massier J, Krebs S, Zischka H, Jae LT. A pathway coordinated by DELE1 relays mitochondrial stress to the cytosol. Nature (2020.0) 579 433-437. DOI: 10.1038/s41586-020-2076-4 24. Heissenberger C, Liendl L, Nagelreiter F, Gonskikh Y, Yang G, Stelzer EM, Krammer TL, Micutkova L, Vogt S, Kreil DP. Loss of the ribosomal RNA methyltransferase NSUN5 impairs global protein synthesis and normal growth. Nucleic Acids Res (2019.0) 47 11807-11825. DOI: 10.1093/nar/gkz1043 25. Jiang Z, Li S, Han MJ, Hu GM, Cheng P. High expression of NSUN5 promotes cell proliferation via cell cycle regulation in colorectal cancer. Am J Transl Res (2020.0) 12 3858-3870. PMID: 32774740 26. Liu ML, Zang F, Zhang SJ. RBCK1 contributes to chemoresistance and stemness in colorectal cancer (CRC). Biomed Pharmacother (2019.0) 118 109250. DOI: 10.1016/j.biopha.2019.109250 27. Sun MS, Yuan LT, Kuei CH, Lin HY, Chen YL, Chiu HW, Lin YF. RGL2 drives the metastatic progression of colorectal cancer via preventing the protein degradation of beta-catenin and KRAS. Cancers (2021.0) 13 1763. DOI: 10.3390/cancers13081763 28. Kim WT, Mun JY, Baek SW, Kim MH, Yang GE, Jeong MS, Choi SY, Han JY, Kim MH, Leem SH. Secretory SERPINE1 expression is increased by antiplatelet therapy, inducing MMP1 expression and increasing colon cancer metastasis. Int J Mol Sci (2022.0) 23 9596. DOI: 10.3390/ijms23179596 29. Xu Y, Zhao J, Dai X, Xie Y, Dong M. High expression of CDH3 predicts a good prognosis for colon adenocarcinoma patients. Exp Ther Med (2019.0) 18 841-847. PMID: 31281458 30. Liang Y, Su Q, Wu X. Identification and validation of a novel six-gene prognostic signature of stem cell characteristic in colon cancer. Front Oncol (2020.0) 10 571655. DOI: 10.3389/fonc.2020.571655 31. Huang G, Xu X, Ju C, Zhong N, He J, Tang XX. Identification and validation of autophagy-related gene expression for predicting prognosis in patients with idiopathic pulmonary fibrosis. Front Immunol (2022.0) 13 997138. DOI: 10.3389/fimmu.2022.997138 32. Liberzon A, Birger C, Thorvaldsdottir H, Ghandi M, Mesirov JP, Tamayo P. The molecular signatures database (MSigDB) hallmark gene set collection. Cell Syst (2015.0) 1 417-425. DOI: 10.1016/j.cels.2015.12.004 33. Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, Paulovich A, Pomeroy SL, Golub TR, Lander ES. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A (2005.0) 102 15545-15550. DOI: 10.1073/pnas.0506580102 34. Mootha VK, Lindgren CM, Eriksson KF, Subramanian A, Sihag S, Lehar J, Puigserver P, Carlsson E, Ridderstrale M, Laurila E. PGC-1alpha-responsive genes involved in oxidative phosphorylation are coordinately downregulated in human diabetes. Nat Genet (2003.0) 34 267-273. DOI: 10.1038/ng1180 35. Kanehisa M, Goto S. KEGG: kyoto encyclopedia of genes and genomes. Nucleic Acids Res (2000.0) 28 27-30. DOI: 10.1093/nar/28.1.27 36. Kanehisa M. Toward understanding the origin and evolution of cellular organisms. Protein Sci (2019.0) 28 1947-1951. DOI: 10.1002/pro.3715 37. Kanehisa M, Furumichi M, Sato Y, Kawashima M, Ishiguro-Watanabe M. KEGG for taxonomy-based analysis of pathways and genomes. Nucleic Acids Res (2023.0) 51 D587-D592. DOI: 10.1093/nar/gkac963 38. Kanehisa M. Inferring Antimicrobial Resistance from Pathogen Genomes in KEGG. Methods Mol Biol (2018.0) 1807 225-239. DOI: 10.1007/978-1-4939-8561-6_17 39. Hanzelmann S, Castelo R, Guinney J. GSVA: gene set variation analysis for microarray and RNA-seq data. BMC Bioinform (2013.0) 14 7. DOI: 10.1186/1471-2105-14-7 40. Charoentong P, Finotello F, Angelova M, Mayer C, Efremova M, Rieder D, Hackl H, Trajanoski Z. Pan-cancer immunogenomic analyses reveal genotype-immunophenotype relationships and predictors of response to checkpoint blockade. Cell Rep (2017.0) 18 248-262. DOI: 10.1016/j.celrep.2016.12.019 41. Ru B, Wong CN, Tong Y, Zhong JY, Zhong SSW, Wu WC, Chu KC, Wong CY, Lau CY, Chen I. TISIDB: an integrated repository portal for tumor-immune system interactions. Bioinformatics (2019.0) 35 4200-4202. DOI: 10.1093/bioinformatics/btz210
--- title: 'Lessons from digital technology-enabled health interventions implemented during the coronavirus pandemic to improve maternal and birth outcomes: a global scoping review' authors: - Imelda K. Moise - Nicole Ivanova - Cyril Wilson - Sigmond Wilson - Hikabasa Halwindi - Vera M. Spika journal: BMC Pregnancy and Childbirth year: 2023 pmcid: PMC10026210 doi: 10.1186/s12884-023-05454-3 license: CC BY 4.0 --- # Lessons from digital technology-enabled health interventions implemented during the coronavirus pandemic to improve maternal and birth outcomes: a global scoping review ## Abstract ### Background Timely access to essential obstetric and gynecologic healthcare is an effective method for improving maternal and neonatal outcomes; however, the COVID-19 pandemic impacted pregnancy care globally. In this global scoping review, we select and investigate peer-reviewed empirical studies related to mHealth and telehealth implemented during the pandemic to support pregnancy care and to improve birth outcomes. ### Methods We searched MEDLINE and PubMed, Scopus, CINAHL and Web of Science for this Review because they include peer-reviewed literature in the disciplines of behavioral sciences, medicine, clinical sciences, health-care systems, and psychology. Because our investigative searches reviewed that there is considerable ‘grey literature’ in this area; we did not restrict our review to any study design, methods, or place of publication. In this Review, peer-reviewed preprints were comparable to published peer-reviewed articles, with relevant articles screened accordingly. ### Results The search identified 1851 peer reviewed articles, and after removal of duplicates, using inclusion and exclusion criteria, only 22 studies were eligible for inclusion in the review published from January 2020 to May 2022. mHealth interventions accounted for $72.7\%$ (16 of 22 studies) and only $27.3\%$ (6 of 22 studies) were telehealth studies. There were only 3 example studies that integrated digital technologies into healthcare systems and only 3 studies that developed and evaluated the feasibility of mobile apps. Experimental studies accounted $68.8\%$ of mHealth studies and only $33.3\%$ studies of telehealth studies. Key functionalities of the pregnancy apps and telehealth platforms focused on mental and physical wellness, health promotion, patient tracking, health education, and parenting support. Implemented interventions ranged from breastfeeding and selfcare to behavioral health. Facilitators of uptake included perceived benefits, user satisfaction and convenience. Mobile apps and short messaging services were the primary technologies employed in the implemented mHealth interventions. ### Conclusion Although our Review emphasizes a lack of studies on mHealth interventions and data from pregnant women during the COVID-19 crisis, the review shows that implementation of digital health interventions during emergencies are inevitable given their potential for supporting pregnancy care. There is also a need for more randomized clinical trials and longitudinal studies to better understand the effectiveness and feasibility of implementing such interventions during disease outbreaks and emergencies. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12884-023-05454-3. ## Contributions to the literature Modalities and facilitators of uptake of mHealth and telehealth interventions implemented during the coronavirus pandemic is investigated as it impacted on obstetric and gynecologic care and outcomes. Key functionalities of the pregnancy apps and telehealth platforms are identified, and the focus of the implemented digital-enabled health interventions was on mental and physical wellness, health promotion, patient tracking, health education, and parenting support. Our findings demonstrate that digital technologies such as mHealth and telehealth have potential for supporting pregnancy care during emergencies and are vital health system strengthening tools globally. ## Background Timely access to key obstetric and gynecologic healthcare has been shown to be an effective method for improving maternal and neonatal outcomes [1–3]; however, over the past 3 years, the COVID-19 pandemic has had severe effects on pregnancy care globally [4–14]. Several reports in the US have linked COVID-19 and resultant “Stay-at-Home” orders to changes in health service seeking behaviors among pregnant women [15, 16], a finding supported by a systematic review ($$n = 56$$ studies) and meta-analysis ($$n = 21$$ studies) conducted by UK researchers of global changes in prenatal care conducted in 2021 [14]. Another study conducted at a tertiary hospital in India noted a decrease in institutional deliveries (by $45\%$), but an increase in high-risk pregnancies including intensive care unit hospitalizations due to changes in pregnant women’s health seeking behaviors [17]. This finding was also reported by a UK study where the number of women seeking prenatal and postnatal appointments decreased [18]. The pandemic has undoubtedly heightened the risk factors generally linked with poor mental health in pregnancy such as anxiety, depression, and posttraumatic symptoms stemming partly from coronavirus infection fears, poor quality prenatal care, and restrictions on societal behaviors [19–21]. A common thread across the studies is in the observed relationship between the augmented need for structural changes to current prenatal care models including timing and frequency of prenatal care to meet the needs of providers, pregnant women, and their babies in different contexts and settings [7, 22–29]. Addressing these needs is key for achieving the United Nations Sustainable Development Goal (SDG) global target of less than 70 maternal deaths per 100,000 live births [30, 31] (from the current level of 152 deaths per 100,000 live births in 2020) [32]. However, while studies have been carried out on the impacts of COVID-19 on pregnancy care, facilitators of uptake of interventions or facilitators by which current prenatal care delivery models have been modified due to the coronavirus impact (e.g., without and /or with a hybrid of in-person and virtual visits) has not been synthesized in the literature. Digital technologies (e.g., mobile phone and tablet apps and telehealth) have become vital health system strengthening tools globally [33, 34] mostly for overcoming healthcare service delivery challenges [35–40], and their use in pregnancy care has been augmented by the coronavirus crisis. Mobile health technology (mHealth) is defined as a component of electronic health used for delivery of health services using information and communication technology [41–44]. Another digital tool is telemedicine, a component of telehealth, and a practice of medicine utilizing technology to deliver care at a distance (telemedicine signifies clinician-patient interactions and consultations that happen remotely via phone, video calls, text messaging, or other formats) [45, 46]. Systematic reviews and studies conducted in both the ‘global North’ [47, 48] and the ‘global South’ [8, 49–56] have associated patients’ timely access to health services and improvements in obstetric and gynecologic outcomes with digital health interventions [43, 55, 57–62] such as test result notification [63], patient management [64], and real-time access to patient information and communication at different points of care [65–67]. Numerous studies have already demonstrated how digital technologies can address healthcare challenges such as those placed by the coronavirus [68, 69] on pregnancy care and health-seeking behaviors of pregnant women during the pandemic [45, 70, 71]. Additionally, studies evaluating provider and patient experiences with digital-enabled consultations and appointments during the coronavirus crisis report high usage, patient and provider satisfaction [45, 72–77], and that both personal and organizational factors motivate implementation [78, 79]. To note, a recent scoping review that used Google searches to evaluate commonly used apps by pregnant women found over 57 unique apps. Of the final evaluated 29 apps, 18 did not have comprehensive information for every stage of pregnancy [80]. This underscores a need to synthesize information regarding the usefulness and benefits of the implemented apps during the coronavirus crisis to support pregnant women, and whether app differences varied by intervention. Furthermore, while some research has been carried out on pregnant women and COVID-19 [14], no single study exists which assessed a combination of digital-enabled technology (mobile phone and telehealth) implemented during the coronavirus crisis to support pregnancy care. Understanding how and why the digital-enabled technology health interventions were implemented and how they impacted pregnancy care (prenatal, pregnancy and postpartum periods) offer a potential opportunity to improve pregnancy care, reduce costs, resources, and time. We undertook a global scoping review to identify peer-reviewed articles that used mobile phone and telehealth health interventions to improve pregnancy care and/or outcomes during the COVID-19 pandemic. To our knowledge, this is the first study to uncover context, digital interventions, uptake facilitators, locations, study designs and outcomes in the implemented interventions aimed at supporting pregnancy care during the COVID-19 pandemic. ## Overview We conducted a scoping review, a method of knowledge synthesis guided by Arksey and O′Malley’s framework for conducting scoping reviews [81]. To ensure reviewer compliance with best practice guidelines for the conduct of scoping reviews, the PRISMA (Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews) checklist and flow diagram [82] was followed closely, with minimal disagreements because we established a priori inclusion and exclusion criteria in consultation with a research librarian. ## Data sources We conducted preliminary searches on COVID-19, pregnant women, postpartum, pregnancy, mobile phone apps, and interventions as well as combining the keywords on Google scholar and Sematic scholar in May 2022. These searches facilitated the delineation of the review scope of this study, research questions, and established eligibility criteria. Subsequently, we selected MEDLINE and PubMed, Scopus, CINAHL and Web of Science for this Review because they include peer-reviewed literature in the disciplines of behavioral sciences, medicine, clinical sciences, health-care systems, and psychology. Appendix A presents the alternative used key search terms used in this study. Because our investigative searches reviewed that there is considerable ‘grey literature’ in this area; study design, and outcome indicators studied varied widely, we did not restrict our review to any study design, methods, or place of publication. Peer-reviewed preprints were comparable to published peer-reviewed articles, with relevant articles screened accordingly. ## Inclusion and exclusion criteria The scoping review included studies that leveraged digital health interventions (e.g., mobile phone, telehealth, or video conferencing) implemented to enhance access to and/or linkage to pregnancy care services such as to support the three stages of pregnancy (perinatal, pregnancy or postpartum) and improve maternal and neonatal outcomes. Two independent reviewers [IKM and NI] searched databases and grey literature for references of identified peer reviewed studies published from January 2019 up to May 2022. We excluded studies published before the COVID-19 pandemic, study protocols [83, 84], commentary [9, 40, 46, 85–87] or viewpoint [26], thesis [88], or if they were qualitative studies focused on perception of users, mHealth conceptual models [26, 89] not published in English language or did not report on COVID-19. We excluded review papers or studies focused on only mobile phone apps and/or pregnant women without a link to COVID-19 [55, 90, 91]. Also excluded were studies that did not focus on an intervention (e.g., mental health, education, test, vaccines, lifestyle, or mental health). ## Screening and data extraction Studies retrieved from each database were imported into Endnote X8 reference management software [92]. These studies were then imported into Covidence, a systematic online review management program [93] which allowed for screening of eligible studies. The screening process was completed in two phases after duplicates were removed. After the two of five reviewers [including IMK, NI] screened all articles independently at each stage, the authors screened all articles based on relevance of information contained in the title and abstract, and then determined their inclusion for full text review. We then compared the individual screening results and resolved discrepancies by consensus via discussion among the five researchers. Only studies that met the inclusion criteria underwent a full text review. Two reviewers abstracted data which was then validated by a third reviewer, with sections assigned based on reviewers’ expertise. ## Quality assessment To evaluate the methodological quality of all primary research studies, we utilized the Joanna Briggs Institute Critical Appraisal Checklist for Prevalence Studies [94]. We used the checklist to assess the extent to which the selected studies report on the likelihood of bias in nine topics of study design, conduct, and analysis as used in previous studies [95], ranging from phase 0 (poor quality) to 2 (high quality), with studies receiving a total quality score (ranged from 0-poor quality to 18- higher quality). We excluded studies with a total score of < 13. Five researchers [IMK, NI, CW, SW, and SV] independently evaluated each included study, and through discussion, we decided on any doubt about the quality of included studies. ## Data synthesis We reviewed attributes of the included studies. Narratively synthesized were key implemented health interventions that supported the three stages of pregnancy. To categorize the selected studies, we adopted a conceptual model developed a priori based on existing literature on mHealth and maternal and child health [43]. Digital interventions were identified based on their purpose, modalities, facilitators for uptake (e.g., choices, perceptions), relevant actors (pregnant women, providers, mothers), study design, context or conditions required for program facilitators to activate or not, and mobile phone or telehealth intervention outcomes [43]. ## Study characteristics Figure 1 presents the PRISMA flow diagram used for the scoping review. The search yielded 1851 peer reviewed articles from select databases: PubMed MEDLINE [1481], CINAHL [220], Web of Science [28], Scopus [12] and grey searches [110]. Of these, we excluded 30 duplicates. Of the remaining 1821 studies, 1713 studies were non-empirical, such as study protocols, thesis, commentary, viewpoints, or reviews. Based on title and abstract review, 108 studies were left for a full-text review, and of these, 85 studies were excluded (with 18 studies excluded after assessment for quality) leaving 22 studies ($$n = 16$$ mHealth studies and 6 telehealth studies).Fig. 1Prisma flowchart of information through the different phases of article extraction from the literature search Table 1 shows the 22 studies that were selected on COVID-19, mHealth and telehealth from the three stages of pregnancy. The included studies were published in English, and between 2020 and 2022. Of the 16 mHealth studies, 7 studies were conducted in the USA ($$n = 4$$) and Iran ($$n = 3$$). The remaining 9 studies were conducted in Australia, Canada, Guatemala, Indonesia, Mozambique, Singapore, Sri Lanka and the UK (one study in each country). One study’s location was unclear. Of the 6 telehealth studies, one was conducted in the USA and the rest in Australia, Germany, Indonesia, Iran and Spain (one study in each country). Figure 2 shows the country locations of the peer-reviewed studies included in this scoping review. Table 1Summary of studies of digital technology-enabled health interventions (mHealth and Telehealth) implemented during the coronavirus crisis to support pregnancy care, globallyAuthorsPublication yearOriginIntervention (s)FacilitatorsStudy designStudy population, age, Sample sizeContext for facilitators OutcomesmHealth for mental and physical wellness (6 studies) Pregnancy Kiani & Pirzadeh [96]2021IranPhysical activity educational intervention delivered via mobile app to increase physical activities in pregnant women during the coronavirus crisis. Perceived app benefits: motivational content and images that emphasize both physical and psychological benefits. Quasi-experimental study93 pregnant women aged 16–20 weeks of gestation participating in the childbirth preparation classes. Training on how to use the application, app did not require an Internet connection. Mobile Apps increased scores of perceived benefits, barriers, social support, enjoyment, and improved levels of physical activity in pregnant women. Smith et al. [ 97]2021USAConsumer-based mobile phone meditation application (app) to help pregnant women self-manage stress and anxiety. Satisfaction with the Calm app, and easy to use. Randomized Controlled TrialPregnant women 18 years and older with confirmed pregnancy between 14 and 34 weeks of gestation attending a university outpatient clinic of obstetrics and gynecology. Ability to regularly access a smart device, self-reported use of app; Training on how to download and use the mobile app. Women who used the prescribed consumer-based mobile meditation app during the coronavirus crisis had significant reductions in perceived stress, depression, anxiety, and sleep disturbance compared with standard care. Kubo et al. [ 98]2021USASelf-paced, guided mindfulness meditations provided through a website or mobile application (iOS and Android) Headspace™ app to improve adherence and efficacy in pregnant women. Convenience of the intervention; ease of use and the short dosage (10–20 min a day).Single-arm trial27 pregnant women age ≥ 18 years with moderate-to-moderately severe depression symptoms and < 28 weeks of gestation identified through the electronic health records, and self- or clinician-referral from obstetrics and gynecology clinical staff or study brochures. Access to a smartphone, tablet, and/or computer with Internet connection. Improvements observed in pre-postintervention scores for depression symptoms, perceived stress, sleep disturbance, and mindfulness; over half of participants used the app ≥$50\%$ of the days during the 6-week intervention. Postpartum period Avalos et al. [ 49]2020USASelf-paced, guided mindfulness meditation training delivered via mobile app Headspace™ app (iOS and Android) or website to deepen mindfulness and encourage routine use in postpartum women. Convenience of the intervention delivered via commercially available mindfulness app (headspace).Mixed methods single-arm feasibility trial13 women aged 18 years; within 6 months of giving birth, with moderate to moderately severe depressive symptom was recruited via electronic health records, self- or clinician referral in obstetrics and gynecology clinics. Seeking standard postpartum care and access to a smartphone, tablet, or computer with internet access. Preliminary efficacy and improvement in depression symptoms, perceived stress, sleep quality, and mindfulness in postpartum women with moderate to moderately severe symptoms of PPD. Dol et al. [ 99]2021CanadaSix-week postpartum mobile phone text message program “Essential Coaching for Every Mother” to improve women’s psychosocial outcomes in the immediate postpartum period. Acceptability of the program in terms of timing and content; perceived support from the providers. Prospective pre-post study88 first time mothers aged 18 years and older between 37 weeks’ gestation or those who gave birth over a three-month period at the study clinic. Daily access to a mobile phone with text message capabilities. Essential Coaching for Every Mother mHealth intervention improved maternal self-efficacy and decreasing anxiety: baseline compared to follow-up Jayasinghe et al. [ 69]2022Sri LankaTelephone interviews providing psychological support to pregnant and postpartum women during the pandemic. Better access to health care; selection biasLarge-scale, population-based pregnancy cohort1438 ($42.6\%$) pregnant women. Availability of mobile phone and fixed access to phones; availability of trained interviewees.476 ($33.1\%$) of interviewed women used messaging apps to receive health messages such as WhatsApp, Viber, IMO, and Facebook Messenger.mHealth for health promotion, tracking and education (5 studies) Pre-pregnancy Leight et al. [ 100]2022MozambiqueShort text message reminders designed to encourage uptake of health facility visits for family planning counselling. Reminder support; timely clinic visits. Two arm randomized controlled trial5623 women of reproductive age who received a referral from a community health worker to visit a health clinic for a family planning consultation. Availability of a phone number on patient record; PSI-provided smartphone to access the platform. The effect of the text reminders were positive and statistically significant. Women who received text reminders more likely to visit a clinic, report receiving a contraceptive method at a clinic and prompt visit to a clinic (conditional on ever reporting a visit). During Pregnancy Krishnamurti et al. [ 101]2021USAPrenatal care app based intimate partner violence (IVP) self-screening tool (MyHealthyPregnancy app to encourage continuation of screening and receiving of support services. Willingness to disclose IPV experiences through an app; social isolation; perceived support. Quality Improvement Pilot Study552 pregnant women who used the app during an in-person visit. Presence of clinical education team; availability of Apple v.1.4.7 and Android v.1.8.App-based screening captured a distinct set of at risk IPV patients, complementing in-person assessments. Incidence of IPV slightly increased during the shelter-in-place order. Varnfield et al. [ 102]2021AustraliamHealth platform M♡THer for gestational diabetes mellitus management in pregnancy. Acceptability of the platform by women; faster interventions. Evaluation design23 women with a first-time diagnosis of gestational diabetes mellitus; between 24 and 28 weeks of gestation; at least 16 years old were recruited at a study hospital. Ownership and ability to use a smart mobile phoneSatisfaction and ease of use of the mHealth platform, with technological challenges around wireless connectivity. Moulaei et al. [ 103]2021IranMobile-based application developed using Java programming language in an Android Studio programming environment to support self-care and self-management for preeclampsia in pregnancy. Perceived knowledge and better attitude towards COVID-19 and preeclampsia. User centered design10 pregnant women (with/without preeclampsia, and with/without COVID-19 infection) and 10 doctors in affiliate hospitals and medical centers. Availability of mobile devices with Android operating systems. Pregnant women rated the usability of the application at a satisfactory level. Moulaei et al. [ 104]2021IranSmart phone-based self-care application developed to help pregnant women against the coronavirus. Educational needs provided via different methods (e.g., texts, educational videos, audios).Perceived usability; user satisfaction with data elements, educational information needs, functions, and lifestyle information. Descriptive applied study conducted in two phases36 pregnant women (> 8 weeks, gestational age 20–50); 11 had a coronavirus diagnosis and without exceptional care, partial or absolute rest recruited at affiliated hospitals. Daily use of smartphone. With an average score of 7.94 (out of 9), pregnant women rated the application at a satisfactory level.mHealth for parenting support (5 studies) Pregnancy Ceballos et al. [ 105]2020GuatemalaText messages (SMS) reminders and phone calls to encourage individuals to visit the health center to monitor the provision of health and nutrition interventions linked to the first 1000 days of life (exclusive breastfeeding, vitamin A, powdered micronutrients, and vaccines).Availability of health services; absence of social conflicts. Clustered randomized controlled trial (cRCT)1542 households with pregnant women and children under two years old who receive key health and nutrition interventions from local public health centers. Access to mobile phone, presence of at least one child under two years old or one pregnant woman; monthly airtime top-ups. Response rate to phone calls was 5 times higher compared to text messages ($75.8\%$ vs. $14.4\%$). The cost for mobile phone call reminders were cheaper than that of SMS. Rhodes et al. [ 106]2020UKBaby Buddy, a pregnancy and parenting app provides trusted, evidence-based information and self-care tools to help expectant and new parents through pregnancy and the first 24 weeks of parenthood (e.g., being pregnant or parenting a young baby, mood, levels of anxiety, key concerns).Low literacy level requirements and extensive video content; accessibility to people not in education, training, or employment and those who do not speak English. Service evaluation study436 expectant ($$n = 244$$, $56.0\%$) and recent ($$n = 192$$, $44.0\%$) parents age < 21- older than 45 years who were Baby Buddy app users. Smartphone ownership: app is free to download and available in all app stores$.97\%$ ($\frac{423}{436}$) of respondents reported that Baby Buddy was currently helping them. Greater speed in updating digital content to reflect changes due to the pandemic. Wulandari et al. [ 107]2022IndonesiaInteractive mHealth message intervention via flyers (text, images), videos, and assistance (consultation, discussion, sharing, and question and answer) to improve safe and effective postpartum care. Perceived easiness to understanding shared information; increased knowledge; existence of communication, interaction, social networks, and the impact of the use of social media. Quasi-experimental design46 pairs of pregnant women (gestational age 28–34 weeks) and their husbands were selected purposively from data on pregnancy visits at the Community Health Center. Availability of mobile phone with WhatsApp applicationKnowledge of mothers and husbands increased on post-partum care, and so was improvement in the mother’s practices related to postpartum visits. Postpartum period Shorey et al. [ 108]2021SingaporeSupportive Parenting App (SPA, iOS, and Android) development procedure to provide perinatal educational intervention for couples with healthy infants. Adequate time; financial budgeting and team cohesion. Multistage iterative development process, and information systems research framework10 new parents and research team members (app developers, clinicians, and research assistants).Availability of smartphone with internet access. Documented the technical details of the SPA and intervention highlights the key aspects needed for future app development. Quifer-Rada et al. [ 109]2022Not clearAutomated breastfeeding consultation system on LactApp, an mHealth Solution for breastfeeding supportFree mobile app. Self-administered tools; functionalities of breastfeeding monitoring; breastfeeding tests and personalized plans. Observational, descriptive, and retrospective study137,327 active usersAvailability of email of registered users, demographic factors of mother and baby. Active users increased by 12, 092; topics of interest for consultations varied but include growth spurts, breastfeeding stages, breastfeeding technique, breast pain and mastitis, problems with infants not gaining weight. Telehealth for mental and physical wellness (2 studies) Pregnancy Hashemzahi et al. [ 110]2022IranSelf-care training via telemedicine to help mothers familiarize with, manage, and follow up risk symptoms, and to reduce their own stress and anxiety. Audio PowerPoints turned into video content, sent through WhatsApp messenger. Quasi-experimental study100 pregnant women aged 18–49 years, with gestational age of 20–28 weeks and referred to comprehensive health centers for pregnancy-related complications and COVID-19 infection. Having a mobile phone or PC and the ability to use them for Internet accessFindings show that telemedicine COVID-19 self-care training significantly reduced perceived stress, and anxiety in pregnant women including rising awareness about coronavirus and reducing false beliefs. Silva-Jose et al. [ 111]2022SpainVirtual supervised exercise program to increase maternal physical activity and improve health outcomes, with classes delivered in an online format using the Zoom platform. Greater availability of time; home confinement; perceived sense of sense of social supportEvaluation design using semi-structured interviews24 women between 8 and 10 and 38–39 weeks of pregnancy and attending online fitness classes during the confinement period. Access to the Zoom platform from home. Pregnant women were receptive to online group exercise classes and liked the accessible option to accommodating physical activity during the pandemic. Telehealth for health promotion, tracking and education (4 studies) Pregnancy Oelmeier et al. [ 124]2022GermanyPrenatal counseling via Telemedicine: video consultations in a tertiary center for obstetric care that was a part of the larger open Video Service project on telemedicine. Perceived satisfaction and feasibility. Prospective single-center trial75 video consultations were carried out with patients requiring prenatal or pre-pregnancy counseling. Being a part of a larger open Video Service project on telemedicine. Patient satisfaction was high ($95\%$, $\frac{71}{75}$) but technical problems occurred in $37\%$ ($\frac{29}{75}$) of the appointments. Nur et al. [ 112]2020IndonesiaAndroid-based electronic technology antenatal care (e-ANC) to enhance participation of midwives and pregnant women in antenatal care (e.g., counseling, high-risk early detection on pregnancy, and monitoring of Hb and Fe tablets).Perceived privacy and confidentiality. Quasi-experimental study using pre- and post-test experiments30 pregnant women (in 2nd trimester) ages < 20- > 35, and 20 midwives at areas around the Public Health Centers. Capacity to use Android devices with the e-ANC feature, speak Indonesiane-ANC increased prenatal care visits particularly counseling, high-risk early detection, monitoring Hb, and provision of Tablet Fe. Postpartum period Palmer et al. [ 113]2021AustraliaTelehealth integration into routine antenatal care and delivered via telephone or video conferencing compared to conventionally delivered care on pregnancy outcomes. Perceived easy of self-monitoring, perceived technology, and communication of appointments support. An interrupted time-series analysis2292 women who gave birth between April 20 and July 26, 2020, across a large health service, with large numbers of births assessed in both periods. Availability of phone or internet with video; implementation of integrated antenatal care. Telehealth successfully integrated into antenatal care, and it enabled the reduction of in-person consultations by $50\%$ without compromising pregnancy outcomes. Reisinger-Kindle et al. [ 114]2021USAMaintenance of telehealth as an option for prenatal and postpartum visits after state-mandated restrictions eased. Perceive significance in choice of appointment, perceived support from providers regarding use of phone systems, and mandatory virtual meetings. Retrospective chart review of all pre- and post-natal care visits558 prenatal patients and 209 postpartum patients receiving prenatal or postpartum care at a large urban academic obstetrics and gynecology practice. Availability of video equipment for those with video telehealth capabilities, and/or availability of audio-only telephone; primary language Spanish; Availability of trained providers. The Reach of the intervention increased from baseline. Adoption was high, with all thirty providers using telehealth, and the telehealth found to be feasible and acceptable based on uptake. Effectiveness measures suggest potential for earlier diagnosis of prenatal conditions. Fig. 2Country loctaion of studies included in the scoping review. Map was genereated using ArcGIS software v. 10.5 (https://www.esri.com/en-us/home) The 16 mHealth included studies were classified into four types of study designs: experimental studies (11 studies, including 5 randomized control trials (RCTs), and 2 quasi-experimental studies), one was a mixed methods study, and two were observational studies. Two of 16 mHealth studies used a multistage iterative development process and a user centered design. We then grouped the identified 6 telehealth interventions into three types of study designs: experinmental studies (2 studies), evalaution studies (3 studies), and one study was a retrospective chart review of all pre- and post-natal care visits. All included studies reported the use of mHealth or telehealth to support the three stages of pregancy during the COVID-19 pandemic. Although included studies (telehealth and mHealth) reported varying degrees of eligiblity criteria for selecting study participants, women had to be pregnant or should have given birth within a stipulated study period. Study particpants in most studies were selected based on the pregnant woman’s age (e.g., at least 18 years), their gestational age in weeks or months (e.g., 16–20 weeks of gestation), or if the pregnant woman was a parent with a child born within 6 months postpartum. Two articles included women of reproductive age. A few studies selected pregnant women based on their area of residence and/or service clinic or if women were active users of the study digital technolgy. ## Overview of interventions examined To classify the chosen studies, we applied the program theory of mHealth programs and maternal and child health [43] and developed a modified model of the major outcomes (Fig. 2). The included 22 studies reported on three outcomes: 1) mental and physical wellness, 2) health promotion, patient tracking and education, and 3) parenting support (Fig. 3). The mHealth interventions were delivered using various technology platforms, including mobile app (iOS and Android) or smartphone (15 studies) with some sending intervention content via WhatsApp Messenger and/or email (1 study). A few studies used a combination of mobile phone or smart phone and/or websites. Fig. 3Digital interventions (mHealth and telehealth) implemented at different stages of pregancy in the inlcuded studies ## Mental and physical wellness A recent review of the psychological impact of COVID-19 pandemic on the mental health of pregnant women, conducted by researchers at the University of Cagliari in Italy in 2021, linked an increase in mental health of pregnant women to the coronavirus pandemic [115] and underscored the need for psychological support during pregnancy to mitigate mental health and the risk of long-term impacts on child development. Three of six mHealth mental and physical wellness studies supported pregnant women (3 studies) while three studies supported postpartum care. Mental and physical wellness interventions included: 1) self-paced guided mediation, 2) meditation training, 3) psychological support, 4) psychosocial coaching, and 5) physical activity. ## mHealth mental and physical wellness interventions for pregnancy support To determine the effect of a mobile application (app)-based health interventions, with motivational multimedia with photos and videos on physical activity, Kiani and Pirzadeh [96] conducted a quasi-experimental study of 93 pregnant women (16–20 weeks of gestation) participating in childbirth preparation classes during the coronavirus crisis. Researchers encouraged women to use the mobile app designed not to require internet connection for a specified period. Findings (pre- and 3 months-post intervention) showed that the perceived benefits and enjoyment of physical activity increased post intervention in the intervention group (compared to the control group), so did the mean score of physical activity in this group. Smith and colleagues evaluated the effect of a consumer-based mobile app-based meditation on wellness by randomizing 101 women (50 women in the treatment group, and 51 women in the control group and standard care). Pregnant women in the treatment group used the mindfulness meditation app Calm for 30 days. Results showed a significant reduction in the intervention group (compared to the control group) in perceived stress, depression, anxiety, and sleep disturbance [97]. ## mHealth mental and physical wellness interventions for postpartum support To evaluate the effect of a six-week postpartum mobile phone text message program (“Essential Coaching for Every Mother”) on maternal self-efficacy, social support, postpartum anxiety, and depression, Dol and colleagues [99] conducted a prospective study with 88 first-time mothers enrolled after giving birth and 6 weeks postpartum between July 15 and September 19, 2020. The study noted an increase in self-efficacy (at follow-up compared to baseline), and a reduction in anxiety, and women’s satisfaction with the program [99]. However, further work is required to establish the viability of this program. Using a large-scale, population-based pregnancy cohort, Jayasinghe and colleagues [68] recruited 3374 first-trimester pregnant women registered with midwives at the field prenatal clinics to assess the feasibility and the coverage and feasibility of app-based interventions and generalizability of telephone interviews for psychological support during the pandemic. The study revealed that mHealth led to selection bias and that mHealth may not be the best strategy for interventions in this remote area. Two studies in the US tested the feasibility and acceptability of offering a self-paced, commercially available mobile-app (Headspace™)-based mindfulness intervention in women with depressive symptoms by using single arm trials [48, 98]. Notably, although one of these studies focused on pregnant women (< 28 weeks of gestation who were not practicing a regular, < 3 times per week) mindfulness/meditation [98]; the other study focused on postpartum women [48]. Both studies noted significant improvements in pre-post-intervention scores for depression symptoms, perceived stress, sleep disturbance, mindfulness, feasibility, and acceptability of the mHealth mindfulness intervention for pregnant and postpartum women. In both studies, women had to follow a self-paced, 6-week mindfulness meditation program using the app 10–20 minutes each day over the 6-week period [98]. ## Health promotion, tracking and patient education Health education during pregnancy plays a critical role in improving maternal and neonatal outcomes (e.g., birth weight, initiation and continuation of breastfeeding, and postpartum strategies) [116–119]. Five of the 16 mHealth studies focused on health promotion, tracking and patient education to support the three stages of pregnancy [100]. These interventions focused on educating women regarding the coronavirus, management of gestational diabetes, preeclampsia, perinatal education, visits, and maintenance including uptake. ## mHealth for health promotion, tracking for pre- pregnancy support To estimate the effects of mobile text messages in encouraging the use of family planning services in Mozambique, Leight and colleagues [100] conducted a randomized controlled trial with 5623 women receiving services from the Integrated Family Planning Program implemented by Population Services International between 20 January and 18 December 2020. Women in the treatment group received a series of text message reminders encouraging uptake of health facility visits for family planning counselling. They observed that women in the treatment group were more likely to receive a contraceptive method at a clinic. ## mHealth for health promotion, tracking and pregnancy support A quality improvement study that screened 552 patients for intimate partner violence (IPV) during the COVID-19 pandemic using a prenatal care app which showed that the use of the IPV screening tool increased during the lock-down period [101]. Another evaluation study focused on the adoption and multidisciplinary care coordination of an mHealth platform in a cohort of women with first-time diagnosis of GDM showed that high blood glucose reviews and antenatal contact among app users (compared to non-app users) [102]. Two studies conducted in Iran designed and developed two mobile apps to facilitate self-care for pregnant women with preeclampsia during COVID-19, and examined the effect of a self-care smartphone-based application on self-care practices for pregnant women against COVID-19. Both studies noted positive reviews, and the apps were rated highly by users [103, 104]. ## Parenting support Parenting can be tough, and parenting apps can lessen the burden for first-time parents through providing access to information and tracking child development [120] especially during the COVID-19 pandemic, when pregnant parents and parents of young children lost usual support networks. Five studies focused on parenting support during pregnancy (3 studies) and the postpartum period (2 studies), and four interventions enabled by mHealth technology: 1) breastfeeding, 2) nutrition support, 3) self-care tools, 4) postpartum care, perinatal education and/or uptake. ## mHealth for parenting support Ceballos and colleagues [105] applied a clustered randomized controlled trial (cRCT) to evaluate the feasibility of using simple calendarized mobile text messages (SMS) and phone calls to improve the timely reception of health services of key health and nutrition interventions linked to a critical period of growth and development (the first 1000 days of life), but did not find any effects. However, mobile phone calls were an effective low-cost tool. Rhodes and colleagues [106] conducted a service evaluation using mixed methods to determine whether the “Baby Buddy”, a pregnancy and parenting app aimed at supporting pregnant women and new parents through the first 24 weeks of parenthood, was meeting users’ needs (436 women). Women reported increased anxiety due to decreased health care delivery and loss of social support from friends and family. In Indonesia, a quasi-experimental study assessed the effectiveness of an interactive mHealth message intervention (text, images), videos, and assistance (consultation, discussion, sharing, and question and answer) using WhatsApp groups to improve postpartum care behavior of 88 mothers and husbands (43 pairs- treatment group and 45 pairs - control group) [107]. The duration of the intervention was 14 days (5 hours per day, followed by random flyers delivered at birth up to 42 days post birth). The researchers report improved postpartum care behavior for mothers and their husbands. ## mHealth for parenting support during the postpartum period One study developed a Supportive Parenting App (SPA) to improve parent and infant outcomes in the perinatal period, and highlight the challenges and lessons learned, which centered around user and technological problems [108]. Likewise, a study in an unspecified location evaluated whether the coronavirus crisis affected breastfeeding consultations by using data from LactApp (a mobile application [app], an mHealth solution designed for breastfeeding support, and revealed an increase in breastfeeding support [109]. ## Telehealth for mental and physical wellness Previous studies conducted prior to the pandemic as well as during the pandemic has demonstrated the significance of telemedicine prenatal care [121, 122]. However, others have raised concerns regarding equity in access to care, particularly in resource-limited settings [123]. The 6-telehealth technology-enable interventions identified in this scoping review focused on exercise, selfcare, prenatal counseling and behavioral health. ## Telehealth for mental and physical wellness support during pregnancy To explore the experiences of pregnant women (8–39 weeks of pregnancy) participating in an online group exercise program, and the role of a virtual group fitness on maternal mental outcomes, Silva-Jose and colleagues [111] applied a phenomenological approach in study conducted between March to October 2020. The researchers found that pregnant women were receptive to the online group exercise classes and liked the accessibility of the physical activity, and the social connection they provided. Three of the telehealth studies focused on antenatal care. One study quasi-experimental study with 100 pregnant women aged 18–49 years and 20–28 weeks gestation age determined the effect of COVID-19 self-care training via telemedicine on perceived stress and corona disease anxiety, and noted a reduction in perceived stress in the treatment group (compared to the control group) [110]. Findings confirm the effectiveness of the selfcare training implemented via telemedicine and during the coronavirus crisis in reducing the perceived stress and anxiety of pregnant women. ## Telehealth for health promotion, tracking and education support during pregnancy Regarding the telehealth technology-enable interventions, Oelmeier and colleagues [124] conducted a prospective single-center trial by use of 75 video consultations in a tertiary center for obstetric care, and the results indicated that patient satisfaction was high, but technical problems were experienced in $37\%$ of the appointments. Nur and colleagues [112] used a purposive sampling technique and post-test experiments on 30 pregnant women and 20 midwives to examine the effect of COVID-19 on antenatal visits on pregnant women. They found differences in the prenatal visits among pregnant women pre-and post the COVID-19 lockdown period, and in midwives participation rates in counseling, high-risk early detection on pregnancy, Hb monitoring, and provision of Fe tablets. ## Telehealth for health promotion, tracking and education support during the postpartum period In Australia, one telehealth technology study used an interrupted time-series design to assess the impact of telehealth integration into antenatal care by comparing the first 3 months of telehealth integrated care delivered between April 20 and July 26, 2020, with conventional care across low-risk and high-risk care models. We found no significant differences in the integrated care period as it relates to the number of babies with fetal growth restriction (e.g., birthweight) or pregnancies complicated by pre-eclampsia, or gestational diabetes. Overall, they noted a reduction in-person interactions during the pandemic, and recommend the use of telehealth in post-pandemic health-care models [113]. Further, a retrospective chart review of electronic health records of all pre-and postnatal care visit encounters (558 patients, and 1788 prenatal visits) from March 19 and August 31, 2020 noted limited effectiveness measures but potential for earlier diagnosis of some prenatal conditions, and that telehealth was a feasible option [114]. ## Discussion We performed a global scoping review to synthesize the published peer-reviewed literature on the mHealth and telehealth studies conducted during the coronavirus crisis to support pregnancy care. These studies showed how mHealth and telehealth studies varied regarding the types of designs, context, interventions and/or modalities, uptake facilitators and outcomes studied. mHealth studies constituted $72.7\%$ of the published studies while $27.3\%$ were telehealth studies. There were only 3 example studies that integrated digital technologies into healthcare systems [28, 98, 113] and only 3 studies that designed, developed and evaluated the feasibility of mobile apps to support one or a combination of the 3 stages of pregnancy [103, 104, 108]. Mobile apps and short-messaging services were the primary technologies employed in mHealth studies. Experimental studies, such as (RCTs) quasi-experimental studies accounted for $68.8\%$ of mHealth studies, and only $33.3\%$ were telehealth studies using experimental study designs. Future studies that determine whether a cause-effect relation exists between treatment and outcome during pandemics or emergencies are therefore recommended. We identified three key functionalities of pregnancy apps and telehealth platforms including mental and physical wellness, health promotion, patient tracking, health education, and parenting support. However, the results of the synthesized studies demonstrate the similarities and differences between participant characteristics and intervention modalities used from prenatal through delivery and postpartum. For instance, there were considerable differences in study participants ages, with most studies using both the birth age of the mother and gestation age, while a few studies did not include age but referenced the age of participants as women of reproductive age [100] or pregnant women [68, 101, 103, 109, 124, 125], or women with children under 2 years old [105, 113, 126]. Likewise, although two studies conducted in the US used a similar mHealth intervention delivered via Headspace™ app and with patients with moderate-to-moderately-severe depression as well as duration (use of app for 10 to 20 minutes a day for 6 weeks), they focused on different stages of pregnancy, gestation age and patient birth ages. For example, Avalos and colleagues [48] applied the mHealth mindfulness program to support women aged at least 18 years, and within 6 months of giving birth, while Kubo and colleagues [98] applied it to support pregnant women age 18 years or older with less than 28 weeks of gestation. In this scoping review, $37.5\%$ of included studies utilized mHealth for physical activity education, mental health and wellness, mindfulness meditation, psychological and psychosocial (e.g., maternal self-efficacy, social support, anxiety -postpartum specific and coronavirus related stress). This provides noble interest and motivation of using mHealth technologies and content-driven education to facilitate maternal self-efficacy. In addition, although only few studies were included, the finding highlights the psychological sequelae of COVID-19 on pregnant women, a concern raised in recent studies [19]. The identified mHealth and telehealth platforms from this scoping review provide good evidence as digital technologies used by both women and providers to support maternity continuum of care during the pandemic. Similarly, the design of self-paced guided and instances of supervised self-care, including the use of apps without needing internet connection can enhance uptake, promote healthy pregnancies and replication in other subpopulations and clinical settings. Our findings are in line with pregnancy apps desired functionalities reported in other related reviews [80, 83, 127]. Moreover, interventions were piloted and evaluated with potential users (women and providers) before implementation. Previous studies have indicated the enormous effect that positive parenting practices can have on children’s social, emotional, and intellectual development, especially during the early years, and parenting apps can help first-time parents access information and track child development. Five of the 16 mHealth studies focused on this important topic and revealed that mHealth technologies play a key role in supporting new parents to access information and receive health interventions because of their ease of accessibility. However, only two of the five mHealth parenting studies used experimental study designs, therefore questions remain. In terms of the telehealth studies, pregnancy, and the COVID-19 pandemic, $37.5\%$ included studies supported pregnancy and postpartum care via prenatal counselling, self-care training, supervised exercises, prenatal care provision, and or maintenance and behavioral health. Most studies used a combination of platforms in the delivery of telehealth interventions such as video consultations, audio PowerPoints turned into video content and sent via mobile WhatsApp, Zoom platform, or telephone. What is surprising is the limited evidence supporting telehealth interventions in obstetrics and gynecology, a finding also reported by a systematic review conducted before the COVID-19 pandemic [60]. In summary, although our scoping review has limitations that are embedded from the included studies to our knowledge, the current study is the first to uncover context, digital interventions, uptake facilitators, locations, study designs and outcomes in the interventions implemented during the COVID-19 pandemic to support pregnancy care. We recognized three key functionalities of both pregnancy apps and telehealth that have been studied to date, including clinical outcomes across the maternity care continuum that has been supported using the two digital technologies during the coronavirus crisis to support pregnancy care based on the theory for mHealth programs. ## Conclusion Our scoping review identified key functionalities of mobile apps and telehealth platforms, pregnancy and intervention outcomes, context, interventions, study designs and approaches in the included studies during the COVID-19 crisis. Most studies focused on mental health and physical wellness, and mobile phone apps were the most used modality, followed by telehealth. Few studies included the use of WhatApp, messenger and websites. Regarding methodological approaches, slightly more than half used experimental studies, and the remaining were evaluation studies, developmental or design studies. Included studies published in the past two and a half years underscore an emergent topic of study, and we anticipate more studies in the near future. ## Supplementary Information Additional file 1: Appendix A. Keyword search strategy. ## References 1. Berhan Y, Berhan A. Antenatal care as a means of increasing birth in the health facility and reducing maternal mortality: a systematic review. Ethiop J Health Sci (2014.0) 24 Suppl 93-104. DOI: 10.4314/ejhs.v24i0.9S 2. Moise IK, Kalipeni E, Jusrut P, Iwelunmor JI. Assessing the reduction in infant mortality rates in Malawi over the 1990–2010 decades. Global Public Health (2017.0) 12 757-779. DOI: 10.1080/17441692.2016.1239268 3. Hollowell J, Oakley L, Kurinczuk JJ, Brocklehurst P, Gray R. The effectiveness of antenatal care programmes to reduce infant mortality and preterm birth in socially disadvantaged and vulnerable women in high-income countries: a systematic review. BMC Pregnancy Childbirth (2011.0) 11 13. DOI: 10.1186/1471-2393-11-13 4. 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 5. Mirzakhani K, Shoorab NJ, Akbari A, Khadivzadeh T. High-risk pregnant women’s experiences of the receiving prenatal care in COVID-19 pandemic: a qualitative study. BMC Pregnancy and Childbirth (2022.0) 22 363. DOI: 10.1186/s12884-022-04676-1 6. Uwambaye P, Nyiringango G, Musabwasoni SMG, Husain A, Nessa K, Razzaque MS. COVID-19 pandemic: adaptation in antenatal Care for Better Pregnancy Outcomes. Front Glob Womens Health (2020.0) 1 599327. DOI: 10.3389/fgwh.2020.599327 7. Schmitt N, Mattern E, Cignacco E, Seliger G, König-Bachmann M, Striebich S, Ayerle GM. Effects of the Covid-19 pandemic on maternity staff in 2020 - a scoping review. BMC Health Serv Res (2021.0) 21 1-25. DOI: 10.1186/s12913-021-07377-1 8. Agarwal S, Labrique A. Newborn health on the line: the potential mHealth applications. JAMA (2014.0) 312 229-230. DOI: 10.1001/jama.2014.6371 9. Boum Y. Beyond COVID-19: scaling up and sustaining mobile health in Africa. Lancet (2021.0) 398 1962-1963. DOI: 10.1016/S0140-6736(21)02349-7 10. Al-Muhanna FA, Alkuwaiti AA, Alwazzeh MJ. Coronavirus disease (COVID-19) and healthcare delivery system. J Fam Community Med (2021.0) 28 63-64. DOI: 10.4103/jfcm.JFCM_572_20 11. Setiawan HW, Pratiwi IN, Nimah L, Pawanis Z, Bakhtiar A, Fauziningtyas R, Ramoo V. Challenges for healthcare workers caring for COVID-19 patients in Indonesia: a qualitative study. Inquiry (2021.0) 58 00469580211060291. PMID: 34879719 12. Onwuzurike C, Meadows AR, Nour NM. Examining inequities associated with changes in obstetric and gynecologic care delivery during the coronavirus disease 2019 (COVID-19) pandemic. Obstet Gynecol (2020.0) 136 37-41. DOI: 10.1097/AOG.0000000000003933 13. Emeruwa UN, Gyamfi-Bannerman C, Miller RS. Health care disparities in the COVID-19 pandemic in the United States: a focus on obstetrics. Clin Obstet Gynecol (2022.0) 65 123-133. DOI: 10.1097/GRF.0000000000000665 14. 14.Dol J, Aston M, Grant A, McMillan D, Tomblin Murphy G, Campbell-Yeo M. Implementing Essential Coaching for Every Mother during the COVID-19 pandemic: A pre-post intervention study. Birth. 2022;49(2):273-80. 10.1111/birt.12603. 15. Burgess A, Breman RB, Bradley D, Dada S, Burcher P. Pregnant Women's reports of the impact of COVID-19 on pregnancy, prenatal care, and infant feeding plans. MCN Am J Matern Child Nurs (2021.0) 46 21-29. DOI: 10.1097/NMC.0000000000000673 16. Czeisler M, Marynak K, Clarke K. Delay or avoidance of medical Care because of COVID-19–related concerns — United State. MMWR Morb Mortal Wkly Rep (2020.0) 69 1250-1257. DOI: 10.15585/mmwr.mm6936a4 17. Goyal M, Singh P, Singh K, Shekhar S, Agrawal N, Misra S. The effect of the COVID-19 pandemic on maternal health due to delay in seeking health care: experience from a tertiary center. Int J Gynecol Obstet (2021.0) 152 231-235. DOI: 10.1002/ijgo.13457 18. Jardine J, Relph S, Magee LA, von Dadelszen P, Morris E, Ross-Davie M, Draycott T, Khalil A. Maternity services in the UK during the coronavirus disease 2019 pandemic: a national survey of modifications to standard care. BJOG Int J Obstet Gynaecol (2021.0) 128 880-889. DOI: 10.1111/1471-0528.16547 19. 19.Ogunbiyi MO. Psychopathology associated with coronavirus disease 2019 among pregnant women. Am J Obstet Gynecol MFM. 2021;3(1):100289–9. 20. Corbett GA, Milne SJ, Hehir MP, Lindow SW, O'Connell MP. Health anxiety and behavioural changes of pregnant women during the COVID-19 pandemic. Eur J Obstet Gynecol Reprod Biol (2020.0) 249 96-97. DOI: 10.1016/j.ejogrb.2020.04.022 21. Iyengar U, Jaiprakash B, Haitsuka H, Kim S. One year into the pandemic: a systematic review of perinatal mental health outcomes during COVID-19. Front Psychiatry (2021.0) 12 674194. DOI: 10.3389/fpsyt.2021.674194 22. 22.Daniels MDK, Monga MDM, Gupta S, Abir MFG, Chanisse M, MDFFC N. A strategy for disaster preparedness in obstetrics. Am J Disaster Med. 2021. SummerDO;16(3). 10.5055/ajdm.2021.0403. 23. Haeri S, Marcozzi D. Emergency preparedness in obstetrics. Obstet Gynecol (2015.0) 125 959-970. DOI: 10.1097/AOG.0000000000000750 24. Maher MJ. Emergency preparedness in obstetrics: meeting unexpected key challenges. J Perinat Neonatal Nurs (2019.0) 33 238-245. DOI: 10.1097/JPN.0000000000000421 25. 25.This Committee Opinion was developed by the American College of Obstetricians and Gynecologists’ Committee on Obstetric Practice in collaboration with committee member Alfred G. Robichaux, MD; the American Academy of Pediatrics’ Council on Environmental Health, liaison member Nathaniel DeNicola, MD, MSc; and Richard H. Beigi, MD. Committee Opinion No. 726: Hospital Disaster Preparedness for Obstetricians and Facilities Providing Maternity Care. Obstet Anesth Dig. 2018;38(2):85–86. 10.1097/01.aoa.0000532279.90472.30. 26. Peahl AF, Smith RD, Moniz MH. Prenatal care redesign: creating flexible maternity care models through virtual care. Am J Obstet Gynecol (2020.0) 223 389.e381-389.e310. DOI: 10.1016/j.ajog.2020.05.029 27. Peahl AF, Powell A, Berlin H, Smith RD, Krans E, Waljee J, Dalton VK, Heisler M, Moniz MH. Patient and provider perspectives of a new prenatal care model introduced in response to the coronavirus disease 2019 pandemic. Am J Obstet Gynecol (2021.0) 224 384.e381-384.e311. DOI: 10.1016/j.ajog.2020.10.008 28. Ehmer AC, Scott SM, Smith H, Ashby BD. Connecting during COVID: the application of teleservices in two integrated perinatal settings. Infant Ment Health J (2022.0) 43 127-139. DOI: 10.1002/imhj.21958 29. Galle A, Semaan A, Huysmans E, Audet C, Asefa A, Delvaux T, Afolabi BB, El Ayadi AM, Benova L. A double-edged sword—telemedicine for maternal care during COVID-19: findings from a global mixed-methods study of healthcare providers. BMJ Glob Health (2021.0) 6 e004575. DOI: 10.1136/bmjgh-2020-004575 30. 30.SDG Goal 3: Ensure healthy lives and promote well-being for all at all ages. https://unstats.un.org/sdgs/report/2017/goal-03/. 31. Moran AC, Jolivet RR, Chou D, Dalglish SL, Hill K, Ramsey K, Rawlins B, Say L. A common monitoring framework for ending preventable maternal mortality, 2015–2030: phase I of a multi-step process. BMC Pregnancy Childbirth (2016.0) 16 250. DOI: 10.1186/s12884-016-1035-4 32. 32.Maternal Mortality: Global Progress and Projections for Maternal Mortality. https://www.gatesfoundation.org/goalkeepers/report/2021-report/progress-indicators/maternal-mortality/. 33. Labrique AB, Vasudevan L, Kochi E, Fabricant R, Mehl G. mHealth innovations as health system strengthening tools: 12 common applications and a visual framework. Glob Health Sci Pract (2013.0) 1 160-171. DOI: 10.9745/GHSP-D-13-00031 34. Mosa AS, Yoo I, Sheets L. A systematic review of healthcare applications for smartphones. BMC Med Inform Decis Mak (2012.0) 12 67. DOI: 10.1186/1472-6947-12-67 35. Moise IK, Ortiz-Whittingham LR, Omachonu V, Sheskin IM, Patel R, Meguro JAS, Lucas AG, Bice W, Thompson LM. The impact of COVID-19 on service delivery systems: evidence from a survey of United States refugee resettlement agencies. BMC Health Serv Res (2022.0) 22 535. DOI: 10.1186/s12913-022-07909-3 36. Moise I, Ortiz-Whittingham L, Omachonu V, Clark M, Xue R-D. Fighting the bite during a crisis: capabilities of Florida Mosquito control districts during the COVID-19 pandemic. Research Square (2020.0) 37. Cruden G, Campbell M, Saldana L. Impact of COVID-19 on service delivery for an evidence-based behavioral treatment for families involved in the child welfare system. J Subst Abus Treat (2021.0) 129 108388. DOI: 10.1016/j.jsat.2021.108388 38. Murphy AA, Karyczak S, Dolce JN, Zechner M, Bates F, Gill KJ, Rothpletz-Puglia P. Challenges experienced by behavioral health organizations in New York resulting from COVID-19: a qualitative analysis. Community Ment Health J (2021.0) 57 111-120. DOI: 10.1007/s10597-020-00731-3 39. 39.Benjamen J, Girard V, Jamani S, Magwood O, Holland T, Sharfuddin N, Pottie K. Access to Refugee and Migrant Mental Health Care Services during the First Six Months of the COVID-19 Pandemic: A Canadian Refugee Clinician Survey. Int J Environ Res Public Health. 2021;18(10):5266. 10.3390/ijerph18105266. 40. Chattu VK, Lopes CA, Javed S, Yaya S. Fulfilling the promise of digital health interventions (DHI) to promote women’s sexual, reproductive and mental health in the aftermath of COVID-19. Reprod Health (2021.0) 18 1-8. DOI: 10.1186/s12978-021-01168-x 41. Majeed-Ariss R, Hall AG, McDonagh J, Fallon D, Swallow V. Mobile phone and tablet apps to support young People's Management of Their Physical Long-Term Conditions: a systematic review protocol. JMIR Res Protoc (2015.0) 4 e40. DOI: 10.2196/resprot.4159 42. Abejirinde IO, Ilozumba O, Marchal B, Zweekhorst M, Dieleman M. Mobile health and the performance of maternal health care workers in low- and middle-income countries: a realist review. Int J Care Coord (2018.0) 21 73-86. PMID: 30271609 43. Kabongo EM, Mukumbang FC, Delobelle P, Nicol E. Explaining the impact of mHealth on maternal and child health care in low- and middle-income countries: a realist synthesis. BMC Pregnancy Childbirth (2021.0) 21 196. DOI: 10.1186/s12884-021-03684-x 44. Kabongo EM, Mukumbang FC, Delobelle P, Nicol E. Understanding the influence of the MomConnect programme on antenatal and postnatal care service utilisation in two south African provinces: a realist evaluation protocol. BMJ Open (2019.0) 9 e029745. DOI: 10.1136/bmjopen-2019-029745 45. Aziz A, Zork N, Aubey JJ, Baptiste CD, D'Alton ME, Emeruwa UN, Fuchs KM, Goffman D, Gyamfi-Bannerman C, Haythe JH. Telehealth for high-risk pregnancies in the setting of the COVID-19 pandemic. Am J Perinatol (2020.0) 37 800-808. DOI: 10.1055/s-0040-1712121 46. Reynolds RM. Telehealth in pregnancy. Lancet Diabetes Endocrinol (2020.0) 8 459-461. DOI: 10.1016/S2213-8587(20)30158-3 47. Bush J, Barlow DE, Echols J, Wilkerson J, Bellevin K. Impact of a Mobile health application on user engagement and pregnancy outcomes among Wyoming Medicaid members. Telemed J E Health (2017.0) 23 891-898. DOI: 10.1089/tmj.2016.0242 48. Avalos LA, Aghaee S, Kurtovich E, Quesenberry C, Nkemere L, McGinnis MK, Kubo A. A mobile health mindfulness intervention for women with moderate to moderately severe postpartum depressive symptoms: feasibility study. JMIR Mental Health (2020.0) 7 e17405. DOI: 10.2196/17405 49. Coleman J, Bohlin KC, Thorson A, Black V, Mechael P, Mangxaba J, Eriksen J. Effectiveness of an SMS-based maternal mHealth intervention to improve clinical outcomes of HIV-positive pregnant women. AIDS Care (2017.0) 29 890-897. DOI: 10.1080/09540121.2017.1280126 50. Venter W, Coleman J, Chan VL, Shubber Z, Phatsoane M, Gorgens M, Stewart-Isherwood L, Carmona S, Fraser-Hurt N. Improving linkage to HIV care through Mobile phone apps: randomized controlled trial. JMIR Mhealth Uhealth (2018.0) 6 e155. DOI: 10.2196/mhealth.8376 51. Wolff-Piggott B, Coleman J, Rivett U. The clinic-level perspective on mHealth implementation: a south African case study. Inf Technol Dev (2018.0) 24 532-553. DOI: 10.1080/02681102.2016.1233858 52. LeFevre A, Cabrera-Escobar MA, Mohan D, Eriksen J, Rogers D, Neo Parsons A, Barre I, Jo Y, Labrique A, Coleman J. Forecasting the value for money of Mobile maternal health information messages on improving utilization of maternal and child health Services in Gauteng, South Africa: cost-effectiveness analysis. JMIR Mhealth Uhealth (2018.0) 6 e153. DOI: 10.2196/mhealth.8185 53. Coleman J, Eriksen J, Black V, Thorson A, Hatcher A. The Mobile Alliance for maternal action text message-based mHealth intervention for maternal Care in South Africa: qualitative user study. JMIR Hum Factors (2020.0) 7 e14078. DOI: 10.2196/14078 54. 54.Kazi AM, Carmichael J-L, Hapanna GW, Wangoo PG, Karanja S, Wanyama D, et al. Assessing Mobile phone access and perceptions for texting-based mHealth interventions among expectant mothers and child caregivers in remote regions of northern Kenya: a survey-based descriptive study. JMIR Public Health Surveill. 2017;3(1):e5–5. 55. Carmichael SL, Mehta K, Srikantiah S, Mahapatra T, Chaudhuri I, Balakrishnan R, Chaturvedi S, Raheel H, Borkum E, Trehan S. Use of mobile technology by frontline health workers to promote reproductive, maternal, newborn and child health and nutrition: a cluster randomized controlled trial in Bihar, India. J Glob Health (2019.0) 9 0204249. DOI: 10.7189/jogh.09.020424 56. Choudhury A, Asan O, Choudhury MM. Mobile health technology to improve maternal health awareness in tribal populations: mobile for mothers. J Am Med Inform Assoc (2021.0) 28 2467-2474. DOI: 10.1093/jamia/ocab172 57. 57.Sondaal SFV, Browne JL, Amoakoh-Coleman M, Borgstein A, Miltenburg AS, Verwijs M, et al. Assessing the effect of mHealth interventions in improving maternal and neonatal Care in low- and Middle-Income Countries: a systematic review. PLoS One. 2016;11(5):e0154664–4. 58. Chen H, Chai Y, Dong L, Niu W, Zhang P. Effectiveness and appropriateness of mHealth interventions for maternal and child health: systematic review. JMIR Mhealth Uhealth (2018.0) 6 e7. DOI: 10.2196/mhealth.8998 59. Carter J, Sandall J, Shennan AH, Tribe RM. Mobile phone apps for clinical decision support in pregnancy: a scoping review. BMC Med Inform Decis Mak (2019.0) 19 219. DOI: 10.1186/s12911-019-0954-1 60. DeNicola N, Grossman D, Marko K, Sonalkar S, Butler Tobah YS, Ganju N, Witkop CT, Henderson JT, Butler JL, Lowery C. Telehealth interventions to improve obstetric and gynecologic health outcomes: a systematic review. Obstet Gynecol (2020.0) 135 371-382. DOI: 10.1097/AOG.0000000000003646 61. Dosani A, Arora H, Mazmudar S. mHealth and perinatal depression in low-and middle-income countries: a scoping review of the literature. Int J Environ Res Public Health (2020.0) 17 7679. DOI: 10.3390/ijerph17207679 62. Feroz A, Abrejo F, Ali SA, Nuruddin R, Saleem S. Using mobile phones to improve young people’s sexual and reproductive health in low- and middle-income countries: a systematic review protocol to identify barriers, facilitators and reported interventions. Syst Rev (2019.0) 8 117. DOI: 10.1186/s13643-019-1033-5 63. Free C, Phillips G, Watson L, Galli L, Felix L, Edwards P, Patel V, Haines A. The effectiveness of mobile-health technologies to improve health care service delivery processes: a systematic review and meta-analysis. PLoS Med (2013.0) 10 e1001363. DOI: 10.1371/journal.pmed.1001363 64. Gagnon M-P, Ngangue P, Payne-Gagnon J, Desmartis M. M-health adoption by healthcare professionals: a systematic review. J Am Med Inform Assoc (2016.0) 23 212-220. DOI: 10.1093/jamia/ocv052 65. 65.Elwood D, Diamond MC, Heckman J, Bonder JH, Beltran JE, Moroz A, Yip J. Mobile Health: Exploring Attitudes Among Physical Medicine and Rehabilitation Physicians Toward this Emerging Element of Health Delivery. PM&R. 2011;3:678–80. 10.1016/j.pmrj.2011.05.004. 66. 66.Ogundaini OO, de la Harpe R, McLean N. Integration of mHealth information and communication technologies into the clinical settings of hospitals in sub-Saharan Africa: qualitative study. JMIR mHealth and uHealth. 2021;9(10):e26358–8. 67. O'Connor Y, Andreev P, O'Reilly P. MHealth and perceived quality of care delivery: a conceptual model and validation. BMC Med Inform Decis Mak (2020.0) 20 41. DOI: 10.1186/s12911-020-1049-8 68. Jayasinghe I, Wickramasinghe Y, Kurera DM, Samarakoon N, Rupasinghe I, Ponmalay P, Agampodi SB. Feasibility of using telephone interviews and internet-based message services during the COVID-19 pandemic in rural Sri Lanka: experiences of the Rajarata pregnancy cohort. Rural Remote Health (2022.0) 22 7442. PMID: 35546146 69. Omowale SS, Casas A, Lai Y-H, Sanders SA, Hill AV, Wallace ML, Rathbun SL, Gary-Webb TL, Burke LE, Davis EM. Trends in stress throughout pregnancy and postpartum period during the COVID-19 pandemic: longitudinal study using ecological momentary assessment and data from the postpartum mothers Mobile study. JMIR Mental Health (2021.0) 8 e30422. DOI: 10.2196/30422 70. Feroz A, Perveen S, Aftab W. Role of mHealth applications for improving antenatal and postnatal care in low and middle income countries: a systematic review. BMC Health Serv Res (2017.0) 17 704. DOI: 10.1186/s12913-017-2664-7 71. Colaci D, Chaudhri S, Vasan A. mHealth interventions in low-income countries to address maternal health: a systematic review. Ann Glob Health (2016.0) 82 922-935. DOI: 10.1016/j.aogh.2016.09.001 72. Madden N, Emeruwa UN, Friedman AM, Aubey JJ, Aziz A, Baptiste CD, Coletta JM, D'Alton ME, Fuchs KM, Goffman D. Telehealth uptake into prenatal care and provider attitudes during the COVID-19 pandemic in new York City: a quantitative and qualitative analysis. Am J Perinatol (2020.0) 37 1005-1014. DOI: 10.1055/s-0040-1712939 73. Jeganathan S, Prasannan L, Blitz MJ, Vohra N, Rochelson B, Meirowitz N. Adherence and acceptability of telehealth appointments for high-risk obstetrical patients during the coronavirus disease 2019 pandemic. Am J Obstet Gynecol MFM (2020.0) 2 100233. DOI: 10.1016/j.ajogmf.2020.100233 74. Jakubowski D, Sys D, Kajdy A, Lewandowska R, Kwiatkowska E, Cymbaluk-Płoska A, Rabijewski M, Torbé A, Kwiatkowski S. Application of telehealth in prenatal care during the COVID-19 pandemic-a cross-sectional survey of polish women. J Clin Med (2021.0) 10 2570. DOI: 10.3390/jcm10122570 75. 75.Tozour JN, Bandremer S, Patberg E, Zavala J, Akerman M, Chavez M, et al. Application of telemedicine video visits in a maternal-fetal medicine practice at the epicenter of the COVID-19 pandemic. Am J Obstet Gynecol MFM. 2021;3(6):100469–9. 76. Özkan Şat S, Yaman Sözbir Ş. 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 77. Quinn LM, Olajide O, Green M, Sayed H, Ansar H. Patient and professional experiences with virtual antenatal clinics during the COVID-19 pandemic in a UK tertiary obstetric hospital: questionnaire study. J Med Internet Res (2021.0) 23 e25549. DOI: 10.2196/25549 78. Rousseau A, Gaucher L, Gautier S, Mahrez I, Baumann S. How midwives implemented teleconsultations during the COVID-19 health crisis: a mixed-methods study. BMJ Open (2022.0) 12 e057292. DOI: 10.1136/bmjopen-2021-057292 79. 79.Ramnund S, Baloyi OB, Nkwanyana NM, Jarvis MA. Midwives’ acceptance of mHealth applications in the dissemination of maternal health education in primary healthcare clinics. Information Development. 2021. 10.1177/02666669211049140. 80. 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 81. Arksey H, O'Malley L. Scoping studies: towards a methodological framework. Int J Soc Res Methodol (2005.0) 8 19-32. DOI: 10.1080/1364557032000119616 82. Moher D, Liberati A, Tetzlaff J, Altman DG. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. BMJ (2009.0) 339 b2535. DOI: 10.1136/bmj.b2535 83. Ronen K, Choo EM, Wandika B, Udren JI, Osborn L, Kithao P, Hedstrom AB, Masinde M, Kumar M, Wamalwa DC. Evaluation of a two-way SMS messaging strategy to reduce neonatal mortality: rationale, design and methods of the Mobile WACh NEO randomised controlled trial in Kenya. BMJ Open (2021.0) 11 e056062. DOI: 10.1136/bmjopen-2021-056062 84. Ronen K, Choo EM, Wandika B, Udren J, Osborn L, Kinthao P, Hedstrom A, Masinde M, Kumar M, Wamalwa D. Evaluation of a two-way SMS mHealth strategy to reduce neonatal mortality: rationale, design and methods of a randomized controlled trial in Kenya (2021.0) 85. Chen H, Selix N, Nosek M. Perinatal anxiety and depression during COVID-19. J Nurse Pract (2021.0) 17 26-31. DOI: 10.1016/j.nurpra.2020.09.014 86. Graseck A, Leitner K. Prenatal education in the digital age. Clin Obstet Gynecol (2021.0) 64 345-351. DOI: 10.1097/GRF.0000000000000608 87. Kingsley JP, Vijay PK, Kumaresan J, Sathiakumar N. The changing aspects of motherhood in face of the COVID-19 pandemic in low- and middle-income countries. Matern Child Health J (2021.0) 25 15-21. DOI: 10.1007/s10995-020-03044-9 88. Saad A. Improving the health equity of women now and in the post COVID-19 era: Mobile technology-assisted mental health interventions for pregnant and postpartum women (2021.0) 89. Delmaifanis D, Siregar K, Prabawa A. Mhealth conceptual model for providing quality antenatal care in health centers during the coronavirus disease 2019 pandemic. Open access Macedonian. J Med Sci (2021.0) 9 828-834 90. Gurol-Urganci I, Waite L, Webster K, Jardine J, Carroll F, Dunn G, Frémeaux A, Harris T, Hawdon J, Muller P. Obstetric interventions and pregnancy outcomes during the COVID-19 pandemic in England: a nationwide cohort study. PLoS Med (2022.0) 19 1-15. DOI: 10.1371/journal.pmed.1003884 91. 91.Jiang H, Jin L, Qian X, Xiong X, La X, Chen W, et al. Maternal Mental Health Status and Approaches for Accessing Antenatal Care Information During the COVID-19 Epidemic in China: Cross-Sectional Study. J Med Internet Res. 2021;23(1):e18722. 10.2196/18722. 92. 92.The EndNote Team: EndNote. In., vol. EndNote X9. Philadelphia, PA: Clarivate; 2013. 93. 93.Covidence Systematic Review Software, Veritas Health Innovation, Melbourne, Australia. Available online: www.covidence.org. Accessed 5 Dec 2022. 94. Joanna Briggs I. The Joanna Briggs institute critical appraisal tools for use in JBI systematic reviews. Critical Appraisal Checklist for Analytical Cross Sectional Studies (2017.0) 95. Borschmann R, Janca E, Carter A, Willoughby M, Hughes N, Snow K, Stockings E, Hill NTM, Hocking J, Love A. The health of adolescents in detention: a global scoping review. Lancet Public Health (2020.0) 5 e114-e126. DOI: 10.1016/S2468-2667(19)30217-8 96. Kiani N, Pirzadeh A. Mobile-application intervention on physical activity of pregnant women in Iran during the COVID-19 epidemic in 2020. J Educ Health Promot (2021.0) 10 328. DOI: 10.4103/jehp.jehp_56_21 97. Smith RB, Mahnert ND, Foote J, Saunders KT, Mourad J, Huberty J. Mindfulness effects in obstetric and gynecology patients during the coronavirus disease 2019 (COVID-19) pandemic: a randomized controlled trial. Obstet Gynecol (2021.0) 137 1032-1040. DOI: 10.1097/AOG.0000000000004316 98. Kubo A, Aghaee S, Kurtovich EM, Nkemere L, Quesenberry CP, McGinnis MK, Avalos LA. mHealth mindfulness intervention for women with moderate-to-moderately-severe antenatal depressive symptoms: a pilot study within an integrated health care system. Mindfulness (2021.0) 12 1387-1397. DOI: 10.1007/s12671-021-01606-8 99. 99.Dol J, Aston M, Grant A, McMillan D, Tomblin Murphy G, Campbell-Yeo M. Effectiveness of the “Essential Coaching for Every Mother” postpartum text message program on maternal psychosocial outcomes: A randomized controlled trial. Digital Health. 2022;8. 10.1177/20552076221107886. 100. Leight J, Hensly C, Chissano M, Safran E, Ali L, Dustan D, Jamison J. The effects of text reminders on the use of family planning services: evidence from a randomised controlled trial in urban Mozambique. BMJ Glob Health (2022.0) 7 e007862. DOI: 10.1136/bmjgh-2021-007862 101. Krishnamurti T, Davis AL, Quinn B, Castillo AF, Martin KL, Simhan HN. Mobile remote monitoring of intimate partner violence among pregnant patients during the COVID-19 shelter-in-place order: quality improvement pilot study. J Med Internet Res (2021.0) 23 e22790. DOI: 10.2196/22790 102. Varnfield M, Redd C, Stoney RM, Higgins L, Scolari N, Warwick R, Iedema J, Rundle J, Dutton W. M♡THer, an mHealth system to support women with gestational diabetes mellitus: feasibility and acceptability study. Diabetes Technol Ther (2021.0) 23 358-366. DOI: 10.1089/dia.2020.0509 103. Moulaei K, Bahaadinbeigy K, Ghaffaripour Z, Ghaemi MM. The design and evaluation of a Mobile based application to facilitate self-care for pregnant women with preeclampsia during COVID-19 prevalence. J Biomed Phys Eng (2021.0) 11 551-560. PMID: 34458202 104. Moulaei K, Sheikhtaheri A, Ghafaripour Z, Bahaadinbeigy K. The development and usability assessment of an mHealth application to encourage self-Care in Pregnant Women against COVID-19. J Healthc Eng (2021.0) 2021 9968451. DOI: 10.1155/2021/9968451 105. Ceballos F, Hernandez MA, Olivet F, Paz C. Assessing the use of cell phones to monitor health and nutrition interventions: evidence from rural Guatemala. PLoS One (2020.0) 15 1. DOI: 10.1371/journal.pone.0240526 106. Rhodes A, Kheireddine S, Smith AD. Experiences, attitudes, and needs of users of a pregnancy and parenting app (baby buddy) during the COVID-19 pandemic: mixed methods study. JMIR Mhealth Uhealth (2020.0) 8 e23157. DOI: 10.2196/23157 107. Wulandari R, Suwandono A, Kartasurya MI, Nugraheni SA. Postpartum care behavior improvement during COVID-19 pandemic in Indonesia using Mobile-health interactive message. Ethiop J Health Sci (2022.0) 32 243-254. PMID: 35693576 108. Shorey S, Tan TC, Mathews J, Yu CY, Lim SH, Shi L, Ng ED, Chan YH, Law E, Chee C. Development of a supportive parenting app to improve parent and infant outcomes in the perinatal period: development study. J Med Internet Res (2021.0) 23 e27033. DOI: 10.2196/27033 109. 109.Quifer-Rada P, Aguilar-Camprubí L, Padró-Arocas A, Gómez-Sebastià I, Mena-Tudela D. Impact of COVID-19 Pandemic in Breastfeeding Consultations on LactApp, an m-Health Solution for Breastfeeding Support. Telemed J E Health. 2022;28(10):1449–57. 10.1089/tmj.2021.0586. 110. Hashemzahi M, Khayat S, Khazayan S. Effect of COVID-19 self-care training via telemedicine on perceived stress and Corona disease anxiety in pregnant women: a quasi-experimental study. J Midwif Reprod Health (2022.0) 10 3066-3074 111. Silva-Jose C, Nagpal TS, Coterón J, Barakat R, Mottola MF. The ‘new normal’ includes online prenatal exercise: exploring pregnant women’s experiences during the pandemic and the role of virtual group fitness on maternal mental health. BMC Pregnancy Childbirth (2022.0) 22 251. DOI: 10.1186/s12884-022-04587-1 112. Nur R, Aulia U, Dwilarasati RR, Patui NS, Mantao E, Hendra S, Ngemba HR, Rusydi M, Dominggo NP. Effects of electronic technology antenatal care (E-anc) on midwives and pregnant women during the covid-19 period. Open Access Maced J Med Sci (2020.0) 8 115-121. DOI: 10.3889/oamjms.2020.4923 113. Palmer KR, Tanner M, Davies-Tuck M, Rindt A, Papacostas K, Giles ML, Brown K, Diamandis H, Fradkin R, Stewart AE. Widespread implementation of a low-cost telehealth service in the delivery of antenatal care during the COVID-19 pandemic: an interrupted time-series analysis. Lancet (2021.0) 398 41-52. DOI: 10.1016/S0140-6736(21)00668-1 114. Reisinger-Kindle K, Qasba N, Cayton C, Niakan S, Knee A, Goff SL. Evaluation of rapid telehealth implementation for prenatal and postpartum care visits during the COVID-19 pandemic in an academic clinic in Springfield, Massachusetts, United States of America. Health Sci Rep (2021.0) 4 e455. DOI: 10.1002/hsr2.455 115. Ahmad M, Vismara L. The psychological impact of COVID-19 pandemic on Women’s mental health during pregnancy: a rapid evidence review. Int J Environ Res Public Health (2021.0) 18 7112. DOI: 10.3390/ijerph18137112 116. 116.Herval ÁM, Oliveira DPD, Gomes VE, Vargas AMD. Health education strategies targeting maternal and child health: a scoping review of educational methodologies. Medicine. 2019;98(26):e16174–4. 117. 117.Lumbiganon P, Martis R, Laopaiboon M, Festin MR, Ho JJ, Hakimi M. Antenatal breastfeeding education for increasing breastfeeding duration. Cochrane Database Syst Rev. 2016;12(12):CD006425. 10.1002/14651858.CD006425.pub4. 118. McFadden A, Gavine A, Renfrew MJ, Wade A, Buchanan P, Taylor JL, Veitch E, Rennie AM, Crowther SA, Neiman S. Support for healthy breastfeeding mothers with healthy term babies. Cochrane Database Syst Rev (2017.0) 2 Cd001141. PMID: 28244064 119. Dyson L, McCormick FM, Renfrew MJ. Interventions for promoting the initiation of breastfeeding. Cochrane Database Syst Rev (2005.0) 2 CD001688 120. 120.Virani A, Duffett-Leger L, Letourneau N: Parenting apps review: in search of good quality apps. mHealth. 2019;5: mHealth 2019. 121. Byrskog U, Olsson P, Essén B, Allvin MK. Being a bridge: Swedish antenatal care midwives’ encounters with Somali-born women and questions of violence; a qualitative study. BMC Pregnancy Childbirth (2015.0) 15 1. DOI: 10.1186/s12884-015-0429-z 122. Butler Tobah YS, LeBlanc A, Branda ME, Inselman JW, Morris MA, Ridgeway JL, Finnie DM, Theiler R, Torbenson VE, Brodrick EM. Randomized comparison of a reduced-visit prenatal care model enhanced with remote monitoring. Am J Obstet Gynecol (2019.0) 221 638.e631-638.e638. DOI: 10.1016/j.ajog.2019.06.034 123. Morgan A, Goodman D, Vinagolu-Baur J, Cass I. Prenatal telemedicine during COVID-19: patterns of use and barriers to access. JAMIA Open (2022.0) 5 ooab116. DOI: 10.1093/jamiaopen/ooab116 124. 124.Oelmeier K, Schmitz R, Möllers M, Braun J, Deharde D, Sourouni M, et al. Satisfaction with and Feasibility of Prenatal Counseling via Telemedicine: A Prospective Cohort Study. Telemed J E Health. 2022;28(8):1193–8. 10.1089/tmj.2021.0309. 125. Padró-Arocas A, Quifer-Rada P, Aguilar-Camprubí L, Mena-Tudela D. Description of an mHealth tool for breastfeeding support: LactApp. Analysis of how lactating mothers seek support at critical breastfeeding points and according to their infant’s age. Res Nurs Health (2021.0) 44 173-186. DOI: 10.1002/nur.22095 126. Shorey S, Chan V. Lessons from past epidemics and pandemics and a way forward for pregnant women, midwives and nurses during COVID-19 and beyond: a meta-synthesis. Midwifery (2020.0) 90 102821. DOI: 10.1016/j.midw.2020.102821 127. Yang Y, Chen H, Qazi H, Morita PP. Intervention and evaluation of Mobile health Technologies in Management of patients undergoing chronic Dialysis: scoping review. JMIR Mhealth Uhealth (2020.0) 8 e15549. DOI: 10.2196/15549
--- title: Physical and mental health impact of the COVID-19 pandemic at first year in a Spanish adult cohort authors: - Pere Castellvi Obiols - Andrea Miranda-Mendizabal - Silvia Recoder - Ester Calbo Sebastian - Marc Casajuana-Closas - David Leiva - Rumen Manolov - Nuria Matilla-Santander - Isaac Lloveras-Bernat - Carlos G. Forero journal: Scientific Reports year: 2023 pmcid: PMC10026238 doi: 10.1038/s41598-023-28336-2 license: CC BY 4.0 --- # Physical and mental health impact of the COVID-19 pandemic at first year in a Spanish adult cohort ## Abstract The COVID-19 pandemic and the political and health measures have profoundly affected the health of our populations. However, very few studies have been published assessing its impact using a prospective cohort. The aim of this study is to describe the impact on physical and mental health due to the COVID-19 pandemic in the general population in Spain, and according to COVID-19 clinical status, during the first year of the pandemic. A longitudinal cohort study with two online surveys were performed on a representative sample of the adult Spanish population before ($$n = 2005$$, October/November 2019) and during the pandemic ($$n = 1357$$, November/December 2020). We assessed disability using the World Health Organisation Disability Assessment Schedule (WHODAS), major depressive episode (MDE) and suicidal thoughts and behaviours (STB), using an adapted version of the Composite International Diagnostic Interview (CIDI 3.0); generalised anxiety disorder (GAD) using the GAD-7 scale; post-traumatic stress disorder (PTSD) symptoms using the PTSD checklist for DSM-5 (PCL-5). For physical health, there was a statistically significant loss of weight (mean/SD) (T0, $\frac{73.22}{15.56}$ vs. T1, $\frac{71.21}{11.94}$), less use of tobacco (T0, $11.4\%$ vs. T1, $9.0\%$) and decreased disability (mean/SD) (T0, $\frac{21.52}{9.22}$ vs. T1, $\frac{19.03}{7.32}$). For mental health, there was a significant increase in MDE (T0, $6.5\%$ vs. T1, $8.8\%$) and in the prevalence of GAD (T0, $13.7\%$ vs. T1, $17.7\%$). The prevalence of STB (T0, $15.1\%$ vs. T1, $7.1\%$) significantly decreased. Individuals who declared they had been diagnosed with COVID-19 ($3.6\%$) showed a worsening in physical health and an increase in mental health problems and PTSD symptoms. Although suicide risk during the first year of the pandemic was significantly less, many suicide risk factors increased: such as the incidence and persistence of MDE and GAD, the presence of PTSD symptoms in those diagnosed with COVID-19, and a worsening in self-assessed health status. We expect an increase in STB in the population in the long-term. Future research should gather information about the long-term impact of the pandemic. ## Introduction The COVID-19 pandemic and the political and health measures taken to curb the spread of SARS-CoV-2 have profoundly affected every aspect of day-to-day life1. Spain, with its first case on 31 January 2020, is one of the countries in Europe most affected by infections, complications, and deaths2. It was not until February 24 when Spain confirmed several new COVID-19 cases related to a recent SARS-CoV-2 outbreak in the north of Italy. Since that date, the number of COVID-19 cases grew exponentially in Spain so that by March 30, there were over 85,199 confirmed cases, and 7424 deaths, according to the official numbers. On March 25, the death toll attributed to COVID-19 in Spain surpassed that of mainland China. The economic and social impact of the COVID-19 pandemic in Spain is without precedent. To combat the pandemic, the Spanish Government implemented a series of social distancing and mobility restriction measures. First, all classes at all educational levels were cancelled in the main hot spots of the disease on March 10. On March 14, the Government of Spain declared a state of emergency for 2 weeks across the entire country closing all schools and university classes, and workers were encouraged to tele-work. Despite these efforts, the daily growth rate in the number of confirmed COVID-19 cases continued to grow. Thus, on March 30, new mobility restriction and social distancing measures were implemented; all nonessential labour activity was to be interrupted for a 2-week period. Moreover, the Spanish Government extended the state of emergency until June 212. Although these interventions put a halt to the normal daily lives of most people in Spain, their impact on people’s economic, physical, and mental well-being were unknown at the time. Many other countries implemented similar measures. Studies show an impact on employment and livelihoods, income, and personal debt3, coupled with increased worries about future job insecurity and probable physical and psychological worsening4–6. Since the emergence of the COVID-19 pandemic, substantial efforts have been made to understand its transmission and to assess the socio-economic and health impact of the pandemic, due to the political measures, economic recession, and social crises. Previous literature suggests a link between pandemics and a worsening in physical health, such as an increase in obesity rates7, chronic physical symptoms, frailty, coronary heart disease, malnutrition, hospital readmission and early mortality8. The lockdown culture, loneliness, socio-economic instability, changes in eating habits and an increase in sedentary, domestic activities might have a further deleterious effect on physical health9. Mixed results have been found regarding the impact of the pandemic on mental health. In Denmark10 and the United Kingdom1, results from a cohort study showed worsening mental health among the general population. However, the COVID-19 pandemic does not seem to have further increased depressive, anxiety and obsessive-compulsive symptom severity, compared with pre-pandemic levels in The Netherlands11. As for Spanish data, the general adult population has mostly reported an increase of depressive12 and anxiety symptoms following the immediate consequences of the first wave of the COVID-19 pandemic (spring 2020). However, these previous studies in Spain used a retrospective cross-sectional design. In an earlier published article using a longitudinal, population-based cohort study of Spaniards just after the first wave of infection (spring/summer 2020), the prevalence of depression and suicidal ideation were not significantly increased13. However, there is a need to know the medium- and long-term mental health impact of the COVID-19 pandemic on the population, using a prospective, longitudinal design study, with assessments before and during the pandemic. Through this a prospective cohort, it allows us to measure changes of some physical and mental health outcomes in the Spanish population and uncover temporality, which is one of the postulates about causality, making comparisons with prepandemic data, and evaluate changes in the health of our population. The aim of this study is to describe the impact on physical and mental health due to the COVID-19 pandemic in the general population in Spain, and according to COVID-19 clinical status, during the first year of the pandemic. ## Study design A longitudinal cohort study, with two assessments from an online survey of Spanish residents was carried out. The baseline reference survey (T0) was acquired as part of the BIOVAL-D study (ISCII-FEDER Exp: PI$\frac{16}{00165}$) during October/November 2019 and assessed mental health prior to the SARS-CoV-2 outbreak. The follow-up survey (T1) was conducted after 12 months (November/December 2020) using the same questionnaire and adding some extra dimensions and variables to identify physical and mental health outcomes, their risk and associated protective factors during the SARS-CoV-2 outbreak during and after lockdown, using clinical characteristics of people diagnosed with COVID-19. The survey had an approximate duration of 30 min. This study was approved by the Ethical Committee of the institution (Reg. No.: MED-2020-02) and has been performed in accordance with the Declaration of Helsinki. An informed consent form was signed by each participant using an online version. Data were recorded in a centralized database and anonymized before statistical analysis and shared for all authors. ## Study sample A Spanish, nationally representative, population-based sample (≥ 18 years old) was chosen, representative of geographical, sex, age and socioeconomic status. At the baseline assessment (T0), participants were recruited from a secure online panel data vendor, resulting in a final sample of 2005 individuals. For the follow-up survey, all 2005 participants were contacted by the panel provider and invited to participate. Participants received an informative email on the study objectives and characteristics, including a link to an informed consent form which acted as a filter for entering the survey. Baseline participants were offered participation in the follow-up and those who did not sign consent forms or did not fully fill in the survey were excluded. From the initial 2005 individuals, 941 participants answered the follow-up survey; a participation rate of $46.9\%$. To ensure representativeness of the post-pandemic sample, an additional representative panel was invited to participate. Participants from the additional panel who were invited to participate at the follow-up were matched by sex, geographical residence and age, to ensure similar characteristics to the baseline participants who did not respond to the follow-up. From this additional panel, 416 participants were recruited; giving a total of 1357 participants included in the analysis at 12 months. ## Socio-demographic variables (T0) Age, sex, marital and employment status were recorded. Age was a continuous variable; Sex had two response options Male and Female; Marital status was recorded as Single, Married or Living with a couple, Separated, Divorced and Widowed; and, finally, Employment status as Employed, Off sick, Unemployed, Homemaker, Student, Both student & employed, Temporal or permanent disability, and Retired. These socio-demographic variables were considered as nominal. ## Health status (T0 & T1) Physical and mental health and smoking status were self-assessed. The short version of the World Health Organisation Disability Assessment Schedule (WHODAS 2.0, 12 items) was used for assessing disability, and it is recommended for epidemiological studies. The WHODAS 2.0 showed have excellent internal consistency in all languages (alpha ≥ 0.90)14.This is a 12-item, self-administered scale. Items are grouped by pairs in 6 domains: Understanding and communicating with the world; Moving and getting around; Self-care; Getting along with people; Activities of daily living (domestic responsibilities, leisure, and work); and Participation in society. Scoring is on a range of 12–60, where 12 means no disability and 60 the highest disability. Response options were using a Likert Scale (None = 1; Mild = 2; Moderate = 3; Severe = 4; Extreme or cannot do = 5). The scale has been validated for the Spanish population15. Internal consistency in our sample was good in both assessments (baseline, alpha = 0.93; follow-up, alpha = 0.87). The distribution of disability using WHODAS 2.0 ranged from 12 to 58 at T0, and from 12 to 49 at T1, showing a normal distribution. This variable was considered as a continuous. Physical and mental health self-assessment and reported health transition one year ago were provided using The MOS 36-Item Short-Form Health Survey (SF-36)16. The SF-36 is a widely used and patient-reported measure of health status. Physical and mental health were assessed using 2 items with 5 response options (Excellent; Very Good; Good; Fair; Poor). Health transition response options were: Much better; Somewhat better now; About the same; Somewhat worse now; Much worse). These variables were considered as an ordinal. The Spanish version of the SF-36 has been used17. Self-reported anthropometric measurements relating to body mass index (BMI) were collected. BMI was recorded as an ordinal scale which: < 18.5 kg/m2: underweight; 18.5–24.9 kg/m2: normal; 25–29.9 kg/m2: overweight; and ≥ 30 kg/m2: obese. ## COVID-19 exposure (T1) Items about COVID-19 exposure were developed ad hoc for this study. Data about having been tested for or diagnosed with COVID-19 were gathered, including related symptoms. Items developed were “Have you ever been tested for COVID-19?”, response options were “Yes/No/No, although I had related symptoms, I was not tested”. If the subject answered Yes, then an additional question was administered “Was this test positive?”. COVID-19 clinical status was classified into 4 groups: Group 1, individuals with no COVID-19 symptoms and no COVID-19 test done; Group 2, those with COVID-19 symptoms with no test done; Group 3, those with COVID-19 test done with a negative result; and Group 4, those with COVID-19 test done with a positive result. Related symptoms administered in case an individual responded “Yes” or “No, although I had related symptoms, I was not tested” were: (a) Cough; (b) Fever; (c) Difficulty breathing or shortness of breath; (d) Sore throat when you drink any liquid; (e) Loss of smell; (f) Loss of taste; (g) Muscle aches; (h) Diarrhoea; (i) Chest pain; (j) Headache; (k) Coughing up blood; (l) Vomiting; (m) Feeling confused; (n) Feeling drowsy; (o) Feeling very tired; (p) Had other related symptoms; (q) Didn't have any symptoms. Information about the number of friends and relatives infected with COVID-19 and their mortality were also assessed using a continuous variable. Finally, stress related to the COVID-19 outbreak and its possible effects (e.g., family finances, increased social isolation and worry about getting infected) were also evaluated using six items using a Likert scale with five response options: Not at all; A bit; Quite; A lot of; Very much. Cronbach’s alpha of stress related to COVID-19 in our sample was 0.83 showing good internal consistency. For some analyses, COVID-19 clinical status was collapsed merging Group 1 and Group 3 versus Group 2 and Group 4 due to the small number of individuals in some groups. ## Major depressive episode (T0 & T1) For the assessment of Major Depressive Episode (MDE), the full screening section (8 items) from the Composite International Diagnostic Interview (CIDI) version 3.0 was used18. This section works as a filter to enter the diagnosis section, meaning that only those who answer some of the items positively go on to answer following questions. The diagnostic section includes 37 items divided into 8 sections: depression and anhedonia (6 items); weight (5items); insomnia (5 items); retardation and agitation (4 items); fatigue (2 items); worthlessness and guilt (5 items); concentration (4 items); suicide (6 items) that evaluate the presence or absence of MDE symptoms for at least two weeks. When five criteria were achieved, individuals must have a high grade of disability of > 50 in WHODAS to establish the diagnosis19. The area under the curve was 0.7520. The CIDI has been translated into many languages, including Spanish18. The prevalence at 12 months was assessed at T0, and since the lockdown started (March 14) at T1. ## Generalised anxiety disorder (T0 & T1) The Generalised Anxiety Disorder-7 scale (GAD-7) was administered, which consists of 7 items answered with a 4-point Likert scale and total scores ranging from 0 to 21. Point prevalence (2 weeks) was assessed at T0, and since the lockdown started (March 14) at T1. For the Spanish version, Cronbach's alpha coefficient of 0.93 was obtained. Taking into account the 10-point cut-off, sensitivity values of $86.8\%$ and specificity of $93.4\%$ were found21. To establish a diagnosis, individuals must also have a high degree of disability of > 50 in WHODAS19. The GAD-7 was administered to all those with positive depression screening using CIDI instrument and, additionally, a randomized $40\%$ with negative screening of the baseline sample ($$n = 722$$) and the entire sample at the follow-up assessment ($$n = 1357$$). ## Symptoms of post-traumatic stress disorder (PTSD) (T1) To assess DSM-5 symptoms of PTSD related to the experience of being infected with COVID-19 or the death of somebody close due to COVID-19, an adapted Spanish version of the PTSD Checklist for DSM-5 (PCL-5) was used (20 items). Responses to each item are rated using a 5-point scale, ranging from 1 (Not at all) to 5 (Extremely), and indicating the extent to which respondents had been bothered by that symptom in the past 7 days. Scoring ranges from 20 to 100. The higher the score, the more symptoms of PTSD. The PCL-5 demonstrated that the scale had solid psychometric properties (alpha = 0.97; ICC = 0.96; and convergent validity with other PTSD symptom scales)22. The differential item functioning of the PCL-5 scale score indicated that the Spanish version is equivalent to the original language23. The PCL-5 was adapted ad hoc in the context of being exposed to the COVID-19 pandemic (e.g. …avoid memories, thoughts or feelings related to being infected or someone has died from COVID-19). The PCL-5 was administered to all those with positive COVID-19 test or with any relative or someone known infected by COVID-19 and, additionally, a randomized $20\%$ of the rest of the follow-up sample ($$n = 720$$). The PCL-5 was administered to all those who underwent a COVID-19 diagnostic test, all those who knew a person who died from COVID-19 and a randomised selection of $20\%$ of the rest of the sample. Cronbach’s alpha in our sample was 0.96 showing good internal consistency. The distribution of the PCL-5 was skewed being against the null hypothesis that it is normally distributed (median/Q1-Q3) ($\frac{14}{7}$–34). ## Suicidal thoughts and behaviours (STB) (T0 & T1) The STB24 was assessed for ideation, plan or attempt with a single item for any symptom (total 4 items) from the CIDI questionnaire. Suicidal ideation was classified as passive “Did you ever think it would be better if you were dead?” and active ideation “Have you ever thought about killing yourself?”; suicidal plan with “Did you make any plans to harm or kill yourself?”; and suicide attempt with “Did you try to harm yourself or attempt suicide?”. Response items were “Yes/No/I don’t know”. 12 months prevalence was assessed at T0, and since the lockdown started (March 14) at T1. MDE, GAD and STB were recorded as follows: No mental health problem (negative in T0 and T1), Incidence (negative in T0 but positive in T1), Persistence (positive in both assessments T0 and T1) and Recovery (positive in T0 and negative in T1). These variables were considered as a dichotomous (Yes/No). ## Sample size Sample size was estimated based on the incidence data between T0 and T1, assuming an annual baseline depression incidence of $2\%$ and COVID-19 exposure affecting $10\%$ of the population, increasing incidence up to $10\%$. Based on these assumptions, with a statistical power of 0.80 at a 0.05 nominal significance level and considering a $40\%$ loss-to-follow up cases, total sample size at follow-up was estimated in 1200 people. Depression incidence was selected for this purpose because it was expected to be especially high in the pandemic context being a good proxy variable for mental health effects. Furthermore, it has been done to be consistent with the criteria used in the baseline study, ensuring comparability. ## Statistical analysis Statistical analyses involve different methods depending on the use of cross-sectional or longitudinal data. In cross-sectional data, the prevalence and mean (plus standard deviation) or median (plus the interquartile range between quartiles 1 and 3) of socio-demographic characteristics, COVID-19 exposure, and physical and mental health were calculated. Prevalence at TO was conducted with 2005 individuals and T1 with 1357 individuals. Longitudinal data analyses were conducted for assessing trajectories before and during the pandemic in physical health and mental health problems. Longitudinal analyses to assess changes of mental health status during the first year of follow-up were conducted with 941 individuals. Due to its online nature, cross-sectional data contained no missing data other than interview skips by design. For the missing values lost to follow up, we corrected using inverse probability weighting (IPW)25, calculated as the inverse of the logistic propensity of completing the follow-up survey, conditioned on observed related covariates. Population weights were applied to restore sex, geographical and age population distribution. McNemar’s test was used to assess changes in the sample between T0 and T1 in categorical variables; the Student’s paired samples t-test was used in continuous variables for assessing mean differences across time. Physical and mental health problems were assessed for their association with COVID-19 clinical status in 2 groups of the COVID-19 clinical status (positive or those with no test done but COVID-19 symptoms vs. Negative test result or No test done and no COVID-19 symptoms) with Chi-squared (χ2) and Cramer’s V (Vc). Finally, group differences between the level of disability and COVID-19 was assessed with the Student’s parametric t-test and, finally, PTSD symptoms and COVID-19 clinical status was assessed with the U-Mann non-parametric test for independent samples because most of individuals had no or few symptoms of PTSD not supporting visually and statistically the hypothesis of normality (Kolmogorov–Smirnov normality test $p \leq 0.001$). Some independent variables of physical and mental health were collapsed because very few individuals were found in some subgroups. Due to the low number of individuals in some comparisons, a sensitivity analysis was done between physical and mental health outcomes after collapsing for increasing the number of individuals in each group and COVID-19 clinical status (Supplementary Table S2). All statistical tests were conducted with R package ipw26 and SPSS v20.0. Significance level was corrected for multiple testing with False Discovery Rates (FDR) method using the Benjamini–Hochberg adjustments27, with a significance level of $5\%$. ## Data availability The study protocol and individual participant data that underlie the results reported in this article, after de-identification, can be shared with investigators whose proposed use of the data has been approved by the ethic committee of the Universitat Internacional de Catalunya (UIC). Data can be provided for meta-analysis or other projects. Requests should be addressed to the senior author at pcastellvi@uic.es. ## Ethics approval The Ethical Committee of the Universitat Internacional de Catalunya approved the follow-up study. The previous study was approved by the Ethical Committee of the IMIM-Parc de Salut Mar. ## Prevalence of physical and mental health before (T0) and during (T1) the COVID-19 pandemic Attrition analyses identified differences among individuals who responded to the T1 subsample as compared to the baseline sample (T0) in gender and age range, but not in the Spanish regions (see Supplementary Table S1). The follow-up sample had a lower percentage of men (T0, $51.1\%$ vs. T1, $44\%$), and a higher percentage of older people (> 65: T0, $14.2\%$ vs. T1, $20.8\%$) than the baseline sample. Baseline characteristics of the whole T0 sample and the follow-up T1 subsample are shown in Table 1. Table 1 summarises the weighted characteristics of the sample, $48.5\%$ were men, $53.8\%$ of the sample had an age range of 40–65 years, $31.3\%$ were single and $56.6\%$ were married, more than half of the sample were employed ($54.6\%$) and $20.4\%$ were retired. Table 1Comparison of sample characteristics at baseline and 12-month follow-up samples after weighting. Baseline $$n = 200512$$-month follow-up $$n = 1357$$N%N%pSocio-demographic characteristicsGender, (men)96948.5Age, (years)18–251256.3 > 25–4044422.2 > 40–65107453.8 > 6535517.8Marital statusSingle62531.3Married/Couple113056.6Separated231.1Divorced1206.0Widowed1005.0Employment statusEmployed109054.6Off sick301.5Unemployed1587.9Homemaker1397.0Student713.6Student & employed572.8Temporal or permanent disability442.2Retired40820.4Physical healthSelf-perception0.570Excellent1005.0644.7Very good42321.228120.8Good112756.474555.2Fair30915.423417.3Poor392.0262.0Current general health self-perception than0.002before the pandemic814.0302.2Much better25312.7574.2Somewhat better140770.595270.5Same23511.828321.0Somewhat worse211.1282.1Much worseWeight (kg) (mean/SD)73.2215.5671.2111.940.002BMI0.589Underweight371.9332.4Normal92046.259043.7Overweight72836.552538.9Obese30715.519514.4Smoking0.002Non-smoker121160.691767.9Former smoker43321.725118.6Occasional1266.3614.5Current22711.41219.0Disability (mean/SD)21.529.2219.037.320.002Mental healthSelf-perception0.008Excellent31015.515911.8Very good66833.543232.0Good88744.461645.6Fair1236.21299.5Poor100.5141.0Major depressive episode (yes)1316.51198.80.036Generalized anxiety disorder (yes)9913.77517.70.002Posttraumatic stress disorder symptoms (median/ Q1-Q3)147–34Thoughts of death0.079Yes48024.037828.0I don’t know643.2342.5Any suicidal thoughts and behaviors29415.19.47.10.002Suicidal ideation (passive)0.002Yes27613.810411.3I don’t know502.5161.8Suicidal ideation (active)0.002Yes874.4312.3I don’t know341.770.5Suicidal plan0.081Yes412.190.7I don’t know110.660.4Suicide attempt0.129Yes261.350.4I don’t know70.310.1Categorical variables were assessed with McNemnar’s test, and continuous variables with paired t-test. p-values were adjusted by multiple comparison with False Discovery Rates (FDR). Statistically significant differences between T0 and T1 were conducted only. Kg, Kilograms; Q1, First quartile; Q3, Third quartile; SD, Standard deviation$.\%$ weighted follow-up sample weight (inverse probability weighting and post-stratification).Significant values are in bold. Overall, there were no statistically significant differences in the prevalence of self-assessment of physical health during the pandemic compared with before ($$p \leq 0.532$$), although more people considered their general health was somewhat or much worse than somewhat or much better during the pandemic, when compared with before the pandemic (somewhat or much worse, $23.1\%$ vs. somewhat or much better, $6.6\%$; $$p \leq 0.002$$). Additionally, the prevalence of current smokers was statistically significantly lower (T0, $11.4\%$ vs. T1, $9.0\%$; $$p \leq 0.002$$); the population had statistically significant lower weight (mean/SD) (T0, $\frac{73.22}{15.56}$ vs. T1, $\frac{71.21}{11.94}$; $$p \leq 0.002$$); and the WHODAS indicated there was a statistically significant decrease in disability (mean/SD) (T0, $\frac{21.52}{9.22}$ vs. T1, $\frac{19.03}{7.32}$; $$p \leq 0.002$$) during the pandemic. Regarding self-assessed mental health, a higher prevalence of fair or poor self-assessment was observed during the pandemic than before (T0, $6.7\%$ vs. T1, $10.5\%$; $$p \leq 0.002$$). For mental disorders, there were statistically significant differences in the prevalence of MDE (T0, $6.5\%$ vs. T1, $8.8\%$; $$p \leq 0.036$$) and GAD (T0, $13.7\%$ vs. $17.7\%$; $$p \leq 0.002$$). Finally, we found that the prevalence of any STB (T0, $15.1\%$ vs. T1, $7.1\%$; $$p \leq 0.002$$), and passive (T0, $13.8\%$ vs. T1, $11.3\%$; $$p \leq 0.002$$) and active suicidal ideation (T0, $4.4\%$ vs. T1, $2.3\%$; $$p \leq 0.002$$) were statistically significantly decreased during the pandemic. We also assessed specific variables related to the COVID-19 pandemic. Results showed that 514 ($38.1\%$) of the total sample received a diagnostic test for COVID-19 and $3.4\%$ reported symptoms related to COVID-19 but they did not undergo any diagnostic test. Of the 514 participants who had a COVID-19 diagnostic test, 48 ($9.3\%$) were positive, which represents $3.6\%$ of the total sample. The most prevalent COVID-19 symptoms were headache ($18.8\%$), cough ($16.3\%$), muscle aches ($15.1\%$), fever ($15.0\%$), feeling very tired ($13.7\%$) and diarrhoea ($10.1\%$). Taking into account the number of people known by participants to be infected, the observed median (Q1-Q3) was 22(16–33). For people known by participants to have died, the observed median (Q1-Q3) was 6(5–8). Individuals reported different reasons for being worried: (i) a lot or very much about the increase in social distancing ($19.5\%$); (ii) difficulties getting the help needed for their loved ones ($19.8\%$); and (iii) the probability of their loved ones becoming infected ($27.5\%$) during the pandemic (Table 2).Table 2Sample characteristics during the COVID-19 pandemic (at 12-month follow-up) of factors associated and stress-related with COVID-19 after weighting.12-Month follow-up $$n = 1357$$N%Factors associated with COVID-19 COVID-19 clinical status Positive test result483.6 Negative test result46634.5 COVID-19 symptoms without test done453.4 No test done and no symptoms79158.5COVID-19 symptoms Cough9216.3 Fever8515.0 Difficulty breathing or shortness of breath478.3 Sore throat when you drink any liquid305.2 Loss of smell539.2 Loss of taste478.1 Muscle aches8615.1 Diarrhoea5610.1 Chest pain254.4 Headache10718.8 Coughing up blood00.0 Vomiting162.8 Feeling confused132.3 Feeling drowsy234.2 Feeling very tired7813.7 Had other related symptoms183.2 Number of known infected people (median/Q1–Q3)2216–33 Number of known death people by COVID-19 (median/Q1–Q3)65–8COVID-19 stress-related Financial problems Not at all67550.0 A bit37627.8 Quite17513.0 A lot of675.0 Very much574.2Increase of social isolationNot at all25819.1A bit47235.0Quite35726.4A lot of16212.0Very much1017.5Difficulties to get the needed help to our loved ones Not at all26719.8 A bit46934.7 Quite34725.7 A lot of15711.6 Very much1118.2Have increased arguments with our family and friendsNot at all72253.5A bit38428.4Quite15711.6A lot of594.4Very much292.2The probability to get infected Not at all36927.3 A bit50337.3 Quite26019.3 A lot of1299.6 Very much896.6The probability about loved ones getting infected Not at all17312.8 A bit41330.6 Quite39429.2 A lot of20815.4 Very much16312.1Q1, First quartile; Q3, Third quartile$.\%$ weighted follow-up sample weight (inverse probability weighting and post-stratification). ## Trajectories of health problems during the COVID-19 pandemic We found that only $0.5\%$ (weighted $$n = 11$$) of non-smokers at baseline started smoking at follow-up and $1.4\%$ (weighted $$n = 30$$) of smokers at baseline quit smoking. When we assessed mental health, the highest incidence in our sample was in GAD ($20.6\%$); MDE was less ($5.6\%$) and for any STB was $2.1\%$. Persistence was also highest in GAD ($6.8\%$), followed by any STB ($4.2\%$) then MDE ($2.3\%$). Finally, the percentage of individuals who recovered from a baseline mental health problem was the highest for any STB ($8.1\%$), followed by GAD ($6.2\%$) and the lowest for MDE ($3.9\%$) (Fig. 1).Figure 1Percentage of sample with No (negative at both assessments T0 and T1), Incidence (negative at T0 but positive at T1), Persistence (positive at both assessments T0 and T1) and Recovery (positive at T0 and negative at T1) of MDE, GAD and any STB before and during the COVID-19. GAD Generalized anxiety disorder; MDE Major depressive episode; STB Suicidal thoughts and behaviors. % weighted follow-up sample weight (inverse probability weighting and post-stratification). Statistical analyses were conducted with 941 individuals. ## Health impact according to COVID-19 clinical status The physical and mental health was compared among individuals according to COVID-19 clinical status: those with COVID-19 symptoms with no test done or those with a positive COVID-19 test result (Group 1); and individuals with no test done and no COVID-19 symptoms or those with a negative COVID-19 test result (Group 2); Results showed that Group 1 reported worse both physical (χ2 = 7.41; Vc = 0.074; $$p \leq 0.025$$) and mental (χ2 = 8.00; Vc = = 0.077; $$p \leq 0.024$$) health than before the pandemic; a worse health self-perception than 1 year ago (χ2 = 22.95; Vc = 0.077; $$p \leq 0.002$$); and increased disability (mean/SD) (Group 1, $\frac{8.17}{8.5}$; Group $\frac{217.30}{6.5}$; $$p \leq 0.002$$) than Group 2. For mental health, there were statistically significant differences in MDE (χ2 = 26.24; Vc = 0.143; $$p \leq 0.002$$), GAD (χ2 = 13.23; Vc = 0.219; $$p \leq 0.006$$) and STB (χ2 = 29.05; Vc = 0.128; $$p \leq 0.002$$). Specifically, new cases (Incidence), those positive in both assessments (Persistence) and those positive but negative during pandemic (Recovery) of MDE, GAD and STB during the COVID-19 pandemic were higher in those individuals with a positive test result or COVID-19 symptoms but no test done (Group 1) comparing with those with a negative test result or with no COVID-19 symptoms and no test done (Group 2) (Table 3). Finally, we assessed the symptomatology of PTSD according to COVID-19 status. A non-parametric U-Mann test analysis showed there were statistically significant differences between groups ($$p \leq 0.025$$). Group 1 had statistically higher PTSD symptoms than Group 2 (median/Q1-Q3) (Group 1, $\frac{13}{5}$–34; Group 2, $\frac{8}{3}$–20).Table 3Association between COVID-19 clinical status (positive or those with no test done but COVID-19-related symptoms vs. Negative test result or No test done and no COVID-19 symptoms) and physical and mental health. COVID-19 clinical statusPositive test result/no test done but COVID-19 symptomsNegative test result/no test done and no COVID-19 symptomsχ2pN%N%Physical health Self-perception7.410.025 Excellent/very good1516.133026.3 Good2555.969355.1 Fair/poor2628.023318.6Current general health self-perception 1 year ago22.950.002Much/Somewhat better66.5816.4Same4850.590572.0Much/Somewhat worse4143.027121.6Disability (mean/SD)20.178.517.306.50.002Mental health Self-perception8.000.024 Excellent/very good3941.555143.9 Good3739.457946.1 Fair/poor1819.11259.9Major depressive episode26.240.002No8273.5155589.2Incidence1614.2885.0Persistence54.4382.2Recovery98.0633.6Generalized anxiety disorder13.230.006No1341.919668.9Incidence725.85820.1Persistence516.1165.7Recovery516.1155.3Any suicidal thoughts and behaviors29.050.002No7067.6145186.7Incidence54.8221.9Persistence109.5653.9Recovery1918.11267.5Posttraumatic stress disorder symptoms (median/ Q1–Q3)135–3483–200.025Q1, First quartile; Q3, Third quartile$.\%$ weighted follow-up sample weight (inverse probability weighting and post-stratification).Student’s t-test for independent samples; ** U-Mann non-parametric test for independent samples. Significant values are in bold. ## Sensitivity analyses Sensitivity analyses conducted after collapsing physical and mental health outcomes and COVID-19 symptoms, significance was maintained in all statistical analyses performed (Supplementary Table S2) suggesting that our results are consistent across groups. ## Main results This study explored the impact in adults over the nine months following the start of the first lockdown response to the COVID-19 pandemic in Spain. Results show there was a substantial impact on physical and mental health in the Spanish population. Specifically, in physical health, individuals reported they had lost weight, but there were no qualitatively substantial changes in BMI; fewer occasional and current smokers; but not a worsening in disability. As for mental health, there was a worsening in mental health self-assessment; a statistically significant higher prevalence of MDE and GAD, but a lower STB. The general Spanish population was mostly affected by GAD, with 1 out of 5 people defining as an incident case, and $6.8\%$ persisting in GAD in both assessments. For MDE, our study population had a MDE incidence rate of $\frac{5.3}{100}$ and $3.9\%$ showed MDE before and during the pandemic. The highest percentage of individuals recovering from all mental health problems was for STB, where the prevalence was lower during the pandemic than before. When we compared mental health status according to COVID-19 symptoms and diagnosis, individuals who had been diagnosed with COVID-19 or who had compatible symptoms had worse self-assessment of their physical and mental health and more disability than before the pandemic. For mental health, individuals who had been diagnosed with COVID-19 had more incidence, persistence and recovery of MDE, GAD and STB. In PTSD, we found that those with greater symptoms were those with related COVID-19 symptoms but who had had no test done and for those with a diagnosis of COVID-19. ## Strengths and limitations The results of this study have to be seen in the light of its limitations. First, the attrition rate ($53.1\%$) at follow-up might have biased comparisons between our baseline results of the total Spanish population. We addressed these limitations by adding an extended additional sample of individuals with similar characteristics to those who did not respond at follow-up. Furthermore, we applied population-based adjustments and inverse-probability weighting to correct for bias in the comparisons, which proved to be an effective method for reducing bias from a lack of response28,29. Secondly, the assessment of mental disorders and suicide risk was based on self-reports and not on direct clinical assessment. Therefore, they would be better considered as “probable cases” of disorder. Nevertheless, good diagnostic agreement was reported with the clinical judgment of the CIDI instrument, which includes an assessment of MDE and STB in our study, with face-to-face20,30, and online assessments31 in the Spanish population. However, although the GAD-7 and PCL-5 are well-validated scales21,23, calibration studies have not been carried out on general population samples. Furthermore, some scales were developed ad hoc for this study, such as the COVID-19 survey, so their diagnostic properties for detecting cases are not available. The urgency of the pandemic situation and the necessity of developing our study to add new knowledge about the impact of the pandemic and the social restrictions during the first year of pandemic motivated us to develop these scales without studying their validity. Thirdly, due to the infrequency of some variables, we had to combine information to avoid numerical problems in statistical analyses and we did sensitivity analyses with collapsed variables for measuring consistency. So, it was not possible to analyse them separately, due to the low frequency of these outcomes. Fourthly, we assessed PTSD symptoms instead of PTSD disorder. So, we do not have the mechanisms for diagnosis, and it is not possible to estimate the prevalence of PTSD in our population. However, although the DSM-5 definition notes that a life-threatening illness or debilitating medical condition is not necessarily a traumatic event. Since the first case of contagion appeared, more than 6 million people died by COVID-19 according to WHO and much more people had sequelae after the infection32. Therefore, although many people were not mental health affected by the consequences of COVID-19 pandemic and contagion, it is true that a subgroup of our sample have suffered some type of trauma related, especially in high-risk groups. Nevertheless, the study has a number of strengths. First, the prospective design, with an assessment before and during the first months of the COVID-19 pandemic over a wide range of health outcomes, especially in mental health, provides a more comprehensive view of the impact of the pandemic on the health of the general Spanish population, and in people diagnosed with, or showing symptoms of, COVID-19. The longitudinal design allowed us to assess the trajectories (number, incidence, persistence, recovery) of many variables of interest. Second, we analysed two mental disorders (MDE, GAD), symptoms of PTSD and STB using adapted and validated instruments. Finally, the online methodology used tends to deliver more reliable information about sensitive information, such as suicide risk, than face-to-face assessments33. ## Comparison with other studies Results showed there were an impact on physical health, with BMI having reduced significantly; although, the difference was only 2 kg. Thus, we consider this change as not clinically relevant. Furthermore, there were no qualitative changes in BMI which is in line with this hypothesis. More important, is that the general health self-assessment was worsened during the pandemic. The lockdown and further restrictions in the first year are having an impact on the general population with an increasing in Disability-Adjusted Life Years34; and many of those experiencing disability did not receive care, due to the closure of outpatient services or an increase in waiting lists35. Our study shows a significant increase in the prevalence of mental health disorders during the pandemic in the Spanish population, from 6.5 to $8.8\%$ in MDE, and from 13.7 to $17.7\%$ in GAD. In two recent meta-analyses of the mental impact of COVID-19 worldwide, both showed a higher prevalence of depression ($26\%$ and $16\%$, respectively) than our study; a similar prevalence of anxiety ($15\%$) was seen in only one meta-analysis, while another meta-analysis showed a higher prevalence of anxiety ($32\%$)36,37. These disparities suggest that the results should be interpreted with caution, because much heterogeneity exists between studies. We used online diagnostic instruments, but most of the studies included in the meta-analyses used scales which pooled results may have overestimated. Also, most of the samples were from China, so these results may not extend well to the Spanish or European population; also, the impact of the pandemic in the population may be different for studies occurring during and after the lockdown. In Spain, two previous population-based, cross-sectional studies38,39 showed that the prevalence of depressive symptoms was $19\%$ and $24\%$, and anxiety symptoms was $22\%$ and $26\%$ during the first wave. While these values are higher than our findings, it must be noted that those studies did not use diagnostic-oriented tools, but a screening instrument assessment. Longitudinal studies with assessments before and during the pandemic showed an increase in depressive symptoms in adolescents from Iceland40. In adults, the British population had an increase in depressive and anxiety symptoms in the early stages of lockdown, which declined fairly rapidly; possibly because individuals adapted to the circumstances41. In The Netherlands, only individuals without previous mental disorders showed an increase in depressive and anxiety symptoms, but not individuals with no previous mental disorders11. Finally, a Spanish cohort sample assessed the prevalence of MDE in the first wave using CIDI, which was the same diagnostic instrument as our study but contacts were made by telephone and an abbreviated version was used; this increased from 7.8 to $9.8\%$ in May/June 2020, but was not statistically significant13. When comparing results, our study shows that that the prevalence of MDE was significantly higher. This result suggests that the impact of the pandemic on mood increased over time more than just during the lockdown. We also assessed the impact of the pandemic for STB in the general population and the results showed a significant decrease in STBs from 15.1 to $7.1\%$. These results are in line with previous studies. The evidence showed a decrease in suicide rates compared with the expected number in 12 countries from 21, and no country showed a significant increase in suicides. Specifically in Spain, the rates of suicides have reduced by $23\%$ compared to before the pandemic42. To the best of our best knowledge, two population-based cross-sectional studies assessed STBs in Spaniards. In one study, the 30-day prevalence of STB was $4.5\%$43, lower than in our study. However, we assessed the presence of STBs from when the lockdown started (9 months later). Another study assessed passive suicide ideation in March 2020, just at the beginning of the lockdown. Results showed a prevalence of $8.8\%$ for passive suicide ideation, which was lower than in our study ($11.3\%$)44. In the aforementioned Spanish longitudinal study13, the prevalence of suicidal ideation was quite similar ($2.2\%$ vs. $2.1\%$). So, although some risk factors are increasing (e.g., MDE and GAD), the prevalence of STB was decreasing during the first year of the pandemic. The lack of increase in suicides and STB since the pandemic started could be attributed to the presence of protective factors or attrition rates in this specific subgroup. Communities might have actively tried to support at-risk individuals, people might have connected in new ways and some relationships might have been strengthened by households spending more time with each other45. For some people, the collective feeling of “we’re all in this together” might have been beneficial42. Further research should assess whether or not there is an increase in STB and suicide rates in the population in the long term, as the exposure of some risk factors for suicide are increasing. Finally, we assessed mental disorder trajectories and COVID-19 status. Those individuals diagnosed with COVID-19 or with compatible symptoms were the most affected mentally by MDE, GAD, STB and PTSD symptoms. These are in line with previous studies, where the prevalence of depression36, anxiety36, PTSD46 and suicidal ideation47 were high ($55\%$ for depression; $56\%$ anxiety; $28\%$ PTSD, and $12\%$ suicidal ideation) and substantially higher than in the general population. The psychiatric consequences of SARS-CoV-2 infection can be caused both by the immune response to coronaviruses, which induces local and systemic production of cytokines, chemokines and other inflammatory mediators48, or by psychological stressors, such as social isolation, the psychological impact of a severe and potentially fatal novel illness, concerns about infecting others and stigma46. As a conclusion, although the initial effect of the pandemic in its first year has been moderate regarding physical and mental health, many risk factors have increased. Incidence, persistence and recovery of MDE, GAD and STB; the presence of PTSD symptoms in those diagnosed with COVID-19 or with compatible symptoms; and a worsening in self-assessed health status in the general population are reasons for concern. Many of these increases are regarded as known suicide risk factors. So, we expect a constant increase in mental disorders and STB in our population. This suggests the development is needed of a broad, population-based prevention approach to help people cope with the consequences of the pandemic. Such an approach should be all-encompassing, including financial measures, while also reducing the physical and mental health impact of COVID-19. Future research should gather information about the long-term impact of the pandemic beyond its initial impact, and the trajectories of some vulnerable groups, such as those with previous psychiatric disorders or those with socio-economic difficulties. Additionally, it would be useful to get the exact timing of the onset or recovery from each mental disorder or STB. ## Supplementary Information Supplementary Information. The online version contains supplementary material available at 10.1038/s41598-023-28336-2. ## References 1. Pierce M. Mental health before and during the COVID-19 pandemic: A longitudinal probability sample survey of the UK population. Lancet Psychiatry (2020.0) 7 883-892. DOI: 10.1016/S2215-0366(20)30308-4 2. Oliver N, Barber X, Roomp K, Roomp K. Assessing the impact of the COVID-19 pandemic in Spain: Large-scale, online, self-reported population survey. J. Med. Internet Res. (2020.0) 22 e21319. DOI: 10.2196/21319 3. 3.Joyce, R. & Xu, X. Sector shutdowns during the coronavirus crisis: Which workers are most exposed? Institute for Fisical Studies Briefing note Preprint at (2020). 4. 4.Office for National Statistics. Coronavirus and the social impacts on Great Britain. (2020). 5. Rubin GJ, Wessely S. The psychological effects of quarantining a city. BMJ (2020.0) 368 m313. DOI: 10.1136/bmj.m313 6. Duan L, Zhu G. Psychological interventions for people affected by the COVID-19 epidemic. Lancet Psychiatry (2020.0) 7 300-302. DOI: 10.1016/S2215-0366(20)30073-0 7. Louie JK. A novel risk factor for a novel virus: Obesity and 2009 pandemic influenza A (H1N1). Clin. Infect. Dis. (2011.0) 52 301-312. DOI: 10.1093/cid/ciq152 8. Morina N, Kip A, Hoppen TH, Priebe S, Meyer T. Potential impact of physical distancing on physical and mental health: A rapid narrative umbrella review of meta-analyses on the link between social connection and health. BMJ Open (2021.0) 11 e042335. DOI: 10.1136/bmjopen-2020-042335 9. Zakka K. SARS-CoV-2 and obesity: ‘CoVesity’-a pandemic within a pandemic. Obes. Surg. (2021.0) 31 1745-1754. DOI: 10.1007/s11695-020-04919-0 10. Sønderskov K, Dinesen P, Santini Z, Østergaard S. The depressive state of Denmark during the COVID-19 pandemic. Acta Neuropsychiatr. (2020.0) 32 226-228. DOI: 10.1017/neu.2020.15 11. Pan K-Y. The mental health impact of the COVID-19 pandemic on people with and without depressive, anxiety, or obsessive-compulsive disorders: A longitudinal study of three Dutch case-control cohorts. Lancet Psychiatry (2021.0) 8 121-129. DOI: 10.1016/S2215-0366(20)30491-0 12. Rodríguez-Rey R, Garrido-Hernansaiz H, Collado S. Psychological impact and associated factors during the initial stage of the Coronavirus (COVID-19) pandemic among the general population in Spain. Front. Psychol. (2020.0) 11 1540. DOI: 10.3389/fpsyg.2020.01540 13. Ayuso-Mateos JL. Changes in depression and suicidal ideation under severe lockdown restrictions during the first wave of the COVID-19 pandemic in Spain: A longitudinal study in the general population. Epidemiol. Psychiatr. Sci. (2021.0) 30 e49. DOI: 10.1017/S2045796021000408 14. Federici S, Bracalenti M, Meloni F, Luciano JV. World Health Organization disability assessment schedule 2.0: An international systematic review. Disabil. Rehabil. (2017.0) 39 2347-2380. DOI: 10.1080/09638288.2016.1223177 15. Vázquez-Barquero JL. Spanish version of the new World Health Organization Disability Assessment Schedule II (WHO-DAS-II): Initial phase of development and pilot study. Cantabria disability work group. Actas Esp. Psiquiatr. (2000.0) 28 77-87. PMID: 10937388 16. Ware J, Sherbourne C. The MOS 36-item short-form health survey (SF-36). I. Conceptual framework and item selection. Med. Care (1992.0) 30 473-483. DOI: 10.1097/00005650-199206000-00002 17. Vilagut G. El Cuestionario de Salud SF-36 español: una década de experiencia y nuevos desarrollos por los investigadores de la Red-IRYSS. Gac. Sanit.* (2005.0) 19 135-150. DOI: 10.1157/13074369 18. Kessler RC, Üstün BB. The world mental health (WMH) survey initiative version of the world health organization (WHO) composite international diagnostic interview (CIDI). Int. J. Methods Psychiatr. Res. (2004.0) 13 93-117. DOI: 10.1002/mpr.168 19. Luciano JV. Utility of the twelve-item world health organization disability assessment schedule II (WHO-DAS II) for discriminating depression ‘caseness’ and severity in Spanish primary care patients. Qual Life Res (2010.0) 19 97-101. DOI: 10.1007/s11136-009-9566-z 20. Haro JM. Concordance of the composite international diagnostic interview version 3.0 (CIDI 3.0) with standardized clinical assessments in the WHO world mental health surveys. Int. J. Methods Psychiatr. Res. (2006.0) 15 167-180. DOI: 10.1002/mpr.196 21. García-Campayo J. Cultural adaptation into Spanish of the generalized anxiety disorder-7 (GAD-7) scale as a screening tool. Health Qual. Life Outcomes (2010.0) 8 8. DOI: 10.1186/1477-7525-8-8 22. Roberts NP, Kitchiner NJ, Lewis CE, Downes AJ, Bisson JI. Psychometric properties of the PTSD Checklist for DSM-5 in a sample of trauma exposed mental health service users. Eur. J. Psychotraumatol. (2021.0) 12 1863578. DOI: 10.1080/20008198.2020.1863578 23. Orlando M, Marshall GN. Differential item functioning in a Spanish translation of the PTSD checklist: Detection and evaluation of impact. Psychol. Assess (2002.0) 14 50-59. DOI: 10.1037/1040-3590.14.1.50 24. Nock MK. Self-Injury. Annu. Rev. Clin. Psychol. (2010.0) 6 339-363. DOI: 10.1146/annurev.clinpsy.121208.131258 25. Seaman SR, White IR. Review of inverse probability weighting for dealing with missing data. Stat. Methods Med. Res. (2013.0) 22 278-295. DOI: 10.1177/0962280210395740 26. van der Wal WM, Geskus RB. Ipw: An R package for inverse probability weighting. J. Stat. Softw. (2011.0) 43 2-23. DOI: 10.18637/jss.v043.i13 27. Benjamini Y, Hochberg Y. Controlling the false discovery rate: A practical and powerful approach to multiple testing. J. Roy. Stat. Soc.: Ser. B (Methodol.) (1995.0) 57 289-300 28. Brick JM. Unit nonresponse and weighting adjustments: A critical review. J. Off. Stat. (2013.0) 29 329-353. DOI: 10.2478/jos-2013-0026 29. Dey EL. Working with low survey response rates: The efficacy of weighting adjustments. Res. High. Educ. (1997.0) 38 215-227. DOI: 10.1023/A:1024985704202 30. Wittchen HU. Reliability and validity studies of the WHO–Composite International Diagnostic Interview (CIDI): a critical review. J Psychiatr Res (1994.0) 28 57-84. DOI: 10.1016/0022-3956(94)90036-1 31. Ballester L. Accuracy of online survey assessment of mental disorders and suicidal thoughts and behaviors in Spanish university students. Results of the WHO World Mental Health-International College Student initiative. PLoS ONE (2019.0) 14 e0221529. DOI: 10.1371/journal.pone.0221529 32. 32.WHO. WHO Coronavirus (COVID-19) Dashboard. (2022). 33. Tourangeau R, Yan T. Sensitive questions in surveys. Psychol Bull (2007.0) 133 859-883. DOI: 10.1037/0033-2909.133.5.859 34. Nurchis MC. Impact of the burden of COVID-19 in Italy: Results of disability-adjusted life years (DALYs) and productivity loss. Int. J. Environ. Res. Public Health (2020.0) 17 4233. DOI: 10.3390/ijerph17124233 35. Negrini S. Up to 2.2 million people experiencing disability suffer collateral damage each day of COVID-19 lockdown in Europe. Eur. J. Phys. Rehabil. Med. (2020.0) 56 361-365. DOI: 10.23736/S1973-9087.20.06361-3 36. Luo M, Guo L, Yu M, Jiang W, Wang H. The psychological and mental impact of coronavirus disease 2019 (COVID-19) on medical staff and general public—A systematic review and meta-analysis. Psychiatry Res. (2020.0) 291 113190. DOI: 10.1016/j.psychres.2020.113190 37. Cénat JM. Prevalence of symptoms of depression, anxiety, insomnia, posttraumatic stress disorder, and psychological distress among populations affected by the COVID-19 pandemic: A systematic review and meta-analysis. Psychiatry Res. (2021.0) 295 113599. DOI: 10.1016/j.psychres.2020.113599 38. Ozamiz-Etxebarria N, Dosil-Santamaria M, Picaza-Gorrochategui M, Idoiaga-Mondragon N. Stress, anxiety, and depression levels in the initial stage of the COVID-19 outbreak in a population sample in the northern Spain. Cad Saude Publica (2020.0) 36 e00054020. DOI: 10.1590/0102-311x00054020 39. González-Sanguino C. Mental health consequences during the initial stage of the 2020 Coronavirus pandemic (COVID-19) in Spain. Brain Behav. Immun. (2020.0) 87 172-176. DOI: 10.1016/j.bbi.2020.05.040 40. Thorisdottir IE. Depressive symptoms, mental wellbeing, and substance use among adolescents before and during the COVID-19 pandemic in Iceland: A longitudinal, population-based study. Lancet Psychiatry (2021.0). DOI: 10.1016/S2215-0366(21)00156-5 41. Fancourt D, Steptoe A, Bu F. Trajectories of anxiety and depressive symptoms during enforced isolation due to COVID-19 in England: A longitudinal observational study. Lancet Psychiatry (2021.0) 8 141-149. DOI: 10.1016/S2215-0366(20)30482-X 42. Pirkis J. Suicide trends in the early months of the COVID-19 pandemic: An interrupted time-series analysis of preliminary data from 21 countries. Lancet Psychiatry (2021.0) 8 579-588. DOI: 10.1016/S2215-0366(21)00091-2 43. Mortier P. Thirty-day suicidal thoughts and behaviours in the Spanish adult general population during the first wave of the Spain COVID-19 pandemic. Epidemiol. Psychiatr. Sci. (2021.0) 30 e19. DOI: 10.1017/S2045796021000093 44. Sáiz PA. Prevalence of passive suicidal ideation in the early stage of the Coronavirus disease 2019 (COVID-19) pandemic and lockdown in a large Spanish sample. J. Clin. Psychiatry (2020.0) 81 13899. DOI: 10.4088/JCP.20l13421 45. Wasserman D, Iosue M, Wuestefeld A, Carli V. Adaptation of evidence-based suicide prevention strategies during and after the COVID-19 pandemic. World Psychiatry (2020.0) 19 294-306. DOI: 10.1002/wps.20801 46. Mazza MG. Anxiety and depression in COVID-19 survivors: Role of inflammatory and clinical predictors. Brain Behav. Immun. (2020.0) 89 594-600. DOI: 10.1016/j.bbi.2020.07.037 47. Na P, Tsai J, Harpaz-Rotem I, Pietrzak R. Mental health and suicidal ideation in US military veterans with histories of COVID-19 infection. BMJ Mil Health (2021.0). DOI: 10.1136/bmjmilitary-2021-001846 48. Cameron MJ, Bermejo-Martin JF, Danesh A, Muller MP, Kelvin DJ. Human immunopathogenesis of severe acute respiratory syndrome (SARS). Virus Res. (2008.0) 133 13-19. DOI: 10.1016/j.virusres.2007.02.014
--- title: PRDM10 directs FLCN expression in a novel disorder overlapping with Birt–Hogg–Dubé syndrome and familial lipomatosis authors: - Irma van de Beek - Iris E Glykofridis - Jan C Oosterwijk - Peter C van den Akker - Gilles F H Diercks - Maria C Bolling - Quinten Waisfisz - Arjen R Mensenkamp - Jesper A Balk - Rob Zwart - Alex V Postma - Hanne E J Meijers-Heijboer - R Jeroen A van Moorselaar - Rob M F Wolthuis - Arjan C Houweling journal: Human Molecular Genetics year: 2022 pmcid: PMC10026250 doi: 10.1093/hmg/ddac288 license: CC BY 4.0 --- # PRDM10 directs FLCN expression in a novel disorder overlapping with Birt–Hogg–Dubé syndrome and familial lipomatosis ## Abstract Birt–Hogg–Dubé syndrome (BHD) is an autosomal dominant disorder characterized by fibrofolliculomas, pulmonary cysts, pneumothoraces and renal cell carcinomas. Here, we reveal a novel hereditary disorder in a family with skin and mucosal lesions, extensive lipomatosis and renal cell carcinomas. The proband was initially diagnosed with BHD based on the presence of fibrofolliculomas, but no pathogenic germline variant was detected in FLCN, the gene associated with BHD. By whole exome sequencing we identified a heterozygous missense variant (p.(Cys677Tyr)) in a zinc-finger encoding domain of the PRDM10 gene which co-segregated with the phenotype in the family. We show that PRDM10Cys677Tyr loses affinity for a regulatory binding motif in the FLCN promoter, abrogating cellular FLCN mRNA and protein levels. Overexpressing inducible PRDM10Cys677Tyr in renal epithelial cells altered the transcription of multiple genes, showing overlap but also differences with the effects of knocking out FLCN. We propose that PRDM10 controls an extensive gene program and acts as a critical regulator of FLCN gene transcription in human cells. The germline variant PRDM10Cys677Tyr curtails cellular folliculin expression and underlies a distinguishable syndrome characterized by extensive lipomatosis, fibrofolliculomas and renal cell carcinomas. ## Introduction Birt–Hogg–Dubé syndrome (BHD, MIM #135150) is an autosomal dominant inherited disorder characterized by fibrofolliculomas (FF), pulmonary cysts, pneumothoraces and renal cell carcinoma (RCC) (1–3). BHD is caused by loss-of-function pathogenic germline variants (PGV) in the FLCN gene, encoding the folliculin protein [4]. As BHD-associated RCCs arise upon loss of the wild-type (WT) FLCN allele, usually due to a somatic second hit, folliculin functions as a tumor suppressor in the kidney [5]. Epigenetic silencing of the WT FLCN allele in RCC has also been reported, but the factors critically controlling FLCN gene transcription are not known [6]. Folliculin is amongst others involved in repressing TFEB/TFE3 transcription factor activity (7–10) and a non-canonical interferon response [11]. Indeed, biallelic FLCN inactivation directs RCC formation in transgenic mice in a manner dependent on TFEB function [7]. However, the exact roles of folliculin and the mechanisms by which reduced cellular folliculin levels cause the tissue-specific clinical features observed in BHD are not understood. BHD can be diagnosed by a PGV in the FLCN gene, by the presence of at least five (histologically confirmed) FF, or by a combination of minor criteria including pulmonary cysts and RCC [12]. BHD should be considered in patients with bilateral or multifocal RCC, chromophobe or hybrid oncocytic kidney tumors, occurrence of RCC before 50 years of age, bilateral basal pulmonary cysts and recurrent pneumothorax, especially when a combination of these has occurred in one family [13]. A PGV in FLCN is found in the vast majority (84–$91\%$) of patients with the clinical diagnosis of BHD (14–16). The remaining BHD patients might have an unidentifiable variant in FLCN, for example in a deep intronic region, or a PGV in a yet unknown gene. Here, we present clinical and molecular data from a large family in which the proband was clinically diagnosed with BHD based on the presence of FF, but without a PGV in the FLCN gene. The phenotype was also distinct from BHD with more skin and mucosal lesions, extensive lipomatosis, and the absence of pulmonary manifestations. By whole exome sequencing (WES), we identified a heterozygous missense variant in PRDM10 which co-segregated with the phenotype. We show that PRDM10 acts as a critical transcriptional regulator of FLCN expression in human cells. ## Results An overview of the extended pedigree of the family is shown in Figure 1A. Informed consent for publication of clinical data, photographs (if applicable) and use of tissue was obtained from patients III-2, III-4, IV-1, IV-2, IV-3 and IV-4. Figure 1: Pedigree and photos of skin and mucosal phenotype. (A) Pedigree of the family. Pr; prostate cancer, uRCC; unclassified renal cell carcinoma, B/M; bilateral and multifocal, ccRCC; clear cell renal cell carcinoma, Pap2 RCC; papillary type 2 renal cell carcinoma, Th folll; follicular thyroid carcinoma, Lu; lung carcinoma, NHL; non-Hodgkin lymphoma. The proband is indicated by an arrow. (B) Photos of affected patients. I (patient IV-1): partly confluent papules on the flank, skin tags in the armpit, intraoral papules, papules on the lip and multiple skin-colored papules on the trunk. II (patient IV-4): papules in the neck. III (patient IV-3): papules on the nipple, intraoral papules. IV (patient III-2): partly confluent papules on the flank and intraoral papules. ## Clinical description The proband (IV-1) was first evaluated at the age of 33 years because of skin and mucosal lesions and multiple lipomas of the trunk. The skin lesions consisted of multiple skin-colored papules in the face and on the trunk, skin tags on the trunk and an area on the right flank with multiple, small, partly confluent, yellow/white papules. Also, there were some intra-oral papules and small papules on the lips. Multiple biopsies were taken and one lesion showed the typical histologic features of FF. Other skin lesions were consistent with perifollicular fibromas. No pulmonary cysts were detected by a chest computed tomography (CT). Magnetic resonance imaging of the kidneys showed a small cyst in the left kidney. At re-evaluation at age 48 years, more than 50 lipomas had been surgically removed and many were still present. Most were located on the trunk and the proximal limbs. Histologic evaluation had repeatedly shown common lipomas without specific features. The proband reported severe pain limiting daily physical activity, possibly related to the many lipomas that had been removed or to those still present. The proband reported a family history of lipomas, skin lesions and RCC in many family members, of which III-2, IV-2, IV-3 and IV-4 were also clinically evaluated. All of these family members had comparable skin lesions and lipomas. Images of the skin and mucosal phenotypes are shown in Figure 1B. Biopsies histologically confirmed the presence of FF in patients IV-3 and IV-4 (Fig. 2A). As in the proband, no pulmonary cysts were detected by a chest CT of patients III-4 and IV-3. Three affected family members were diagnosed with RCC: III-3 died of clear cell RCC at age 54 years, IV-3 had clear cell RCC at age 54 years and III-4 had bilateral and multifocal RCC of papillary and unclassified histology at age 68 years. In addition to the pedigree shown in Figure 1A, the proband mentioned nine more family members affected with lipomas. Of these, one had a history of prostate cancer and two had an unknown type of cancer at age 49 and 56 years, respectively. Another family member was mentioned to have a skin phenotype without lipomas. These additional family members did not consent with genetic testing or publication of clinical data. Several other types of cancer occurred in affected family members, namely lung adenocarcinoma, follicular thyroid cancer and non-Hodgkin’s lymphoma. Figure 2: Histology of skin and mucosal phenotype and PRDM10 sequence analyses. (A) Histology of skin lesions from patient IV-3: panel I is H&E staining of skin biopsy of the cheek showing strands of epithelium surrounded by stromal cells in loose connective tissue with mucin. This histological picture is consistent with a diagnosis of a fibrofolliculoma (squared red). Panel II is H&E staining of scrotum biopsy showing a skin tag of fibrous stroma covered by squamous cell epithelium. Scale bar is 200 μm. (B) Sanger sequencing on DNA isolated from tissues derived from family members (III-4) and (IV-3) confirms the PRDM10 c.2030G > A mutation without loss of the second PRDM10 allele. ## Genetic testing No PGV was detected in FLCN in the proband by conventional Sanger sequencing and multiplex ligation dependent probe amplification (MLPA). To investigate whether a shared haplotype of the FLCN gene co-segregated within the family, we employed microsatellite marker analysis using five microsatellite markers flanking FLCN (Supplementary Material, Fig. S1). A common haplotype could be identified in all affected individuals, except for individual IV-2. This individual has different haplotypes compared to all the other affected (including the D17S2196 marker closest to FLCN at 200kbp distance), suggesting that a common FLCN haplotype is unlikely to underlie the phenotype in this family. Even though no other features of PTEN hamartoma tumor syndrome were present, Sanger sequencing and MLPA of PTEN were performed in IV-2 but no variants predicted to be pathogenic were detected. The proband and two other affected family members (III-2 and IV-2) consented for WES and the results were analyzed for shared rare variants. A list of all detected shared rare variants with their considerations is shown in Supplementary Material, Table S1. Four shared predicted missense variants were detected, of which two did not co-segregate with the phenotype and one had a relatively high frequency in GnomAD [17]. Therefore, we considered the remaining missense variant in PRDM10 (NM_020228.3; c.2030G > A, p.(Cys677Tyr)) to be the most likely causal variant. Importantly, this PRDM10 variant was detected in a fourth affected family member (III-4) and subsequently also identified in her affected children IV-3 and IV-4. Thus PRDM10Cys677Tyr co-segregated with the phenotype within the family (nine informative meioses). When all non-genotyped individuals are set to an unknown genotype, this resulted in a Logarithm of Odds (LOD) score of 0.18, while the LOD score was 2.7 when all non-genotyped affected individuals were set to carry the PRDM10 variant. This latter LOD score is approximately the maximal LOD score that can be obtained from a family with this structure. Before family members IV-3 and IV-4 were linked to the proband, a WES-based hereditary cancer gene panel (Supplementary Material, Table S2) was performed in IV-3, but no other potential PGVs were detected. The identified PRDM10 variant is present in GnomAD (v2.1.1, all exomes) with an allele frequency of 4.8e-$4\%$ [17]. The Grantham score for cysteine and tyrosine is 194 and the variant is predicted to be deleterious by SIFT but unlikely to interfere with protein function by Align-GVGD (18–20). The cysteine at this location is highly conserved from mammals to zebrafish. To assess PRDM10 loss of heterozygosity, DNA was isolated from RCC and lipoma tissue from III-4 and from RCC, lung carcinoma and follicular thyroid carcinoma tissue from IV-3. Sanger sequencing displayed the PRDM10 c.2030G > A, p.(Cys677Tyr) variant without loss of the second allele (Fig. 2B). ## PRDM10Cys677Tyr in vitro model The PRDM10Cys677Tyr variant is located in the seventh C2H2 zinc finger domain of the PRDM10 protein (Fig. 3A). We used Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR)-based prime editing (PE) to establish a homozygous PRDM10Cys677Tyr human embryonic kidney cell line 293 T (HEK293T) to model this variant without interference of the WT allele (Fig. 3B). First, we assessed cellular localization of both WT and mutant PRDM10 using immunofluorescence. As shown in Supplementary Material, Figure S2A, PRDM10 was predominantly localized in the nucleus in both conditions, in line with its proposed transcriptional regulatory function [21]. Using quantitative reverse transcriptase-PCR (q RT-PCR) and western blotting (Fig. 3C and D), we found higher expression of mutant PRDM10 when compared to WT PRDM10 at both the mRNA and protein level. Flcn, together with Eif3b and Bccip, has been reported as a putative transcriptional target gene bound and regulated by Prdm10 in mice [21]. Analysis of mRNA expression levels of FLCN, EIF3B and BCCIP in our cell model revealed a significant decrease in FLCN expression in the PRDM10Cys677Tyr cells, while the expression levels of EIF3B and BCCIP were increased (Fig. 3C). Next, we assessed whether PRDM10Cys677Tyr affected folliculin protein levels by western blot (Fig. 3D). In line with the strong decrease of FLCN mRNA expression levels, folliculin protein was almost undetectable in PRDM10Cys677Tyr. For validation, the effects on FLCN expression were confirmed in a second, independent PRDM10Cys677Tyr 293 T cell line clone (Supplementary Material, Fig. S2B and C). Moreover, we did not observe significant changes in PTEN expression in the PRDM10Cys677Tyr cell lines (Supplementary Material, Fig. S2C). Figure 3: PRDM10Cys677Tyr phenocopies cellular effects of FLCN loss in human embryonic kidney cells. (A) Schematic presentation of PRDM10 with known domains. Variant Cys677Tyr is located in seventh zinc finger and indicated in red. There are seven isoforms of PRDM10 described at Uniprot, the longest is 1160. (B) Creation of endogenous mutants in 293 T cells. Sanger sequence chromatogram of DNA derived from CRISPR prime edited cells shows a homozygous endogenous PRDM10Cys677Tyr c.2030G > A mutation (indicated in orange). To prevent further genome editing, a silent mutation (indicated by asterisk) was introduced to disrupt the PAM site. (C) qPCR (n = 4) shows downregulation of FLCN expression upon PRDM10Cys677Tyr. Expression levels of EIF3B and BCCIP are increased. (D) Western blot (n = 3) shows downregulation of FLCN protein upon PRDM10Cys677Tyr. We detected higher PRDM10 protein levels in the PRDM10Cys677Tyr mutant as compared to the wild type (WT). GAPDH was used as a loading control. (E) qPCR showed that PRDM10Cys677Tyr resulted in upregulation of specific genes, which are also induced upon FLCN loss (FLCNKO) in 293 T cells. Results are representative for three independent experiments with at least two technical replicates. To determine quantitative gene expression levels, data were normalized to the geometric mean of two housekeeping genes. Note that FLCN mRNA is still detectable in the FLCNKO cell line, possibly due to incomplete nonsense-mediated mRNA decay or a transcriptional feedback mechanism. (F) Western blots (n = 2) of 293 T mutant cell lines. FLCN protein is absent in both PRDM10Cys677Tyr and FLCNKO 293 T. Induction of GPNMB in PRDM10Cys677Tyr was similar to induction levels observed in FLCNKO cells. Actin was used as loading control. (G) Both PRDM10Cys677Tyr and FLCNKO cells grew slower when compared to wild type. Cells were seeded in equal densities and total cell number was counted for six consecutive days (n = 2). ## Comparison to folliculin knockout Recently, we showed that folliculin loss in renal epithelial cells (RPTEC/TERT1) upregulates transcription of many genes, including RRAGD, GPNMB, CCND1 and MX1 [11]. In PRDM10Cys677Tyr 293 T cells, we observed qualitatively similar effects on the induction of these genes on mRNA levels as well as induction of GPNMB protein when compared to folliculin knockout 293 T cells (Fig. 3E and F). Note that FLCN transcript is still detectable in the FLCNKO cell line (Fig. 3E) yet folliculin protein expression is disrupted by a premature stop codon resulting from the gene editing procedure (Fig. 3F). This was also observed previously [11]. Detection of FLCN mRNA may relate to incomplete nonsense-mediated mRNA decay combined with active FLCN gene transcription. Growth curves of PRDM10WT, PRDM10Cys677Tyr and FLCNKO 293 T show that both mutant cell lines display a similar alteration in growth properties when compared to WT cells (Fig. 3G). ## PRDM10Cys677Tyr ChIP-qPCR We hypothesized that FLCN is a direct transcriptional target of PRDM10 in human cells. To test whether the Cys677Tyr variant alters PRDM10 binding to the FLCN promoter region, we performed ChIP-qPCR experiments. For these experiments, two independent primer sets surrounding the predicted PRDM10 binding motif (GGTGGTACGGCTCA) [21] were designed (Fig. 4A). Enrichments were calculated compared to a random region ~20 kB upstream (Fig. 4B, left plot) or downstream (Fig. 4B, right plot) of the FLCN promoter region. Fold enrichments of FLCN promoter DNA bound by PRDM10 in WT (PRDM10WT) or PRDM10Cys677Tyr mutant cells show that the FLCN promoter region is bound by PRDM10WT. Clearly, PRDM10Cys677Tyr results in a strong decrease in FLCN promoter binding, hampering its transcription. Figure 4: FLCN is a transcriptional target of PRDM10. (A) Schematic overview of primer locations used for X-ChIP qPCR experiments as described in (B). (B) FLCN promoter binding by PRDM10 was assessed by PRDM10 X-ChIP of 293 T PRDM10 wildtype and Cys677Tyr (C677Y) mutant cells. Fold enrichment of binding capacities between wild-type and C677Y mutant PRDM10 was determined by qPCR, showing that the mutation strongly diminished promoter binding. Bar graphs are representative of differences observed in three independent ChIP experiments, with two technical replicates per qPCR, and normalized to input and a predicted non-binding region ~20 kB upstream (left) or downstream (right) of the FLCN promoter. Two different primer sets surrounding the predicted PRDM10 motif (GGTGGTACGGCTCA) in the FLCN promoter were used. (C) qPCR (n = 2) shows that inducible overexpression (OE) of PRDM10Cys677Tyr slightly repressed FLCN expression within 72 h, while this effect did not occur upon overexpression of PRDM10WT in 293 T cells. (D) Volcano plot showing gene expression changes upon five days induction of PRDM10Cys677Tyr in RPTEC/TERT1. 6087 genes are differentially expressed (P-value ≤ 0.05). Top 20 of most significant differential genes are indicated, plus, as a reference, these marker genes: PRDM10, FLCN, GPNMB, RRAGD, PTEN, CCND1, CDH1, CDH3, ACP5, PDGFRB and CCL2. Asterisks indicate genes of potential interest overlapping between FLCNKO and PRDM10Cys677Tyr cell lines. (E) Volcano plot showing gene expression changes upon FLCNKO in RPTEC/TERT1. A total of 3703 genes are differentially expressed (P-value ≤ 0.05). Top 20 of most significant differential genes are indicated, plus, as a reference, these marker genes: PRDM10, FLCN, GPNMB, RRAGD, PTEN, CCND1, CDH1, CDH3, ACP5, PDGFRB and CCL2. Asterisks indicate genes of potential interest overlapping between FLCNKO and PRDM10Cys677Tyr cell lines. (F) Overlap of gene expression changes upon FLCN loss or PRDM10Cys677Tyr induction in RPTECs. A total of 446 genes are significantly upregulated in both conditions and 473 genes are significantly downregulated in both conditions. Gene lists are provided as Supplementary Material, Table S3. ## PRDM10Cys677Tyr inducible overexpression To investigate possible dominance of PRDM10Cys677Tyr over PRDM10WT, we performed experiments with inducible overexpression (OE) of PRDM10Cys677Tyr in the presence of endogenous, WT PRDM10 in 293 T cells. By qPCR, it was shown that OE of PRDM10Cys677Tyr slightly represses FLCN expression within 72 h, and this effect does not occur upon OE of PRDM10WT (Fig. 4C). However, a specific decrease of folliculin protein was not yet visible after 72 h of PRDM10Cys677Tyr OE (Supplementary Material, Fig. S2D and E). Prolonged, 120-h induction of PRDM10Cys677Tyr in RPTECs modestly repressed FLCN but strongly altered the expression of multiple other genes (6087 differentially expressed genes, Fig. 4D). The gene expression changes induced by PRDM10Cys677Tyr showed similarities, but also differences, with the effects of knocking out FLCN (3703 differentially expressed genes, Fig. 4E). These results show that PRDM10Cys677Tyr, apart from curtailing FLCN expression, alters the expression of a variety of other genes, too. Furthermore, the transcriptional effects of increased PRDM10Cys677Tyr levels show overlap but are not identical to the gene expression patterns resulting from loss of FLCN (Fig. 4F and Supplementary Material, Table S3). ## Discussion We here present a family initially clinically diagnosed with BHD but without a detectable PGV in FLCN. Many years later, the family was re-evaluated upon their own request and additional genetic testing was performed. A missense germline variant in the PRDM10 gene was detected that co-segregated with the phenotype within the family. The presence of nine informative meiosis equals a probability of $\frac{1}{512}$ that co-segregation of the variant in this family occurred by chance [22]. Therefore, we considered the PRDM10 variant as the most likely causal variant underlying this novel syndrome. This is the first association of a germline variant in PRDM10 with a Mendelian human disease (MIM #618319). The clinical features of the here presented family overlap with BHD, and a comparison of the clinical phenotype of the here described family and BHD is shown in Table 1. The initial clinical diagnosis of BHD was based on the presence of FF and multiple perifollicular fibromas, which are lesions in the spectrum of FF [23,24]. However, the location, color and morphology of the skin lesions were partly different from those in BHD. Since the initial presentation, three family members had developed RCC, which is one of the hallmark manifestations of BHD. Phenotypically, there were two main differences with BHD. First, no pulmonary cysts were present in three patients who underwent a chest CT, whereas pulmonary cysts are present in 70–$100\%$ of BHD patients (2,16,25–30). Second, extensive lipomatosis was present in many patients in the family described here, which is not an established feature of BHD. In our cohort of more than 300 BHD and BHD-like patients, we have no other families with such extensive lipomatosis and no families with definite BHD without an identifiable FLCN pathogenic variant. Our cohort does contain some families with BHD-like features in which we have attempted to identify the genetic cause, for example by WES, but in none a (possible) pathogentic variant in PRDM10 was identified. Multiple lipomas can occur in the presumed autosomal dominant conditions familial multiple lipomatosis (MIM #151900) and Dercum disease (MIM #103200) [31]. No genes associated with these diseases have been identified yet. Multiple lipomas can also occur in PTEN hamartoma tumor syndrome (MIM #158350), which is an autosomal dominant condition caused by PGVs in the PTEN gene [32,33]. We considered the diagnosis of PTEN hamartoma tumor syndrome unlikely, since no other features of PTEN hamartoma tumor syndrome were present and no PGV was detected in PTEN. Table 1 \| Unnamed: 0 \| Birt–Hogg–Dubé syndrome \| PRDM10 family \| \| --- \| --- \| --- \| \| Renal cell carcinoma \| \| \| \| Prevalence \| 7–34% (2,15,16,25–27,34) \| 3/10 carriers and obligate carriers3/22 including family members affected with lipomas/skin phenotype by hearsay \| \| Mean age at first RCC \| 50 (16,26,34,35) \| 59 \| \| Subtype \| Most subtypes can occurSubtypes with chromophobe component most common (2,15,26,34,36,37) \| Clear cell, papillary type 2, unclassified \| \| Pneumothorax Prevalence \| 22–74% (2,14–16,25,27–30) \| 0 \| \| Pulmonary cysts Prevalence \| 70–100% (2,16,25–30) \| 0/3 \| \| Skin phenotype \| \| \| \| Prevalence \| 80–90% (in Europe and USA) (15,27,28) \| All assessed patients (n = 5) \| \| Morphology \| White to yellow papules on the face, neck and upper trunk (1,38) \| White to yellow and skin-colored papules on the face, neck, trunk, nipples and scrotum. Areas of partly confluent papules on the trunk Mucosal papules \| \| Histology \| Fibrofolliculoma spectrum (38–41) \| Fibrofolliculoma spectrum \| \| Lipomas \| No proven association \| All assessed patients (n = 5)9 more family members by hearsay \| We also performed WES analysis in two other familial cases of suspected BHD without an identifiable PGV in FLCN, but no other variants in PRDM10 were detected. PRDM10 is 1 of 19 PRDM genes currently known in humans [42]. It belongs to the PR/SET (positive regulatory domain-binding factor 1 (PRDI-BF1) and retinoblastoma interacting zinc finger (RIZ) homology domain containing/Su(var)3-9, enhancer-of-zeste and trithorax) transcription factor family which share a conserved N-terminal PR domain that has similarities with the lysine methyltransferase SET domain, followed by variable C2H2-type zinc finger repeats [43,44]. In mice, homozygous knockout of Prdm10 leads to preweaning lethality and Prdm10 functions as a sequence-specific transcription factor which is essential in supporting global translation during early development [21]. In humans, PRDM10 is expressed in almost all tissues including skin, kidney and adipose tissue [45,46]. We found a significant decrease of FLCN expression in the PRDM10Cys677Tyr cells. Since Flcn was suggested to be a target of Prdm10 in mice [21] and because the Cys677Tyr variant is located in the DNA-binding domain of PRDM10, we hypothesized that FLCN could be a direct transcriptional target of PRDM10 in humans. Indeed, we observed that WT PRDM10 binds to the FLCN promoter region and that PRDM10Cys677Tyr results in a strong decrease in promoter binding, curtailing FLCN expression in human cells. These results identify PRDM10 as an upstream regulator of the folliculin tumor suppressor. Indeed, PRDM10Cys677Ty resulted in several effects comparable to those of the loss of folliculin in vitro, such as induction of GPNMB protein and reduced proliferation. While reduced proliferation may seem paradoxical given the proven role of FLCN as a tumor suppressor, this observation is in line with our previous studies of FLCNKO RPTECs, which experience a growth disadvantage upon folliculin loss due to induction of a non-canonical interferon response, counterbalancing the TFE3/TFEB-directed hyperproliferation upon folliculin loss [11]. So, in the family described here, a reduction in folliculin expression levels caused by the PRDM10 variant might largely explain the occurrence of FF and RCC, phenocopying BHD. However, the absence of a pulmonary phenotype in this family may reflect a specific role of PRDM10, where tissue specific expression patterns may play a role. Indeed, the protein levels of PRDM10 in human lung may be lower than in tissues that are affected in this family [46,47]. Hypothetically, FLCN expression in the lungs may be under control of additional transcription factors, overruling the effects of the PRDM10 variant. Although the development of the pulmonary phenotype in BHD has not been investigated in great detail, it was shown that FLCN deficiency hampers the cellular E-cadherin-LKB1-AMPK axis and that folliculin is required for alveolar epithelial cell survival [48]. It will be interesting to investigate whether PRDM10 could act as an upstream regulator of the FLCN-E-cadherin-LKB1-AMPK axis in lung cells and how the variant may affect this pathway. The lipomatosis could be either a direct effect of the PRDM10 variant, for example by gene expression changes other than FLCN reduction, or it may be an adipose tissue-specific effect of folliculin repression and downstream TFE3/TFEB activation, which is described to play a role in adipose tissue (49–52). Heterozygous loss of Prdm10 expression was reported to result in a reduction of fat mass in male mice [53], confirming role for Prdm10 in adipose tissue. Based on these mouse data, with an apparent opposite adipose tissue phenotype than observed in the family reported here, it could be hypothesized that the lipomatosis in this family is not the result of reduced folliculin expression, but due to a change in the expression of a specific gene target of PRDM10Cys677Tyr (Figs 3C and 4D). In addition to RCC, several other types of cancer have occurred in the affected members of this family. Future research will be needed to understand whether these might be caused by the PRDM10 variant. Alterations in the PRDM genes are linked to various tumors, as recently reviewed by Casamassimi et al., so it is possible that the variant in this family may predispose for cancers other than RCC as well [54]. Also, there is some evidence for a role of somatic aberrations of PRDM10 in cancer. PRDM10 gene fusions with either MED12 or CITED2 have been reported in a small proportion of undifferentiated pleomorphic sarcomas. These tumors have a relatively benign course, a distinct expression pattern and no or limited other mutations or numerical chromosomal aberrations [55,56]. PRDM10 increases B-cell lymphoma-2 (Bcl-2) expression in vitro [57]. The Bcl-2 gene is upregulated in a wide variety of cancers including RCC and selective Bcl-2 inhibition was reported as a potential strategy in the treatment of RCC [58,59]. Furthermore, PRDM10 is overexpressed in breast, colon and liver cancer samples [42,60]. A relative OE in kidney cancer samples was also present, but the difference with normal tissue was not significant [42]. These data also suggest that PRDM10 may play a role in cancer, but the mechanism requires further studies. Remarkably, in our in vitro experiments in human cells, is that the mutant PRDM10 was expressed at higher levels than the WT PRDM10. We have not studied the mechanism underlying the increased expression in detail, but a possible explanation for the increased PRDM10 protein levels is the existence of a feedback loop leading to increased PRDM10 promoter activity, as indicated by the increased mutant PRDM10 mRNA levels. In the Eukaryotic Promoter Database [61,62], we noted that the predicted PRDM10 binding motif is also present in its own promoter region, suggesting that PRDM10 may be capable of regulating its own expression, and that this may be affected by the Cys677Tyr variant. Alternatively, another upstream regulator of PRDM10 expression could be affected by the Cys677Tyr variant which results in an increase of PRDM10 expression, or the Cys677Tyr variant increases PRDM10 mRNA stability. Whether PRDM10 functions as a more canonical tumor suppressor gene, requiring functional inactivation of the second allele for development of neoplasms, is not clear yet. In the family described here, the tumors and lipoma lesions investigated did not show loss of the second PRDM10 allele, although epigenetic silencing or a second hit of the second PRDM10 allele was not excluded. PRDM10Cys677Tyr also affects gene expression in a way independent of the reduced folliculin levels. Based on these combined results, we propose that PRDM10Cys677Tyr could act as either a neomorphic or hypomorphic allele, and may have a dominant effect over the second PRDM10 allele, at least for some transcriptional targets. It is likely that external factors or (epi)genetic aberrations in other genomic regions played a role in the development of the observed neoplasms, too. In conclusion, we identified a distinguishable syndrome partly overlapping with BHD, consisting of multiple lipomas, FF and RCC, caused by a missense variant in PRDM10. To further pinpoint the roles of PRDM10 in different tissues and its role as a hereditary cancer gene, future studies in custom-made in vivo models are required. Our observations can serve as the basis for further functional studies into the roles of PRDM10 as a disease gene, and provide further insight into BHD, lipomatosis and RCC pathogenesis. ## Germline genetic testing All germline genetic testing (Sanger sequencing, MLPA and WES) was performed in the diagnostic setting in laboratories in the Netherlands accredited in accordance with ISO15189. Segregation analysis of the PRDM10 variant in III-9, IV-14 and IV-15 was performed by Sanger sequencing using the following primers: Fw 5’ CCCCGATAAACTGCGACT 3′ and Rev 5’ GAGAACCACCTTGGGCTG 3′. WES was performed as described before [63]. Median and average coverage of the exome target region was >100x for each sample. Variant prioritization was performed using Alissa Interpret (Agilent Technologies, Santa Clara, CA, USA). In short, a classification tree was used to select for variants present in all three affected individuals and virtually absent in control cohorts dbSNP build 142 (http://www.ncbi.nlm.nih.gov/projects/SNP), 1000 Genomes Phase 3 release v5.20130502, and ESP6500 (http://evs.gs.washington.edu/EVS/) as well as in house controls. Prerequisite was that these variants had been genotyped in at least 200 alleles. Subsequently, the remaining variants were manually inspected and further prioritized based on literature, predicted (deleterious) effects on protein function by e.g. truncating the protein, affecting splicing, amino acid change and evolutionary conservation. ## LOD score calculation A two-point LOD score calculation was performed using Superlink-Online SNP [64] for the PRDM10 variant using the pedigree and phenotype as depicted in Figure 1A. ## Microsatellite marker haplotypes analysis Microsatellite marker haplotypes analysis was performed using five microsatellite markers flanking the FLCN gene on an ABI 3700 Genetic Analyzer (Applied Biosystems, Carlsbad, CA, USA) [65]. Data were analyzed using the genemapper 5.0 software (Applied Biosystems, Carlsbad, CA, USA). ## Study approval The Medical Research Involving Human Subjects Act (WMO) did not apply to this study, since all genetic testing was performed in the diagnostic setting. Informed consent for publication of clinical data, photographs and use of tumor tissues was obtained. ## DNA isolation and sanger sequencing DNA was extracted from blood and Formalin-Fixed Paraffin-Embedded (FFPE) tissues and equal amounts of DNA were amplified by PCR. Tubes were placed in a thermal cycler (Veriti, Thermo Fisher Scientific Inc, Waltham, MA, USA) for amplification with specific PCR primer mixes (10 μm). PCR program used for amplification was 1 cycle of 94°C for 3 minutes, 5 cycles of 94°C for 30 sec, 65°C for 30 sec, 72°C for 120 sec, 30 cycles of 94°C for 30 sec, 60°C for 30 sec, 72°C for 2 minutes, 72°C for 10 minutes and ending in a rapid thermal ramp to 10°C. After PCR purification (QIAquick PCR Purification Kit, Qiagen, Germany) took place, samples were further analyzed by sequencing. Sequencing was either performed in-house or at Eurofins Genomics. For PCR and sequencing of PRDM10, following primers were used: Fw 5’ CCCGATAAACTGCGACTCCACAT 3′ and Rev 5’ GGTCCAGTTCATCAGAGGTGGGTG 3’. ## Cell culture and genome editing Human embryonic kidney cells (HEK293T, ATCC CRL-3216™) were maintained in Dulbecco’s Modified Eagle Medium (DMEM, Gibco, Life Technologies, Thermo Fisher Scientific Inc, Waltham, MA, USA) supplied with $8\%$ fetal bovine serum (FBS, F0804, Sigma-Aldrich, St. Louis, MO, USA). Renal proximal tubular epithelial cells (RPTEC/TERT1, ATCC CRL-4031™) were maintained in DMEM/F12 (Gibco, Life Technologies, Thermo Fisher Scientific Inc, Waltham, MA, USA) according to the manufacturer’s protocol with addition of $2\%$ fetal bovine serum. Cell lines were cultured in a humidified atmosphere at 37°C and $5\%$ CO2 and were regularly tested to exclude Mycoplasma infections. To introduce the PRDM10 variant endogenously, CRISPR PE was used [66]. First, a doxycycline inducible PE protein plasmid, containing a Cas9 nickase fused to a RT domain, was cloned and stably expressed in 293 T cells. Then, cells were seeded in the presence of doxycycline (10 ng/ml, Sigma-Aldrich, St. Louis, MO, USA) and the next day pegRNA plasmid, designed to both introduce the PRDM10 c.2030G > A variant and disrupt the PAM-site (available on request), was transfected using Lipofectamine 3000 reagent (Thermo Fisher Scientific Inc, Waltham, MA, USA) into Cas9-RT expressing 293 T cells. The FLCN knockout cell lines were created using Synthego’s Synthetic cr:tracrRNA Kit and corresponding manual. Cas9/gRNA (FLCN_exon 4 GAGAGCCACGAUGGCAUUCA + modified EZ scaffold) RNP complexes were transfected transiently using Neon Electroporation System (Thermo Fisher Scientific Inc, Waltham, MA, USA). For all CRISPR experiments, cells were grown in limiting dilution in 96-wells plates to generate single cell clones and genome editing status was assessed by Sanger sequencing. Sequenced samples were analyzed by manual alignment and using the Synthego ICE analysis (ice.synthego.com) tool which gives a quantitative score of editing efficiency. ## Virus production and infection To create inducible cell lines that overexpress PRDM10WT (293 T) and PRDM10Cys677Tyr (293 T and RPTEC/TERT1), lentiviral production and transduction took place according to the Lenti-X Tet-On 3G Inducible Expression System (Clontech, Takara Bio, Japan) technical manual. In short, PRDM10WT and PRDM10Cys677Tyr cDNA were cloned into the pLVX-Tre3G plasmid where after Tre3G-Cas9 and Tet3G lentiviral particles were produced in 293 T cells. For transduction, cells were seeded in a 6-wells plate. The next day growth media was changed for 1 ml media containing viruses. Cells were incubated overnight and after 24 hours media was replaced with 2 ml fresh media. The next day cells were transferred to 10 cm plates and Puromycin (3 μg/ml, Sigma-Aldrich, St. Louis, MO, USA) was added to select for successfully transduced cells. ## RNA isolation and quantitative RT-PCR The High Pure RNA Isolation Kit (Roche, Penzberg, Germany) was used to extract RNA from the dry cell pellet. For qRT-PCR we used Biorad iScript cDNA Synthesis Kit and LightCycler 480 FastStart DNA Master SYBR Green I (Roche, Penzberg, Germany). Measurements were performed with LightCycler 480 System and corresponding software (Roche, Penzberg, Germany). To determine the quantitative gene expression data levels were normalized to the geometric mean of two housekeeping genes. All experiments were at least performed in duplicate with three technical replicates per experiment. Primer sequences used in this study are as follows: PRDM10 Fw 5’ CAGGAACTGAAGGTGTGGTATG 3’ Rev 5’ GCTCTCGAAGAACTTTCCTTTCT 3’ CCND1 Fw 5’ GCGGAGGAGAACAAACAGAT 3’ Rev 5’ GAGGGCGGATTGGAAATGA 3’ EIF3B Fw 5’ GGAGACCGCACTTCCATATTC 3’ Rev 5’ CTTAGGAGACCAACGCACATAC 3’ BCCIP Fw 5’ AGAACCATATTGGGAGTGTGATTA 3’ Rev 5’ ACACTGGGTACCCTTTCTTTC 3’ FLCN Fw 5 ‘GGAGAAGCTCGCTGATTTAGAAGAGGA 3‘ Rev 5’ ACCCAGGACCTGCCTCATG 3′ MX1 Fw 5’ GACAATCAGCCTGGTGGTGGTC 3’ Rev 5’ GTAACCCTTCTTCAGGTGGAACACG 3’ GPNMB Fw 5’ CCTCGTGGGCTCAAATATAAC 3’ Rev 5’ TTTCTGCAGTTCTTCTCATAGAC 3’ RRAGD Fw 5’ CCTGGCTCTCGTTTGCTTTGTCAG 3’ Rev 5’ GGGGTGGCTCTCTTTTTCTTCTGC 3’ HPRT1 Fw 5’ TGACACTGGGAAAACAATGCA 3‘Rev 5 ‘GGTCCTTTTCACCAGCAAGCT 3‘ TBP Fw 5 ‘TGCACAGGAGCCAAGAGTGAA 3‘ Rev 5’ CACATCACAGCTCCCCACCA 3‘ ## Immunoblot For western blotting, dry cell pellets were lysed in RIPA lysis buffer (89900, Thermo Scientific) supplemented with protease and phosphatase inhibitors (Roche, Penzberg, Germany). Lysates were boiled at 70°C for 5 min in 1x NuPAGE LDS sample buffer (Novex NP0007, Thermo Fisher Scientific Inc, Waltham, MA, USA) with $10\%$ 1 M DTT (Sigma) and equal amounts were separated by 4–$15\%$ sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS-PAGE) (BioRad, Hercules, CA, USA) and blotted onto polyvinylidene fluoride transfer membranes (Merck Millipore, MA, USA). Subsequently, membranes were blocked for 1 hour at room temperature with $5\%$ milk (ELK, Campina, Amersfoort, Netherlands) in TBST. The primary antibody incubation was overnight at 4°C in $2.5\%$ milk in TBST. The following day, membranes were washed and incubated with appropriate secondary antibodies (Dako, Agilent, CA, USA) for 3 h at 4°C in $2.5\%$ milk in TBST. After incubation, the membranes were thoroughly washed and bands were visualized by chemiluminescence (ECL prime, Amersham, VWR, Radnor, PA, USA) in combination with ChemiDoc Imaging Systems (BioRad Laboratories, CA, USA). The following antibodies were used according to individual datasheets: PRDM10 Bethyl (A303-204A, Bethyl laboratories, TX, USA), FLCN (D14G9, CST 3697S, Cell Signaling Technologies, MA, USA), GPNMB (AF2550-SP, R&D systems, MN, USA), PTEN (sc-7974, Santa Cruz, TX, USA), β-Actin (sc-47778, Santa Cruz, TX, USA) and GAPDH (sc-47724, Santa Cruz, TX, USA). ## ChIP-qPCR For chromatin immunoprecipitation experiments, cells were cross-linked with $1\%$ formaldehyde for 10 minutes at room temperature, quenched with 125 mm glycine and washed twice in cold PBS. Chromatin extracts were obtained by consecutive rounds of lysis in LB1 (50 mm Tris, pH 7.5, 140 mm NaCl, 1 mm EDTA, $10\%$ glycerol, $0.5\%$ NP-40, $0.25\%$ Triton X-100), LB2 (10 mm Tris, pH = 8.0, 200 mm NaCl, 1 mm EDTA, 0.5 mm EGTA) and LB3 (10 mm Tris, pH = 8.0, 100 mm NaCl, 1 mm EDTA, 0.5 mm EGTA, $0.1\%$ Na-Deoxycholate, $0.5\%$ N-lauroylsarcosine), supplemented with protease inhibitor cocktail (Roche, Penzberg, Germany). Chromatin DNA was sheared to a size range of 100–500 bp using sonication for 12,5 minutes (30 sec on, 30 sec off at high amplitude) using a Bioruptor sonicator (Diagenode, NJ, USA). Triton X-100 was added to a final concentration of $1\%$ and lysates were cleared by centrifugation, where after 50 μl input sample was taken. Subsequently, 40 μl A/G agarose beads (sc-2003, Santa Cruz, TX, USA) and 5 μg PRDM10 antibody (pre-coupled overnight at 4°C) were added to sonicated chromatin and incubated overnight with rotation at 4°C. The next day, beads were washed with RIPA lysis buffer (89 900, Thermo Fisher Scientific Inc, Waltham, MA, USA) 10 times. To elute and reverse cross-links, samples were incubated with 200 μl elution buffer ($2\%$ SDS in TE buffer) overnight at 65°C then treated with proteinase K. DNA was then purified by phenol:chloroform extraction. ChIP and input DNA were measured using quantitative RT-PCR, performed as described above. Fold enrichments of binding capacities were calculated between WT and mutant PRDM10, and normalized to input and a predicted non-binding region ~20 kB upstream or downstream of the FLCN promoter. Two different primer sets surrounding the predicted PRDM10 motif (GGTGGTACGGCTCA) in the FLCN promoter were used: FLCN_promoter_set1_Fw 5’ CTGTGTTCCTGGGCTTGC 3’ FLCN_promoter_set1_Rev 5’ CCGGGTTCAGGCTCTCA 3’ FLCN_promoter_set2_Fw 5’ AGTTGTAGGACTCGGACTGTG 3’ FLCN_promoter_set2_Rev 5’ AGCTGGCAGAACCAGGA 3’ FLCN_promoter_upstream_Fw 5’ CAGTCTGGGCAACTAAGTAAGA 3’ FLCN_promoter_upstream_Rev 5’ CAAGGGAACCTCCTGTTTCA 3’ FLCN_promoter_downstream_Fw 5’ AGGTGTCAATGTCATGGAAGTTA 3’ FLCN_promoter_downstream_Rev 5’ AGGAGATACTACAGGACCCATC 3’ ## Immunofluorescent staining 293 T cells were grown on cover slips, fixed in $2\%$ paraformaldehyde for 15 minutes at room temperature and subsequently in $70\%$ ice cold EtOH for 1 hour. Next, cells were permeabilized in $0.3\%$ Triton X-100 for 5 min, blocked in $3\%$ BSA with $0.3\%$ Triton X-100 for 45 minutes, incubated with rabbit anti-PRDM10 antibody (1:100, HPA026997, Atlas antibodies, Sweden) for 1.5 hour and secondary antibody for 1 hour at room temperature. Cells were mounted using ProLong™ Gold Antifade Mountant with DAPI (Invitrogen, Thermo Fisher Scientific Inc, Waltham, MA, USA) and examined using fluorescence microscopy (Leica, Germany). ## Overexpression experiments and RNA sequencing For overexpression (OE) experiments, cells were treated with doxycycline (250 ng/ml, Sigma-Aldrich, St. Louis, MO, USA) to induce expression of PRDM10WT or PRDM10Cys677Tyr. For Illumina-based RNA sequencing specifically, RPTEC/TERT1 WT, RPTEC/TERT1 FLCNKO and RPTEC/TERT1 PRDM10Cys677Tyr (upon 120 hours of induction) cell lines were harvested in duplicate and RNA was extracted from the dry cell pellets. Then, samples were prepped using TruSeq Stranded mRNA Library Preparation Kit according to TruSeq Stranded mRNA Sample Preparation Guide. Sequencing was performed on an Illumina HiSeq 4000 (Illumina, San Diego, CA, USA) using run mode SR50. Reads were trimmed using sickle-1.33 [67] and aligned to hg19 using hisat2-2.0.4 [68]. The alignments were assigned to genes and exons using featurecount-1.5.0-p3 [69] using the gene annotation provided by the iGenomes resource [70]. To compare RNA-sequencing profiles between PRDM10Cys677Tyr OE and WT RPTECs, the R package edgeR was used [71]. Obtained P-values were corrected for multiple testing using the Benjamini–Hochberg false discovery rate step-up procedure [72]. The volcano plots in Figure 4D and E were generated with VolcaNoseR web app [73]. The raw count data are deposited in the Zenodo Repository and are openly available via https://doi.org/10.5281/zenodo.6420633. ## References 1. Birt A.R., Hogg G.R., Dube W.J.. Hereditary multiple fibrofolliculomas with trichodiscomas and acrochordons. Arch. Dermatol. (1977) 113 1674-1677. PMID: 596896 2. Zbar B., Alvord W.G., Glenn G., Turner M., Pavlovich C.P., Schmidt L., Walther M., Choyke P., Weirich G., Hewitt S.M.. Risk of renal and colonic neoplasms and spontaneous pneumothorax in the Birt-Hogg-Dube syndrome. Cancer Epidemiol. Biomark. Prev. (2002) 11 393-400 3. Toro J.R., Glenn G., Duray P., Darling T., Weirich G., Zbar B., Linehan M., Turner M.L.. Birt-Hogg-Dube syndrome: a novel marker of kidney neoplasia. Arch. Dermatol. (1999) 135 1195-1202. PMID: 10522666 4. Nickerson M.L., Warren M.B., Toro J.R., Matrosova V., Glenn G., Turner M.L., Duray P., Merino M., Choyke P., Pavlovich C.P.. Mutations in a novel gene lead to kidney tumors, lung wall defects, and benign tumors of the hair follicle in patients with the Birt-Hogg-Dubé syndrome. Cancer Cell (2002) 2 157-164. PMID: 12204536 5. Hudon V., Sabourin S., Dydensborg A.B., Kottis V., Ghazi A., Paquet M., Crosby K., Pomerleau V., Uetani N., Pause A.. Renal tumour suppressor function of the Birt-Hogg-Dube syndrome gene product folliculin. J. Med. Genet. (2010) 47 182-189. PMID: 19843504 6. Khoo S.K., Kahnoski K., Sugimura J., Petillo D., Chen J., Shockley K., Ludlow J., Knapp R., Giraud S., Richard S., Nordenskjöld M., Teh B.T.. Inactivation of BHD in sporadic renal tumors. Cancer Res. (2003) 63 4583-4587. PMID: 12907635 7. Napolitano G., Di Malta C., Esposito A., Araujo M.E.G., Pece S., Bertalot G., Matarese M., Benedetti V., Zampelli A., Stasyk T.. A substrate-specific mTORC1 pathway underlies Birt-Hogg-Dube syndrome. Nature (2020) 585 597-602. PMID: 32612235 8. El-Houjeiri L., Possik E., Vijayaraghavan T., Paquette M., Martina J.A., Kazan J.M., Ma E.H., Jones R., Blanchette P., Puertollano R.. The transcription factors TFEB and TFE3 link the FLCN-AMPK signaling axis to innate immune response and pathogen resistance. Cell Rep. (2019) 26 3613-3628.e6. PMID: 30917316 9. Endoh M., Baba M., Endoh T., Hirayama A., Nakamura-Ishizu A., Umemoto T., Hashimoto M., Nagashima K., Soga T., Lang M.. A FLCN-TFE3 feedback loop prevents excessive glycogenesis and phagocyte activation by regulating lysosome activity. Cell Rep. (2020) 30 1823-1834.e5. PMID: 32049013 10. Petit C.S., Roczniak-Ferguson A., Ferguson S.M.. Recruitment of folliculin to lysosomes supports the amino acid-dependent activation of Rag GTPases. J. Cell Biol. (2013) 202 1107-1122. PMID: 24081491 11. Glykofridis I.E., Knol J.C., Balk J.A., Westland D., Pham T.V., Piersma S.R., Lougheed S.M., Derakhshan S., Veen P., Rooimans M.A.. Loss of FLCN-FNIP1/2 induces a non-canonical interferon response in human renal tubular epithelial cells. elife (2021) 10 12. Menko F.H., Steensel M.A., Giraud S., Friis-Hansen L., Richard S., Ungari S., Nordenskjold M., Hansen T.V., Solly J., Maher E.R.. Birt-Hogg-Dube syndrome: diagnosis and management. Lancet Oncol. (2009) 10 1199-1206. PMID: 19959076 13. Schmidt L.S., Linehan W.M.. Molecular genetics and clinical features of Birt-Hogg-Dube syndrome. Nat. Rev. Urol. (2015) 12 558-569. PMID: 26334087 14. Toro J.R., Wei M.H., Glenn G.M., Weinreich M., Toure O., Vocke C., Turner M., Choyke P., Merino M.J., Pinto P.A.. BHD mutations, clinical and molecular genetic investigations of Birt-Hogg-Dube syndrome: a new series of 50 families and a review of published reports. J. Med. Genet. (2008) 45 321-331. PMID: 18234728 15. Houweling A.C., Gijezen L.M., Jonker M.A., Doorn M.B., Oldenburg R.A., Spaendonck-Zwarts K.Y., Leter E.M., Os T.A., Grieken N.C., Jaspars E.H.. Renal cancer and pneumothorax risk in Birt-Hogg-Dube syndrome; an analysis of 115 FLCN mutation carriers from 35 BHD families. Br. J. Cancer (2011) 105 1912-1919. PMID: 22146830 16. Schmidt L.S., Nickerson M.L., Warren M.B., Glenn G.M., Toro J.R., Merino M.J., Turner M.L., Choyke P.L., Sharma N., Peterson J.. Germline BHD-mutation spectrum and phenotype analysis of a large cohort of families with Birt-Hogg-Dube syndrome. Am. J. Hum. Genet. (2005) 76 1023-1033. PMID: 15852235 17. Karczewski K.J., Francioli L.C., Tiao G., Cummings B.B., Alföldi J., Wang Q., Collins R.L., Laricchia K.M., Ganna A., Birnbaum D.P.. The mutational constraint spectrum quantified from variation in 141,456 humans. Nature (2020) 581 434-443. PMID: 32461654 18. Vaser R., Adusumalli S., Leng S.N., Sikic M., Ng P.C.. SIFT missense predictions for genomes. Nat. Protoc. (2016) 11 1-9. PMID: 26633127 19. Grantham R.. Amino acid difference formula to help explain protein evolution. Science (1974) 185 862-864. PMID: 4843792 20. Tavtigian S.V., Deffenbaugh A.M., Yin L., Judkins T., Scholl T., Samollow P.B., Silva D., Zharkikh A., Thomas A.. Comprehensive statistical study of 452 BRCA1 missense substitutions with classification of eight recurrent substitutions as neutral. J. Med. Genet. (2006) 43 295-305. PMID: 16014699 21. Han B.Y., Seah M.K.Y., Brooks I.R., Quek D.H.P., Huxley D.R., Foo C.S., Lee L.T., Wollmann H., Guo H., Messerschmidt D.M., Guccione E.. Global translation during early development depends on the essential transcription factor PRDM10. Nat. Commun. (2020) 11 3603. PMID: 32681107 22. Jarvik G.P., Browning B.L.. Consideration of cosegregation in the pathogenicity classification of genomic variants. Am. J. Hum. Genet. (2016) 98 1077-1081. PMID: 27236918 23. Shvartsbeyn M., Mason A.R., Bosenberg M.W., Ko C.J.. Perifollicular fibroma in Birt-Hogg-Dube syndrome: an association revisited. J. Cutan. Pathol. (2012) 39 675-679. PMID: 22725638 24. Starink T.M., Houweling A.C., Doorn M.B., Leter E.M., Jaspars E.H., Moorselaar R.J., Postmus P.E., Johannesma P.C., Waesberghe J.H., Ploeger M.H.. Familial multiple discoid fibromas: a look-alike of Birt-Hogg-Dubé syndrome not linked to the FLCN locus. J. Am. Acad. Dermatol. (2012) 66 259.e1-259.e9 25. Guo T., Shen Q., Ouyang R., Song M., Zong D., Shi Z., Long Y., Chen P., Peng H.. The clinical characteristics of east Asian patients with Birt-Hogg-Dube syndrome. Ann Transl Med (2020) 8 1436. PMID: 33313181 26. Furuya M., Yao M., Tanaka R., Nagashima Y., Kuroda N., Hasumi H., Baba M., Matsushima J., Nomura F., Nakatani Y.. Genetic, epidemiologic and clinicopathologic studies of Japanese Asian patients with Birt-Hogg-Dube syndrome. Clin. Genet. (2016) 90 403-412. PMID: 27220747 27. Toro J.R., Pautler S.E., Stewart L., Glenn G.M., Weinreich M., Toure O., Wei M.H., Schmidt L.S., Davis L., Zbar B.. Lung cysts, spontaneous pneumothorax, and genetic associations in 89 families with Birt-Hogg-Dube syndrome. Am. J. Respir. Crit. Care Med. (2007) 175 1044-1053. PMID: 17322109 28. Kluger N., Giraud S., Coupier I., Avril M.F., Dereure O., Guillot B., Richard S., Bessis D.. Birt-Hogg-Dube syndrome: clinical and genetic studies of 10 French families. Br. J. Dermatol. (2010) 162 527-537. PMID: 19785621 29. Liu Y., Xu Z., Feng R., Zhan Y., Wang J., Li G., Li X., Zhang W., Hu X., Tian X., Xu K.F., Zhang X.. Clinical and genetic characteristics of Chinese patients with Birt-Hogg-Dube syndrome. Orphanet J. Rare Dis. (2017) 12 104. PMID: 28558743 30. Liu K., Xu W., Tian X., Xiao M., Zhao X., Zhang Q., Qu T., Song J., Liu Y., Xu K.F., Zhang X.. Genotypic characteristics of Chinese patients with BHD syndrome and functional analysis of FLCN variants. Orphanet J. Rare Dis. (2019) 14 223. PMID: 31615547 31. Lemaitre M., Chevalier B., Jannin A., Bourry J., Espiard S., Vantyghem M.C.. Multiple symmetric and multiple familial lipomatosis. Presse Med. (2021) 50 104077. PMID: 34687914 32. Liaw D., Marsh D.J., Li J., Dahia P.L., Wang S.I., Zheng Z., Bose S., Call K.M., Tsou H.C., Peacocke M.. Germline mutations of the PTEN gene in Cowden disease, an inherited breast and thyroid cancer syndrome. Nat. Genet. (1997) 16 64-67. PMID: 9140396 33. Hanssen A.M., Fryns J.P.. Cowden syndrome. J. Med. Genet. (1995) 32 117-119. PMID: 7760320 34. Pavlovich C.P., Grubb R.L., Hurley K., Glenn G.M., Toro J., Schmidt L.S., Torres-Cabala C., Merino M.J., Zbar B., Choyke P.. Evaluation and management of renal tumors in the Birt-Hogg-Dube syndrome. J. Urol. (2005) 173 1482-1486. PMID: 15821464 35. Benusiglio P.R., Giraud S., Deveaux S., Mejean A., Correas J.M., Joly D., Timsit M.O., Ferlicot S., Verkarre V., Abadie C.. Renal cell tumour characteristics in patients with the Birt-Hogg-Dube cancer susceptibility syndrome: a retrospective, multicentre study. Orphanet J. Rare Dis. (2014) 9 163. PMID: 25519458 36. Furuya M., Tanaka R., Okudela K., Nakamura S., Yoshioka H., Tsuzuki T., Shibuya R., Yatera K., Shirasaki H., Sudo Y.. Pulmonary neoplasms in patients with Birt-Hogg-Dube syndrome: histopathological features and genetic and somatic events. PLoS One (2016) 11 e0151476. PMID: 26974543 37. Furuya M., Hasumi H., Yao M., Nagashima Y.. Birt-Hogg-Dube syndrome-associated renal cell carcinoma: histopathological features and diagnostic conundrum. Cancer Sci. (2020) 111 15-22. PMID: 31777168 38. Vincent A., Farley M., Chan E., James W.D.. Birt-Hogg-Dube syndrome: a review of the literature and the differential diagnosis of firm facial papules. J. Am. Acad. Dermatol. (2003) 49 698-705. PMID: 14512919 39. Schulz T., Hartschuh W.. Birt-Hogg-Dube syndrome and Hornstein-Knickenberg syndrome are the same. Different sectioning technique as the cause of different histology. J. Cutan. Pathol. (1999) 26 55-61. PMID: 10189247 40. Collins G.L., Somach S., Morgan M.B.. Histomorphologic and immunophenotypic analysis of fibrofolliculomas and trichodiscomas in Birt-Hogg-Dube syndrome and sporadic disease. J. Cutan. Pathol. (2002) 29 529-533. PMID: 12358810 41. De la Torre C., Ocampo C., Doval I.G., Losada A., Cruces M.J.. Acrochordons are not a component of the Birt-Hogg-Dube syndrome: does this syndrome exist? Case reports and review of the literature. Am. J. Dermatopathol. (1999) 21 369-374. PMID: 10446780 42. Sorrentino A., Federico A., Rienzo M., Gazzerro P., Bifulco M., Ciccodicola A., Casamassimi A., Abbondanza C.. PR/SET domain family and cancer: novel insights from the cancer genome atlas. Int. J. Mol. Sci. (2018) 19 43. Huang S., Shao G., Liu L.. The PR domain of the Rb-binding zinc finger protein RIZ1 is a protein binding interface and is related to the SET domain functioning in chromatin-mediated gene expression. J. Biol. Chem. (1998) 273 15933-15939. PMID: 9632640 44. Di Tullio F., Schwarz M., Zorgati H., Mzoughi S., Guccione E.. The duality of PRDM proteins: epigenetic and structural perspectives. FEBS J. (2021) 289 1256-1275. PMID: 33774927 45. Fagerberg L., Hallstrom B.M., Oksvold P., Kampf C., Djureinovic D., Odeberg J., Habuka M., Tahmasebpoor S., Danielsson A., Edlund K.. Analysis of the human tissue-specific expression by genome-wide integration of transcriptomics and antibody-based proteomics. Mol. Cell. Proteomics (2014) 13 397-406. PMID: 24309898 46. 46. www.proteinatlas.org/ENSG00000170325-PRDM10 (v21.0). 47. Uhlen M., Zhang C., Lee S., Sjostedt E., Fagerberg L., Bidkhori G., Benfeitas R., Arif M., Liu Z., Edfors F.. A pathology atlas of the human cancer transcriptome. Science (2017) 357 48. Goncharova E.A., Goncharov D.A., James M.L., Atochina-Vasserman E.N., Stepanova V., Hong S.B., Li H., Gonzales L., Baba M., Linehan W.M.. Folliculin controls lung alveolar enlargement and epithelial cell survival through E-cadherin, LKB1, and AMPK. Cell Rep. (2014) 7 412-423. PMID: 24726356 49. Fujimoto Y., Nakagawa Y., Satoh A., Okuda K., Shingyouchi A., Naka A., Matsuzaka T., Iwasaki H., Kobayashi K., Yahagi N.. TFE3 controls lipid metabolism in adipose tissue of male mice by suppressing lipolysis and thermogenesis. Endocrinology (2013) 154 3577-3588. PMID: 23885019 50. Pastore N., Vainshtein A., Klisch T.J., Armani A., Huynh T., Herz N.J., Polishchuk E.V., Sandri M., Ballabio A.. TFE3 regulates whole-body energy metabolism in cooperation with TFEB. EMBO Mol. Med. (2017) 9 605-621. PMID: 28283651 51. Salma N., Song J.S., Kawakami A., Devi S.P., Khaled M., Cacicedo J.M., Fisher D.E.. Tfe3 and Tfeb transcriptionally regulate peroxisome proliferator-activated receptor gamma2 expression in adipocytes and mediate adiponectin and glucose levels in mice. Mol. Cell. Biol. (2017) 37 52. Gong X.M., Li Y.F., Luo J., Wang J.Q., Wei J., Wang J.Q., Xiao T., Xie C., Hong J., Ning G.. Gpnmb secreted from liver promotes lipogenesis in white adipose tissue and aggravates obesity and insulin resistance. Nat. Metab. (2019) 1 570-583. PMID: 32694855 53. Dickinson M.E., Flenniken A.M., Ji X., Teboul L., Wong M.D., White J.K., Meehan T.F., Weninger W.J., Westerberg H., Adissu H.. High-throughput discovery of novel developmental phenotypes. Nature (2016) 537 508-514. PMID: 27626380 54. Casamassimi A., Rienzo M., Di Zazzo E., Sorrentino A., Fiore D., Proto M.C., Moncharmont B., Gazzerro P., Bifulco M., Abbondanza C.. Multifaceted role of PRDM proteins in human cancer. Int. J. Mol. Sci. (2020) 21 55. Hofvander J., Tayebwa J., Nilsson J., Magnusson L., Brosjo O., Larsson O., Vult von Steyern F., Mandahl N., Fletcher C.D., Mertens F.. Recurrent PRDM10 gene fusions in undifferentiated pleomorphic sarcoma. Clin. Cancer Res. (2015) 21 864-869. PMID: 25516889 56. Puls F., Pillay N., Fagman H., Palin-Masreliez A., Amary F., Hansson M., Kindblom L.G., McCulloch T.A., Meligonis G., Muc R.. PRDM10-rearranged soft tissue tumor: a clinicopathologic study of 9 cases. Am. J. Surg. Pathol. (2019) 43 504-513. PMID: 30570551 57. Chen N., Hu T., Gui Y., Gao J., Li Z., Huang S.. Transcriptional regulation of Bcl-2 gene by the PR/SET domain family member PRDM10. PeerJ (2019) 7 e6941. PMID: 31143550 58. Nayman A.H., Siginc H., Zemheri E., Yencilek F., Yildirim A., Telci D.. Dual-inhibition of mTOR and Bcl-2 enhances the anti-tumor effect of Everolimus against renal cell carcinoma in vitro and in vivo. J. Cancer (2019) 10 1466-1478. PMID: 31031856 59. Yin P., Jia J., Li J., Song Y., Zhang Y., Chen F.. ABT-737, a Bcl-2 selective inhibitor, and chloroquine synergistically kill renal cancer cells. Oncol. Res. (2016) 24 65-72. PMID: 27178823 60. Zhang L., Huang Y., Ling J., Zhuo W., Yu Z., Shao M., Luo Y., Zhu Y.. Screening and function analysis of hub genes and pathways in hepatocellular carcinoma via bioinformatics approaches. Cancer Biomark. (2018) 22 511-521. PMID: 29843214 61. Dreos R., Ambrosini G., Perier R.C., Bucher P.. The Eukaryotic Promoter Database: expansion of EPDnew and new promoter analysis tools. Nucleic Acids Res. (2015) 43 D92-D96. PMID: 25378343 62. Dreos R., Ambrosini G., Groux R., Cavin Perier R., Bucher P.. The eukaryotic promoter database in its 30th year: focus on non-vertebrate organisms. Nucleic Acids Res. (2017) 45 D51-D55. PMID: 27899657 63. Kapferer-Seebacher I., Waisfisz Q., Boesch S., Bronk M., Tintelen P., Gizewski E.R., Groebner R., Zschocke J., Knaap M.S.. Periodontal Ehlers-Danlos syndrome is associated with leukoencephalopathy. Neurogenetics (2019) 20 1-8. PMID: 30535813 64. Silberstein M., Weissbrod O., Otten L., Tzemach A., Anisenia A., Shtark O., Tuberg D., Galfrin E., Gannon I., Shalata A.. A system for exact and approximate genetic linkage analysis of SNP data in large pedigrees. Bioinformatics (2013) 29 197-205. PMID: 23162081 65. Reed P.W., Davies J.L., Copeman J.B., Bennett S.T., Palmer S.M., Pritchard L.E., Gough S.C., Kawaguchi Y., Cordell H.J., Balfour K.M.. Chromosome-specific microsatellite sets for fluorescence-based, semi-automated genome mapping. Nat. Genet. (1994) 7 390-395. PMID: 7920657 66. Anzalone A.V., Koblan L.W., Liu D.R.. Genome editing with CRISPR-Cas nucleases, base editors, transposases and prime editors. Nat. Biotechnol. (2020) 38 824-844. PMID: 32572269 67. 67. Joshi, N.A., Fass J. (2011) Sickle: A sliding-window, adaptive, quality-based trimming tool for FastQ files (Version 1.33) [Software].. Sickle: A sliding-window, adaptive, quality-based trimming tool for FastQ files (Version 1.33) [Software] (2011) 68. Kim D., Langmead B., Salzberg S.L.. HISAT: a fast spliced aligner with low memory requirements. Nat. Methods (2015) 12 357-360. PMID: 25751142 69. Liao Y., Smyth G.K., Shi W.. featureCounts: an efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics (2014) 30 923-930. PMID: 24227677 70. 70. Illumina . (Retrieved Nov 2021), in press.. (Retrieved Nov 2021) 71. Robinson M.D., McCarthy D.J., Smyth G.K.. edgeR: a Bioconductor package for differential expression analysis of digital gene expression data. Bioinformatics (2010) 26 139-140. PMID: 19910308 72. Benjamini Y., Hochberg Y.. Controlling the false discovery rate - a practical and powerful approach to multiple testing. J. R. Stat. Soc. Ser. B (1995) 57 289-300 73. Goedhart J., Luijsterburg M.S.. VolcaNoseR is a web app for creating, exploring, labeling and sharing volcano plots. Sci. Rep. (2020) 10 20560. PMID: 33239692
--- title: Anti‐cancer effect of entacaponeon esophageal cancer cells via apoptosis induction and cell cycle modulation authors: - Fahimeh Ramedani - Seyyed Mehdi Jafari - Marie Saghaeian Jazi - Zeinab Mohammadi - Jahanbakhsh Asadi journal: Cancer Reports year: 2022 pmcid: PMC10026269 doi: 10.1002/cnr2.1759 license: CC BY 4.0 --- # Anti‐cancer effect of entacaponeon esophageal cancer cells via apoptosis induction and cell cycle modulation ## Abstract ### Background Esophageal cancer (EC) is the sixth leading cause of cancer‐related death, despite many advances in treatment, the survival of patients still remains poor. In recent years, the N6‐methyladenosine (m6A) has been introduced as one of the most important modifications at the epitranscriptome level, with an important role in the mRNA regulation in various diseases, such as cancers. The m6A is regulated by different factors, including FTO as a demethylase. The m6A modification, especially through FTO overexpression has an oncogenic role in different cancer types such as EC. Recent studies showed that entacapone, a catechol‐o‐methyl transferase (COMT) inhibitor currently applied for Parkinson's disease, can inhibit FTO enzyme. ### Aims In this study, we aimed to investigate the effect of entacapone as an FTO inhibitor on the m6A level and also apoptosis and cell cycle response in KYSE‐30 and YM‐1 of esophageal squamous cancer cell (ESCC) lines. ### Methods Cell toxicity and IC50 of entacapone were evaluated using The MTT assay in YM‐1 and KYSE‐30 cells. Cells were treated into two groups: DMSO (control) and entacapone (mean IC50). Total RNA was extracted, and m6A levels were measured via the ELISA method. Subsequently, the apoptosis and cell cycle dys‐regulation were detected by annexin‐V‐FITC/PI staining and PI staining via flow cytometry. ### Results Entacapone has the cytotoxicity effect on both esophageal cancer cell lines compared to normal PBMC cells. As well, entacapone treatment (140 μM) can induce apoptosis (KYSE‐30: $50\%$. YM‐1:$22.6\%$) and has a modulatory effect on cell cycle progression in both YM‐1 and KYSE‐30 cells (p‐value<.05). However, no significant difference in the m6A concentration was observed. ### Conclusion Our findings suggested that entacapone has the inhibitory effect on ESCC cell lines through induction of the apoptosis and modulation of the cell cycle without toxicity on the normal PBMC. ## INTRODUCTION Esophageal cancer (EC) is the eighth most common malignancy and sixth leading cause of cancer‐related death worldwide. 1 Esophageal cancer is divided into two common histologic subtypes: esophageal adenocarcinoma (EAC) and esophageal squamous cell carcinoma (ESCC). 2 In Asia, ESCC is the most frequent histologic subtype of esophageal cancer. 3, 4, 5 In recent years, despite of the great improvements in treatment of ESCC patients, the prognosis of patients remains poor and overall survival rate is still low. 2, 6 Cancer is considered as a complex disease and different environmental, genetic and epigenetic factors can play role in cancer development or progression. More recently, epitranscriptome changes particularly with N6‐methyladenine (m6A) modification, has been reported with association to multiple diseases, such as cancer. 7, 8, 9 The m6A is the common chemical modification of messenger RNA (mRNA), involved in mRNA splicing, stability, transcription, splicing, localization, translation regulation process. 10 The level of m6A modification in mRNAs is regulated by multiple proteins called, “writer,” “eraser”, and “reader” proteins. 11, 12 The writers are consisted of methyl‐transferase complex, METTL3, METTL14 and WTAP. 13 In contrast the eraser proteins, namely FTO and ALKBH5 have the role of m6A demethylation. 13 The Reader protein family includes the YTH domain‐containing family proteins YTHDF$\frac{1}{2}$/3, YTHDC1, and YTHDC2 in which the YTH domain is responsible for recognition and binding to the m6A sites. 14, 15 The previous studies have suggested that overexpression of the FTO has an oncogenic role in different cancer types 16 such as acute myeloid leukemia, 17 gastric cancer, 18 cervical squamous cell carcinoma, 19 and ESCC. 20 It has been reported that down‐regulation of the FTO expression inhibits cell proliferation, migration and invasion abilities of ESCC cells. 20 Recent studies have established, entacapone, a catechol‐o‐methyl transferase (COMT) inhibitor, can compete for the binding to the cofactor‐ and substrate binding sites on FTO and thereby can inhibit FTO activity. 21 The entacapone, is the FDA approved drug which is currently in use for treatment of the Parkinson's disease in combination with levodopa. 22 In this study, we aimed to investigate the potential effect of entacapone on m6a concentration and also its anti‐cancer effect on the cell viability and apoptosis of two different esophageal cancer cell lines, YM‐1 and KYSE‐30, in vitro. ## Reagent and cell line In this study two ESCC cell lines were used. The YM‐1 cell line, which was previously established in the lab of Dr. Jahanbakhsh Asadi (Metabolic Disorders Research Center; Golestan University of Medical Sciences, Gorgan, Iran) from an ESCC female patient. 23 The other cell line, KYSE‐30, was obtained from the cell bank of the Pasture Institute (Tehran, Iran). The cell culture reagents including the RPMI1640 culture medium, streptomycin/penicillin, EDTA Trypsin were purchased from Gibco, Life Technologies. Inc. The entacapone was provided as a gift from the Pharma Manufacturing at Fanda Pharma (Cat. No: ENT/$\frac{516}{06}$/20M;Rechem Lab). The m6A Elisa Kit (Cat. No:ZB‐15178c‐H9648; Zell Bio GmbH) and Nuclease P1 (Cat. No: Mo660s; BioLabs) and Alkaline phosphatase (P0114; sigmaaldrich) were purchased as indicated. annexin V/PI apoptosis detection kit was obtained from Mab Tag, GmbH. ## Cell culture In the present study, the two cell lines of human ESCC, YM‐1 and KYSE‐30, were cultured in RPMI‐1640 medium supplemented with $10\%$ FBS and $1\%$ streptomycin/penicillin. Cells were kept in a humidified incubator containing $5\%$ CO2, at 37°C. ## PBMC Isolation The peripheral blood mononuclear cells isolation was performed using the ficoll reagent. Thus, 10 ml of peripheral blood containing anticoagulants was collected from a healthy person. The blood was diluted with equal volume of PBS and then was added to ficoll, in two separate layers. Then, to separate the PBMC containing layer, solution was centrifuged at 2500 rpm for 30 min. cells were rinsed by PBS. The peripheral blood mononuclear cells (PBMC) then were suspended in 1 ml of RPMI supplemented with $10\%$ FBS and $1\%$ streptomycin/penicillin and used for next experiments. ## Viability assay The toxicity of entacapone was evaluated in 48 h at different doses, in both esophageal squamous carcinoma cell lines, YM1 and KYSE 30, using MTT assay. Cells were seeded into 96‐well plates at a density of 1 × 104 cells/well after 24 incubations, the culture medium was removed and then then cells were treated with different concentrations (25, 50, 75, 100, 125, and 150 μM) or IC50 (140 μM) of entacapone with a final volume of 200 μl. As vehicle control cells were treated with DMSO ($0.3\%$) with same condition. After 48 h treatment, 10 λ of MTT assay (10 mg/mL) solution, (3‐(4,5‐dimethylthiazol‐2‐yl)‐2,5‐diphenyltetrazolium bromide), was added to each well and incubated at 37°C for 4 h. Then, the supernatant was aspirated off and 100 μl of DMSO was added to solve the precipitates. The absorbance of each well was measured at wavelength 570 nm using a micro‐plate reader. Also the viability was checked with dye exclusion assay using Trypan blue dye. Then the viable cell percentage was calculated as (trypan blue negative cell/total cell) × 100. ## Apoptosis assay To perform the apoptosis assay, 4 × 105 cells were seeded in 6‐well plates. After 24 h cells were treated with $0.3\%$ DMSO (control) and 140 μM entacapone for 48 h. Cells were stained by annexin‐V and propidium iodide (PI) at room temperature (RT) for 15 min, according to the manufacturer's protocol. The annexin V binds to phosphatidylserine (PS) exposed to the plasma membrane of cells undergoing apoptosis. This feature allows the living cell discriminated from early (stained only with annexin V) and late (stained with annexin V and PI) apoptotic cells. Briefly, 104–106 cells were re‐suspended in 90 μl of diluted binding buffer. After addition of 5 μl annexin V and 5 μl PI, the cell suspension was incubated for 20 min in the dark. Subsequently, 400 μl diluted binding buffer was added and then the supernatant was removed following centrifugation at 400g for 5 min. Finally the stained cells were re‐suspended in 200 μl diluted binding buffer and were analyzed by flow cytometry. The non‐stained cells also were used to distract the background in flow cytometer instrument. We counted the events for the living (annexin V−/PI−), necrotic (annexin V−/PI+), early apoptotic (annexin V+/PI−), and late apoptotic (annexin V+/PI+) cells using a BD Accuri™ C6 flow cytometer. The results were analyzed using the software supplied in the instrument. ## RNA extraction Total RNA was isolated using the RNX‐PLUSKit (SINACLON, Iran) according to the manufacturer's protocol. The number of 3 × 104 cells of YM‐1 or KYSE‐30 cells were cultured in the flask. After 24 h incubation, cells were treated with DMSO ($0.3\%$ as control) or entacapone (IC50 = 140 μM) for 48 h. Then, cells were harvest with trypsin solution and washed twice with PBS. To extract the RNA, briefly, cells were disrupted and homogenized with RNX‐PLUS reagent at room temperature (RT) for 5 min. Then, chloroform was added to the samples and shacked vigorously for 5–10 s, then incubated for 15 min. Subsequently, using centrifugation (12 000 rpm at 4°C for 15 min), the upper aqueous phase containing the nucleic acids was isolated. Then, it was precipitated by adding equal volume of isopropanol followed with incubation for 15 min on ice and centrifugation. The RNA was washed twice by adding 1 ml $75\%$ ethanol, and centrifugation at 7500 rpm at 4°C for 8 min. Finally, the purified RNA samples were dissolved in RNase‐free water and stored at −80°C. ## Measurement of m6a by ELISA method The level of m6A modification in RNA samples, was measured by m6 A Elisa Kit (Zellbio, Germany), following the manufacturer's protocol. Briefly, a total amount of 5 μg RNA for each group was used to determine the amount of m6A. To remove any RNA secondary structure, RNA samples were heated at 95°C for 5 min, and then was rapidly chilled on ice. The RNA was digested to nucleosides; by incubating the denatured RNA with 5 unit of nuclease P1 for 1 h at 37°C. Subsequently, 5 units of alkaline phosphatase plus sufficient of tris buffer to a final concentration of 100 mM Tris, pH 7.5 was added, and then was incubated for 1 h at 37°C. The supernatant was collected and was used for the ELISA experiment. The absorbance was measured at the wavelength 450 nm using a microplate reader. The standard curve was constructed to calculate the concentration of the samples. ## Cell cycle assay The cell cycle analysis was performed using flow cytometer with PI staining method. Labeling DNA with PI allows for fluorescence‐based analysis of cell cycle according the DNA content of each cell in different cell cycle phase (G1:2N, S:2N‐4N, G2/$M = 4$N). The number of 106 cells were seeded in 25 cm2 flasks and allowed to attach for 24 h. Cells were then treated with DMSO ($0.3\%$) and entacapone (140 μM) for 48 h. The cells were detached by trypsin and then fixed with $70\%$ ethanol for 1 h. The cells were washed twice with PBS and then were incubated with stain solution containing 10 μg/ml propidium iodide and 40 μg/ml RNase in phosphate‐buffered saline and triton X‐100 ($0.1\%$) for 30 min at 37°C. The percentages of cells in different phases of the cell cycle (G0/G1, S, and G2/M) can be quantified by BD Accuri™ C6 flow cytometry. The results were analyzed using the software supplied in the instrument. ## Statistics In this study the p‐value <.05 was considered as significant level. All the experiments were carried out in replicates ($$n = 3$$–5) for each group. For data visualization, the Graph Pad Prism v.5.04 was used and the data was reported as mean ± SE. for statistical analysis, SPSS v.19 was used. The normality of data was checked with the Shapirowik test and then in the case of normal distribution one‐way Anova test was used. In the case of non‐normal distribution the non‐parametric test was used for data analysis. ## Entacapone has selective significant cytotoxicity in ESCC To evaluate the toxicity of entacapone in esophageal cancer cell lines (YM‐1 and KYSE30), cell viability was measured by MTT staining method for different concentrations. The percentage of cell viability after 48 h of treatment is shown in Figure 1A. Analysis of MTT data showed that different concentrations of entacapone inhibited cancer cell viability in both cell lines in comparison to the control (DMSO $0.3\%$). As shown in Figure 1A, the cytotoxicity of entacapone on YM‐1 and KYSE‐30 cells was started from 75 to 50 μM concentration, respectively. Our results showed a dose dependent cytotoxicity in both cell lines with declines cellular viability in higher concentrations. After 48 h of treatment, the IC 50 value was calculated for YM‐1 (IC50 = 149 μM, y = −0.3558x + 103.99, R 2 = 0.9404) and KYSE‐30 (IC50 = 131.8 μM, y = −0.3347x + 93.592, R 2 = 0.9085) cell lines. For the next experiments, the mean IC50 value of two cell lines (140 μM) was used. When comparing the two cell lines together, the cytotoxic effect of entacapone was almost the same in both cell lines, however the KYSE‐30 showed more sensitivity in general with a significant difference with YM‐1 in concentration of 50 μM (Figure 1A). As a normal cell control, the PBMC, isolated from healthy individual was used. The viability was evaluated at a dose of 140 μM of entacapone after 48 h (mean IC50 for cancer cells). Our findings illustrated that the mean survival of PBMC cells treated with entacapone (Mean ± SE:100 ± 1.39) was not significantly different (p‐value:.693) from control group(Mean ± SE: 95.70 ± 2.56),indicating the selective toxicity of entacapone in cancer cells (Figure 1B). The dye exclusion assay using trypan blue also indicated selective toxicity of entacapone (140 μM) on YM‐1 (p‐value =.004) and KYSE‐30 cells (p‐value =.008) but no toxicity in PBMC was detected (Figure 1C). FIGURE 1: The cellular viability of two ESCC cancer cell lines (YM1 and KYSE‐30) and PBMC after entacapone treatment. Different concentrations of entacapone were used for treating cancer cells for 48 h, and the viability was measured using MTT. The cellular viability of YM1 and KYSE30 was compared to the control for each concentration (N = 3 in each concentration) (A). The viability of PBMC cells isolated from healthy as normal cells were also calculated (N = 18 replicates for control, N = 10 replicates for entacapone) (B). The viability of the cells was measured using trypan blue exclusion assay in control (DMSO 0.3%) and entacapone (140 μM) cells (N = 3 in each group) (C). Represents the p‐value comparing to control which was calculated using the student t‐test. # shows p‐value < .01 YM‐1 vs KYSE‐30 at 50 μM entacapone* ## Entacapone treatment induces apoptotic cell death in ESCC The programmed cell death was measured using annexinV/PI staining protocol and flow cytometry technique. As shown in Figure 2A, B, the percentage of positive annexinV+/PI+ cells and positive PI cells treated with entacapone was significantly higher than the control in both cell lines. The entacapone was found to significantly increase the number of YM1 cells in the apoptotic phase (up to mean ± SE: 22.65 ± 3.25, p‐value =.02), as well as necrotic cells (up to mean ± SE = 17.55 ± 0.85, p‐value =.002) at mean IC50 concentration. Also, in the KYSE‐30 cell line, treatment with 140 μM entacapone resulted to increase in the apoptotic cells including both early and late apoptosis (up to mean% ± SE: 50.05 ± 1.55, p‐value =.009), as well as necrotic cells (up to mean% ± SE: 8.05 ± 0.35, p‐value =.002). There was no significantly higher number of early apoptotic phase at mean IC50 (140 μM) in KYSE‐30 versus control, while the number of cells in the early apoptotic phase increased significantly in YM‐1 (mean% ± SE: 3.8 ± 0.2 vs. 0.7 ± 0.10, p‐value =.0052. data not shown). FIGURE 2: The apoptotic cell death in YM1 (A) and KYSE 30 (B) cell lines with entacapone treatment. The graph shows the dot plot of annexin (FL1‐A) and PI (FL2‐A) positive cells. The percentage of the live cells (double negative), apoptotic cells (early and late apoptotic), and necrotic cells (PI positive) were visualized in the bar chart for each cell line. Each bar represents mean of N = 3 replicates ## Measurement of m6A concentration in ESCC cell lines treated with entacapone The entacapone previously was introduced as inhibitor of FTO demethylase which removes the methyl group from m6A, then we hypothesized it may increase the m6A level as substrate of FTO. To test this hypothesis, the concentration of m6A in total RNA samples extracted from both cell lines was measured using ELISA in entacapone treated in comparison to the control cells. We found a small increase in m6A level in entacapone treated ESCC; however the change was not statistically significant (Figure 3). FIGURE 3: The concentration of m6A in total RNA of ESCC cells. Each bar represents the mean ± SE and ns: p‐value > .05. The control cell was treated with the DMSO 0.3% as vehicle control. Each bar represents mean of N = 3 replicates ## Entacapone treatment changes the cell cycle progression in ESCC For more investigation, the ESCC cells were stained for cell cycle using the PI and the cell cycle progression was measured by flowcytometery. Or findings showed that the entacapone treatment decreasescell in G1 phase (YM1:56.40 ± $2.36\%$ vs. 86.95 ± $0.25\%$, p‐value =.005; KYSE30: 71.25 ± $0.20\%$ vs. 82.35 ± $1.47\%$, p‐value =.016) while increasing the percentage of cells in S phase (YM1: 13.40 ± $0.46\%$ vs. 4.35 ± $0.25\%$, p‐value =.000; KYSE30: 9.45 ± $0.086\%$ vs. 3.95 ± $0.49\%$, p‐value = 0.007) and G2M phases of cell cycle in both cell lines (YM1:22.46 ± $0.77\%$ vs. 7.05 ± $0.14\%$, p‐value = 0.002; KYSE30: 16.10 ± $0.23\%$ vs. 9.95 ± $1.58\%$, p‐value =.058). This result shows a modulatory effect in cell cycle progression increasing cancer cells in S/G2M phase of cell cycle in both YM1 and KYSE 30 cells in vitro (Figure 4). FIGURE 4: The cell cycle progression (G1, S, and G2/M) of YM‐1 and KYSE‐30 cells treated with entacapone. Each bar represents the mean ± SE and shows the p‐value calculated using the student t‐test. Cells were treated with a mean IC50 concentration of entacapone (140 μM) for 48 h, and the control cell was treated with the DMSO 0.3% as vehicle control. Each bar represents mean of N = 3 replicates* ## DISCUSSION The esophagus squamous cell carcinoma is one of the most common types of esophageal malignancies with aggressive nature 24 and poor survival rate. 20 There are many factors playing role in ESCC pathogenesis, including genetic and epigenetic, as well as epitranscriptomic changes. 24, 25 The role of epitranscriptome changes in cancer progression of different types has been investigated in recent years. 26 The most common type of the epitranscriptome modification is N6‐methyladenosine (m6A) which can be modulated by methyltransferases METTL$\frac{3}{14}$, demethylases FTO and ALKBH5 and m6A‐binding proteins called YTHDF1‐3. 25, 27 Increased expression of FTO as m6A demethylase has an oncogenic role in various types of cancers including ESCC. In recent years, the use of FO inhibitors as regulator of m6A, has been investigated as new potential therapeutic strategy for cancer. Available pharmaceutical compounds such as meclofenamic acid, nafamostet and eantacapone, in addition to siRNA and chemical compounds, have been introduced as FTO inhibitors. 21, 28, 29, 30 Studies have shown that FTO enhances the proliferation and migration of esophageal cancer cells, and inhibiting FTO with siRNA potentially reduces the cell growth, proliferation, and migration of human esophageal cancer lines including KYSE150, Eca‐109 and TE‐1. 20 In other hand a recent drug virtual screening introduced entacapone as a potential inhibitor of FTO which can directly bind to it and suppress its activity in vitro. 21 Then here in current study we evaluated the potential anti‐cancer properties of the entacapone as FTO inhibitor in ESCC cells. We found a significant dose‐dependent toxicity of entacapone in both ESCC cells (YM‐1 and KYSE‐30) with the mean IC50 of 140 μM after 48 h of treatment without any significant toxicity in normal PBMC. Similar to our findings; Grimes et al. in 2018, showed the toxicity of entacapone alone (IC50 = 100 μM) or in combination with anthocyanin (IC50 = 50 μM), in colon (Caco‐2 and HT‐29) and Breast (MDA‐MB‐231) cancer cell lines after 72 h of treatment. 31 In another study, Forester et al. in 2014, investigated the simultaneous effect of entacapone, tolcapone and epigallocatechin‐3‐gallate (EGCG) in human lung (H1299) and mouse lung (CL‐13) cancer lines. They reported the inhibitory concentration of IC50 = 76.8 μM and 50.7 μM after 72 h for the two cell lines respectively. 32 The difference in IC50 value between their study and ours result could be related to the time of exposure and also the difference in cancer cell lines. Our findings indicated that the $50\%$ cellular death found in ESCC cells in response to the IC50 of the entacapone is through apoptosis with $22.65\%$ ± 3.25 and $50.05\%$ ± 1.55 of annexin/PI positive cells in YM‐1 and KYSE‐30 cell lines. Other studies also reported apoptotic cell death in acute myeloid leukemia cells treated with FTO shRNA 17 and GC1 spermatogonia cell line treated with meclofenamic acid as FTO inhibitor. 33 Regarding the inhibitory effect of entacapone on FTO we expected to observe significant increase in m6A as substrate of FTO, however unexpectedly our results showed a minor non‐significant increase in m6A of total RNA from entacapone treated ESCC (mean difference YM‐1 = +6 ng/ml and KYSE‐30 = +8 ng/mL). This could be explained by the complexity of m6A regulation through different regulators including demethylase, methyl transferases and also m6A‐binding proteins. Although overexpression of FTO demethylasehas been reported in esophagus tumor in associated with lower survival rate and poorer prognosis 20; but overexpression of METTL3 methyl transferases, 34 and ALKBH5 as another m6A demethylase 35 is also reported to be associated with poor prognosis in esophageal tumors. So the other m6A modulators may potentially affect the m6A level compensating the effect of entacapone in FTO demethylase inhibition. Also the cytotoxic effect of entacapone can be related to other mechanisms independent to FTO or m6A modulation. Moreover we found a significant change in the cell cycle progression of entacapone treated ESCC cells by decreased G1 and S, G2/M increase. Nagaki et al reported a cell cycle delay with G0/G1 arrest in ALKBH5 siRNA treated ESCC cell lines through up‐regulation of the CDKN1A (p21) consequent to increased m6A and stability of its mRNA. 35 In another study it was reported that entacapone or tolcapone in combination with EGCG can reduce expression of cyclin D1 and consequently, resulting to G1 arrest induction in H1299 cells, whereas in CL‐13 lung cancer cells they found significant G2/M arrest, which was in parallel to our findings. 32 Similar studies support the role of the FTO in cell cycle progression through regulation of cyclin D1 m6A modification. It has been reported that FTO inhibition can prolong G1 phase by cyclin D1 suppression. 36 Altogether although there are various effects of FTO siRNA, or chemical compounds in cell cycle progression in different cell lines, but all of these share common outcome in general in cell cycle inhibition. ## CONCLUSION Our findings indicated selective anti‐cancer effect of entacapone in esophageal cancer cells in vitro through apoptosis induction and cell cycle regulation. These results suggest potential application of entacapone in ESCC treatment however further investigations to discover the underlying mechanisms are needed. ## AUTHOR CONTRIBUTIONS Fahimeh Ramedani: Investigation (equal); writing – original draft (equal); writing – original draft (equal). seyyed mehdi Jafari: Investigation (equal); supervision (equal). Marie Saghaeian Jazi: Investigation (equal); methodology (equal); writing – review and editing (equal). Zeinab Mohammadi: Investigation (equal); writing – review and editing (equal). Jahanbakhsh Asadi: Project administration (equal); supervision (equal); validation (equal). ## FUNDING INFORMATION This project was extracted from a master student thesis which was financially supported by Golestan University of Medical Sciences (grant number: 111290). ## CONFLICT OF INTEREST The authors declare that there are no conflicts of interests. ## ETHICS STATEMENT This project was approved by ethical committee of Golestan University of Medical Sciences (Approval code: IR.GOUMS.REC.1398.357). ## DATA AVAILABILITY STATEMENT The data that support the findings of this study are available from the corresponding author upon reasonable request. ## References 1. Domper Arnal MJ, Ferrández Arenas Á, Lanas AÁ. Esophageal cancer: risk factors, screening and endoscopic treatment in Western and eastern countries. World J Gastroenterol (2015) 21 7933-7943. PMID: 26185366 2. Uhlenhopp DJ, Then EO, Sunkara T, Gaduputi V. Epidemiology of esophageal cancer: update in global trends, etiology and risk factors. Clin. J Gastroenterol (2020) 13 1010-1021 3. Qin B, Dong M, Wang Z. Long non‐coding RNA CASC15 facilitates esophageal squamous cell carcinoma tumorigenesis via decreasing SIM2 stability via FTO‐mediated demethylation. Oncol Rep (2021) 45 1059-1071. PMID: 33650646 4. Zhao H, Xu Y, Xie Y. m6A regulators is differently expressed and correlated with immune response of esophageal cancer. Front cell. Dev Biol (2021) 9 5. Chen P, Li S, Zhang K. N(6)‐methyladenosine demethylase ALKBH5 suppresses malignancy of esophageal cancer by regulating microRNA biogenesis and RAI1 expression. Oncogene (2021) 40 5600-5612. PMID: 34312488 6. Kato H, Nakajima M. Treatments for esophageal cancer: a review. Gen Thorac Cardiovasc Surg (2013) 61 330-335. PMID: 23568356 7. Taketo K, Konno M, Asai A. The epitranscriptome m6A writer METTL3 promotes chemo‐ and radioresistance in pancreatic cancer cells. Int J Oncol (2018) 52 621-629. PMID: 29345285 8. Dai D, Wang H, Zhu L, Jin H, Wang X. N6‐methyladenosine links RNA metabolism to cancer progression. Cell Death Dis (2018) 9 124. PMID: 29374143 9. Sun T, Wu R, Ming L. The role of m6A RNA methylation in cancer. Biomed Pharmacother (2019) 112 108613. PMID: 30784918 10. Liu ZX, Li LM, Sun HL, Liu SM. Link between m6A modification and cancers. Front Bioeng Biotechnol (2018) 6 89. PMID: 30062093 11. Chen H, Zhao T, Sun D, Wu M, Zhang Z. Changes of RNA N(6)‐methyladenosine in the hormesis effect induced by arsenite on human keratinocyte cells. Toxicol In Vitro (2019) 56 84-92. PMID: 30654086 12. Liu X, Liu L, Dong Z. Expression patterns and prognostic value of m(6)A‐related genes in colorectal cancer. Am J Transl Res (2019) 11 3972-3991. PMID: 31396313 13. Yang Y, Hsu PJ, Chen Y‐S, Yang Y‐G. Dynamic transcriptomic m6A decoration: writers, erasers, readers and functions in RNA metabolism. Cell Res (2018) 28 616-624. PMID: 29789545 14. Wu L, Wu D, Ning J, Liu W, Zhang D. Changes of N6‐methyladenosine modulators promote breast cancer progression. BMC Cancer (2019) 19 326. PMID: 30953473 15. Yang Y, Hsu PJ, Chen YS, Yang YG. Dynamic transcriptomic m(6)a decoration: writers, erasers, readers and functions in RNA metabolism. Cell Res (2018) 28 616-624. PMID: 29789545 16. Deng X, Su R, Stanford S, Chen J. Critical enzymatic functions of FTO in obesity and cancer. Front Endocrinol (2018) 9 17. Li Z, Weng H, Su R. FTO plays an oncogenic role in acute myeloid leukemia as a N6‐methyladenosine RNA demethylase. Cancer Cell (2017) 31 127-141. PMID: 28017614 18. Xu D, Shao W, Jiang Y, Wang X, Liu Y, Liu X. FTO expression is associated with the occurrence of gastric cancer and prognosis. Oncol Rep (2017) 38 2285-2292. PMID: 28849183 19. Zhou S, Bai ZL, Xia D. FTO regulates the chemo‐radiotherapy resistance of cervical squamous cell carcinoma (CSCC) by targeting β‐catenin through mRNA demethylation. Mol Carcinog (2018) 57 590-597. PMID: 29315835 20. Liu S, Huang M, Chen Z. FTO promotes cell proliferation and migration in esophageal squamous cell carcinoma through up‐regulation of MMP13. Exp Cell Res (2020) 389. PMID: 32035950 21. Peng S, Xiao W, Ju D. Identification of entacapone as a chemical inhibitor of FTO mediating metabolic regulation through FOXO1. Sci Transl Med (2019) 11 22. Cacabelos R. Parkinson's disease: from pathogenesis to pharmacogenomics. Int J Mol Sci (2017) 18 23. Ayyoob K, Masoud K, Vahideh K, Jahanbakhsh A. Authentication of newly established human esophageal squamous cell carcinoma cell line (YM‐1) using short tandem repeat (STR) profiling method. Tumour Biol (2016) 37 3197-3204. PMID: 26432330 24. Yang N, Ying P, Tian J. Genetic variants in m6A modification genes are associated with esophageal squamous‐cell carcinoma in the Chinese population. Carcinogenesis (2020) 41 761-768. PMID: 32047883 25. Xu LC, Pan JX, Pan HD. Construction and validation of an m6A RNA methylation regulators‐based prognostic signature for esophageal cancer. Cancer Manag Res (2020) 12 5385-5394. PMID: 32753956 26. Barbieri I, Kouzarides T. Role of RNA modifications in cancer. Nature reviews. Cancer (2020) 20 303-322 27. Helm M, Motorin Y. Detecting RNA modifications in the epitranscriptome: predict and validate. Nat Rev Genet (2017) 18 275-291. PMID: 28216634 28. Han X, Wang N, Li J, Wang Y, Wang R, Chang J. Identification of nafamostat mesilate as an inhibitor of the fat mass and obesity‐associated protein (FTO) demethylase activity. Chem Biol Interact (2019) 5 80-84 29. Wang R, Han Z, Liu B. Identification of natural compound Radicicol as a potent FTO inhibitor. Mol Pharm (2018) 15 4092-4098. PMID: 30063141 30. Qiao Y, Zhou B, Zhang M. A novel inhibitor of the obesity‐related protein FTO. Biochemistry (2016) 55 1516-1522. PMID: 26915401 31. Grimes KL, Stuart CM, McCarthy JJ, Kaur B, Cantu EJ, Forester SC. Enhancing the cancer cell growth inhibitory effects of table grape anthocyanins. J Food Sci (2018) 83 2369-2374. PMID: 30070707 32. Forester SC, Lambert JD. Synergistic inhibition of lung cancer cell lines by (−)‐epigallocatechin‐3‐gallate in combination with clinically used nitrocatechol inhibitors of catechol‐O‐methyltransferase. Carcinogenesis (2014) 35 365-372. PMID: 24148818 33. Huang T, Guo J, Lv Y. Meclofenamic acid represses spermatogonial proliferation through modulating m(6)A RNA modification. J Anim Sci Biotechnol (2019) 10 63. PMID: 31333841 34. Chen X, Huang L, Yang T. METTL3 promotes esophageal squamous cell carcinoma metastasis through enhancing GLS2 expression. Front Oncol (2021) 11. PMID: 34094960 35. Nagaki Y, Motoyama S, Yamaguchi T. m6A demethylase ALKBH5 promotes proliferation of esophageal squamous cell carcinoma associated with poor prognosis. Genes Cells (2020) 25 547-561. PMID: 32449584 36. Hirayama M, Wei FY, Chujo T. FTO demethylates cyclin D1 mRNA and controls cell‐cycle progression. Cell Rep (2020) 31. PMID: 32268083
--- title: Screening for obstructive sleep apnoea in post‐treatment cancer patients authors: - Harini Subramanian - Veronika Fuchsova - Elisabeth Elder - Alison Brand - Julie Howle - Anna DeFazio - Graham J. Mann - Terence Amis - Kristina Kairaitis journal: Cancer Reports year: 2022 pmcid: PMC10026305 doi: 10.1002/cnr2.1740 license: CC BY 4.0 --- # Screening for obstructive sleep apnoea in post‐treatment cancer patients ## Abstract ### Background and aims For cancer patients, comorbid obstructive sleep apnea (OSA) poses additional risk to their surgical/anaesthetic outcomes, quality of life, and survival. However, OSA screening is not well‐established in oncology settings. We tested two screening tools (STOP‐Bang questionnaire [SBQ] and the at‐home monitoring device, ApneaLink™Air), for predicting polysomnography (PSG) confirmed OSA in post‐treatment cancer patients. ### Methods Breast ($$n = 56$$), endometrial ($$n = 37$$) and melanoma patients ($$n = 50$$) were recruited from follow‐up clinics at Westmead Hospital (Sydney, Australia). All underwent overnight PSG, 137 completed SBQ, and 99 completed ApneaLink™Air. Positive (PPV) and negative (NPV) predictive values for PSG‐determined moderate‐to‐severe OSA and severe OSA, were calculated using an SBQ threshold ≥3 au and ApneaLink™Air apnoea‐hypopnea index thresholds of ≥10, ≥15 and ≥30 events/h. ### Results Both SBQ and ApneaLink™Air had high NPVs ($92.7\%$ and $85.2\%$–$95.6\%$ respectively) for severe OSA, but NPVs were lower for moderate‐to‐severe OSA ($69.1\%$ and $59.1\%$–$75.5\%$, respectively). PPV for both tools were relatively low (all <$73\%$). Combining both tools did not improve screening performance. ### Conclusions These screening tools may help identify cancer patients without severe OSA, but both are limited in identifying those with moderate‐to‐severe or severe OSA. PSG remains optimal for adequately identifying and managing comorbid OSA in cancer patients. ## INTRODUCTION Obstructive sleep apnea (OSA) has a reported prevalence of $9\%$–$38\%$ for adults. 1 It is characterised by recurrent episodes of complete (apnoea) or partial (hypopnoea) upper airway collapse during sleep 2 resulting in repetitive nocturnal oxygen desaturation events of varying severity and frequency. 2 OSA has numerous recognised health impacts including; [1] reduced cognitive function and higher accident risk linked to daytime somnolence, 3, 4 [2] poorer wellbeing and quality of life associated with OSA symptomology, 5 [3] increased risk for longer‐term cardiovascular consequences, e.g. hypertension, 6 [4] increased risk for stroke 7 and [5] worsened post‐surgical outcomes. 8 For cancer patients, co‐morbid OSA may pose additional risks, since epidemiological studies suggest associations with both higher cancer incidence 9 and increased mortality rates. 10 A number of published studies report relatively high prevalence rates for moderate‐to‐severe OSA (AHI≥15 events/h) in cancer patient cohorts such as; breast cancer ($13.3\%$–$58\%$), 11, 12 endometrial cancer ($57\%$), 11 melanoma ($60.7\%$), 13, 14 lung cancer ($50\%$) 15, 16 and prostate cancer ($48.7\%$). 17 Overall, these prevalence rates tend to be higher than those reported ($21.2\%$–$49.7\%$) for community‐based cohorts, 18, 19 and even approach those seen in sleep‐clinic settings ($51.7\%$–$67.4\%$). 20, 21, 22, 23 Notably, OSA prevalence in breast and endometrial cancer patients, 11 exceed rates observed in women from the general population ($13.2\%$–$23.4\%$) 18, 19 and from women in sleep clinics ($38.8\%$). 20 These findings highlight the extent of unrecognised OSA likely present in cancer patient cohorts, making OSA a relatively prevalent co‐morbidity for cancer patients. The aetiology of OSA in cancer cohorts is not well understood. However, a likely candidate is the presence of common risk factors. Among these, age and obesity are well‐known risk factors for both OSA 24, 25, 26 and a number of cancers. 27, 28, 29 OSA is a treatable condition, but is often left unrecognised, undiagnosed, and untreated. 30, 31 Implementing protocols expediting OSA diagnoses in oncology‐clinic settings can facilitate timely OSA treatment, to reduce the burden of comorbid OSA in cancer patients. An OSA diagnosis is made using overnight, attended, in‐laboratory polysomnography (PSG), or unattended home PSG, both of which are finite, labour intensive resources. 32 Alternatively, tools such as questionnaires 33, 34 and at‐home limited channel monitoring studies, which record between 2 and 8 physiological variables, 35 can be used to identify patients with a high probability of OSA for streamlining the utilisation of finite PSG resources. This approach has been utilised in primary care, 36 and pre‐anaesthetic clinics, 33 but has not been widely applied in oncology clinics. The aim of the present study was to investigate the relative ability of two widely used and previously validated community/sleep clinic screening tools to accurately identify cancer patients with a high probability for OSA when deployed in an oncology clinic setting: [1] the STOP‐Bang questionnaire (SBQ), 33 and [2] the ApneaLink™Air device (ResMed, Australia). 35 We recruited patients from three different post‐treatment oncology clinics: [1] breast cancer; a female cohort with a reported high prevalence of sleep disturbance, 37 [2] endometrial cancer; a female cohort with a high prevalence of obesity (a shared risk factor for OSA), 38 and [3] melanoma; a mixed sex cohort with a proposed cancer‐OSA pathological linkage. 39 ## Patients The Western Sydney Local Health District Ethics Committee approved this study. Written informed consent was obtained from all patients. The study was conducted in accordance with the Declaration of Helsinki. Patients were recruited from melanoma, breast, or endometrial cancer outpatient clinics at Westmead Hospital, Sydney Australia; from August 2016 until November 2019. PSG data for breast and endometrial cancer patients included in the present study have been reported previously. 11 Patients were eligible to participate if they: [1] were ≥ 18 years of age, [2] had a confirmed diagnosis of either breast cancer, endometrial cancer or melanoma, [3] had completed their treatment regimen (i.e., surgery, chemotherapy, radiotherapy) for a minimum of 2 months (endometrial cancer), 6 months (melanoma), or 12 months (breast cancer) prior to study recruitment, [4] were able to understand instructions relevant to study requirements, and [5] provided informed written consent. Patients were also screened for serious medical conditions via a physician‐conducted telephone interview, and were excluded if they had serious respiratory, cardiovascular, hepatic, renal, neurological or psychiatric conditions, or if they were pregnant. ## Polysomnography All patients underwent a standard clinical practice (single‐night) 32 overnight in‐laboratory PSG session in the Sleep Research Facility, at the Westmead Institute for Medical Research (Sydney, Australia). The following signals were measured: nasal pressure, oronasal thermistor signals, snoring auditory signals, thoracic and abdominal inductive plethysmography, pulse oximetry (SpO2), EEG, EOG, and chin, diaphragm, and pre‐tibial EMG. PSG data were collected, scored, and analysed by an experienced sleep technician using American Academy of Sleep Medicine (AASM‐2012) criteria. 39 The following metrics were calculated: [1] total sleep time (TSTPSG), [2] apnoea‐hypopnoea index (AHIPSG), [3] oxygen desaturation index (ODIPSG; number of desaturations of ≥$3\%$ per hour), and [4] proportion of time asleep when SpO2 levels were <$90\%$ (TsatPSG < $90\%$). PSG metrics and prevalence values for moderate‐to‐severe and severe OSA, stratified by cancer subgroup, are displayed in Table 2. TABLE 2 \| Unnamed: 0 \| BRC (n = 56) \| ENDO (n = 37) \| MEL (n = 50) \| Total (n = 143) \| p a \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| PSG metrics \| PSG metrics \| PSG metrics \| PSG metrics \| PSG metrics \| PSG metrics \| \| TSTPSG (min) \| 357.5 (313.3–416.8) \| 372.0 (313.8–419.3) \| 353.0 (324.4–386.1) \| 356.5 (318.0–398.0) \| .4125 \| \| AHIPSG (events/h) \| 17.5 (7.4–34.8) \| 15.7 (10.0–33.4) \| 12.8 (7.4–22.0) \| 14.6 (7.5–31.6) \| .2421 \| \| ODIPSG (events/h) \| 5.7 (2.1–17.0) \| 10.0 (2.4–20.9) \| 3.3 (1.9–8.8) \| 5.5 (2.1–15.1) \| .0613 \| \| TsatPSG < 90% (min) \| 0.4 (0.0–3.5) \| 2.4 (0.2–9.7) \| 0.2 (0.0–1.1) \| 0.5 (0.0–3.3) \| .0014 \| \| OSA severity categories \| OSA severity categories \| OSA severity categories \| OSA severity categories \| OSA severity categories \| OSA severity categories \| \| AHIPSG ≥ 15 (n [%]) \| 30 (53.6%) \| 21 (56.8%) \| 20 (40.0%) \| 71 (49.7%) \| .2283 \| \| AHIPSG ≥ 30 (n [%]) \| 17 (30.4%) \| 12 (32.4%) \| 8 (16.0%) \| 37 (25.9%) \| .1382 \| ## STOP‐Bang questionnaire The SBQ is an eight‐item binary (yes/no) validated tool, measuring recognised OSA phenotypic characteristics. 33 The ‘STOP’ section contains four questions, each capturing a self‐reported OSA symptom (i.e., snoring, tiredness, observed apnoea and high blood pressure). The ‘Bang’ section records four additional demographic conditions (BMI ≥35 kg/m2, age ≥ 50 years, neck circumference ≥40 cm and male gender). Patients self‐completed the questionnaire, prior to undergoing PSG. ‘Yes’ responses were scored ‘1,’ and ‘No’ responses were scored ‘0.’ A total SBQ score was determined by summing the ‘yes’ responses. The total SBQ score can range between 0 and 8 arbitrary units (au). 33 The SBQ is a straightforward, easy‐to‐administer OSA screening tool with a simple scoring system. 33 The tool has been validated across many clinical settings 33, 57 and adopted in certain clinical guidelines 32 for the appropriate triage of patients with suspected OSA for further clinical assessment. The majority ($$n = 82$$; ~$60\%$) of patients in the present study who completed the SBQ ($$n = 137$$) scored ≥3 au, indicating most patients had an elevated OSA risk. 49 SBQ≥3 au had good discriminative power (NPV ≈ $93\%$) to identify patients without severe OSA, but performed poorly for positively identifying those with severe OSA (PPV = $39\%$). This trend of high NPV values and low PPV values, resembles previously reported data for sleep clinic, 52 surgical clinic, 34 and general population‐based cohorts. 46 A low PPV for SBQ and may be related to: [1] underlying prevalence of OSA in the particular study cohort, 34 or the [2] gender‐specific technical issues with STOP‐Bang. 50, 54 The present cohort contained a high proportion of female participants who can only obtain a maximum SBQ score of 7 au. SBQ has been previously demonstrated to underestimate the risk for OSA in female populations. 50, 54 When matched for OSA severity, males tend to produce a significantly higher mean SBQ score than females, a finding attributed to the SBQ being focused on the male gender and typical male OSA symptoms. 57 However, despite this SBQ performed similarly in our predominantly‐female, oncology clinic‐recruited cohort to previous reports 46, 51 assessing mixed sex cohorts recruited from other clinical settings. Consistent with previous reports, overall, an SBQ≥3 au. cutoff in the present study effectively identified patients unlikely to have severe OSA. 51 ## ApneaLink™Air Patients received an at‐home limited channel monitoring device kit (ApneaLink™Air kit; ResMed, Sydney, Australia) with printed instructions, either during recruitment or upon PSG completion. The device records respiratory airflow from a nasal cannula, respiratory effort via a thoracic movement ‘effort’ sensor, and blood oxygen saturation and pulse rate using a finger pulse oximeter. After self‐administered usage at home for one night, the kit was returned in‐person or via a pre‐paid postal envelope. The failure rate for obtaining technically acceptable ApneaLink™Air data from those who received the device ($$n = 117$$) was $15.4\%$, this includes patients who were provided the device but did not subsequently use it. For those who used the device ($$n = 105$$), the technical failure rate was $5.7\%$, similar to previous reports for cohorts recruited from sleep clinics or other settings. 23, 47, 52, 58 Based on ApneaLink™Air data alone, moderate‐to‐severe OSA was present in ~$38\%$ of patients ($$n = 38$$) and severe OSA in ~$11\%$ ($$n = 11$$). These prevalence rates are less that obtained via PSG, for the same patient cohort. ApneaLink™Air had moderate utility in predicting moderate‐to‐severe OSA, but greater utility for identifying patients without severe OSA, particularly when using an AHIAL ≥ 10 events/h threshold. The PPV for predicting an AHIPSG ≥ 15 events/h using the AHIAL ≥ 15 events/h threshold, in this study is similar to some previous validation studies, 55 but not others. 48, 56 The NPV for ApneaLink™Air for moderate‐to‐severe OSA when using the AHIAL ≥ 15 threshold, was substantially lower than most previously reported values ($68.9\%$ vs. $51.4\%$, 56 $82.5\%$ 48 and $93.5\%$ 55). The NPV for an AHIAL ≥ 30 events/h threshold to identify patients without severe OSA was better than reported by Delesie et al., 2021 56 ($63.6\%$ vs $70\%$ 56). Differences between our study and with other validation studies that have used different ApneaLink™ devices may be a consequence of: [1] the prevalence of moderate‐to‐severe or severe OSA within the particular study cohort 50; [2] the criteria used to score OSA‐related events, 40, 59 [3] the minimum acceptable evaluation period, 53 [4] the use of manual or automatic scoring methods, 47 and [5] the particular version of the ApneaLink™ device used. 47 ## ApneaLink™Air data ApneaLink™Air data were downloaded onto a computer for analysis. Data were analysed and scored automatically using commercially available software (ApneaLink™Air Application Software Multilingual, ResMed Australia). The evaluation period for each recording was the total recording period minus the first 10 min, and sections with poor quality signals. If the evaluation period was <120 min, the recording was excluded. The recordings were reviewed by an experienced sleep technician for acceptable technical quality and to confirm all identified events met AASM‐2012 criteria. 40 Events were re‐scored by the sleep technician if required. An event was scored as an apnoea if it involved a ≥ $90\%$ drop in respiratory airflow from pre‐event baseline, lasting ≥10 s. Hypopneas were scored if events involved a ≥ $30\%$ drop in respiratory airflow lasting ≥10 s. The following outputs were generated: [1] evaluation period (min), [2] apnoea‐hypopnea index (AHIAL: includes obstructive, central, and mixed apnoea classes and hypopneas per hour), [3] oxygen desaturation index (ODIAL: number of desaturations of ≥$3\%$ per hour) and [4] percentage of time asleep with SpO2AL ≤ $90\%$ (TsatAL ≤ $90\%$). Sleep duration was considered equivalent to the evaluation period. Of the 117 patients who received an ApneaLink™Air device, 12 ($10.3\%$) did not subsequently use the device. Recordings from 6 ($5.7\%$) of the 105 patients who used the device, were considered technically unacceptable. Overall, the failure rate for obtaining technically acceptable recordings from those who received an ApneaLink™Air was $15.4\%$. The prevalence of ApneaLink™Air‐determined moderate‐to‐severe OSA (AHI AL ≥ 15 events/h) and severe OSA (AHI AL ≥ 30 events/h) is illustrated in Figure 2. For those with technically acceptable data ($$n = 99$$), the median evaluation period was 406 (298–462) min. Group median (IQR) values were: [1] AHIAL: 10.9 (4.9–22.2) events/h, [2] ODIAL: 14.0 (6.6–23.8) events/h, and [3] TsatAL < $90\%$: 7.9 (1.4–$27.9\%$) events/h. FIGURE 2: Frequency histogram for ApneaLink™Air‐derived apnoea‐hypopnea indices (AHIAL) values (n = 99). Bin width = 5 events/h. Coloured lines represent indicated AHIAL threshold values, with associated percentages indicating prevalence for each AHIAL‐determined OSA severity category. Across all threshold levels, PPV values were at the most modest for predicting both moderate‐to‐severe and severe OSA. NPVs were highest for predicting severe OSA, but lower for predicting moderate‐to‐severe OSA (Table 3). ## Statistical analysis Statistical analysis was performed using Statistical Package for Social Sciences version 24 (IBM SPSS Inc., Chicago, Illinois, USA), GraphPad PRISM® version 9.3.1 (GraphPad Inc., San Diego, CA) and Microsoft Excel version 7 (Microsoft, Redmond, Washington, USA). Individual data were grouped and expressed as median (interquartile range [IQR]). Kruskal‐Wallis tests were used to compare continuous data, and chi‐square tests for categorical data. A two‐tailed p‐value <.05 was considered statistically significant. The ability of the SBQ and ApneaLink™Air data to predict PSG‐confirmed OSA, was determined using positive predictive values (PPV) and negative predictive values (NPV). PPV=True positive cases/Test positive cases×100. NPV=True negative cases/Test negative cases×100. For this study, an AHIPSG ≥ 15 events/h was used to confirm the presence of moderate‐to‐severe OSA, while an AHIPSG ≥ 30 events/h established severe OSA. PPV and NPV values were calculated for: [1] a SBQ ≥3 au, a threshold score previously established to possess high discriminative power to screen for moderate‐to‐severe and severe OSA, 33, 41 [2] AHIAL thresholds of 10, 15 and 30 events/h, and [3] a combination of SBQ≥3 au and the AHIAL thresholds. PPV and NPV values for both SBQ and ApneaLink™Air data were also assessed by cancer type and age (<60 years vs. ≥60 years). ## RESULTS We recruited 56 breast cancer, 37 endometrial cancer and 50 melanoma patients (Table 1). TABLE 1 \| Unnamed: 0 \| BRC (n = 56) \| ENDO (n = 37) \| MEL (n = 50) \| Total (n = 143) \| p a \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| Age (years) \| 60.0 (54.0–67.0) \| 59.0 (54.5–67.0) \| 61.5 (52.8–69.0) \| 60.0 (54.0–67.0) \| .8745 \| \| Gender (F/M) \| 56/0 \| 37/0 \| 18/32 \| 111/32 \| <.0001 \| \| Height (cm) \| 160.0 (155.1–164.0) \| 160.0 (153.8–162.5) \| 171.1 (167.0–175.1) \| 162.0 (156.6–170.0) \| <.0001 \| \| Weight (kg) \| 73.5 (65.7–83.8) \| 78.4 (72.0–101.1) \| 87.9 (76.6–95.1) \| 79.2 (69.2–93.3) \| .0015 \| \| BMI (kg/m2) \| 29.1 (25.6–31.3) \| 32.6 (28.4–38.0) \| 30.0 (26.0–33.0) \| 29.9 (26.4–33.5) \| .0035 \| \| Neck circumference (cm) \| 35.0 (33.6–37.0) \| 37.0 (34.5–39.5) \| 40.0 (35.5–42.9) \| 36.5 (34.0–40.5) \| .0004 \| ## Patient characteristics Anthropometric and demographic data for all patients are displayed in Table 1. PSG data were obtained from 143 patients, with SBQ data available for 137 of these patients. The ApneaLink™Air device was provided to 117 patients, however only 99 patients had recordings meeting technical criteria (see 2.5.3). There were 53 patients with PSG data who had both an SBQ score ≥3 au and a technically acceptable ApneaLink™Air recording. ## SBQ scores The median (IQR) SBQ score was 3 (2.0–5.0) au., with 82 patients scoring ≥3 au (Figure 1). FIGURE 1: Bar chart showing distribution of STOP‐Bang questionnaire (SBQ) scores (n = 137) An SBQ≥3 au, produced modest PPVs for predicting both moderate‐to‐severe and severe OSA, but a high NPV for severe OSA (see Table 3). TABLE 3 \| Unnamed: 0 \| AHIPSG ≥ 15 \| AHIPSG ≥ 15.1 \| AHIPSG ≥ 30 \| AHIPSG ≥ 30.1 \| \| --- \| --- \| --- \| --- \| --- \| \| \| PPV \| NPV \| PPV \| NPV \| \| SBQ only \| SBQ only \| SBQ only \| SBQ only \| SBQ only \| \| SBQ ≥3 au \| 63.4 \| 69.1 \| 39.0 \| 92.7 \| \| AL only \| AL only \| AL only \| AL only \| AL only \| \| AHIAL = 10 \| 61.1 \| 75.6 \| 33.3 \| 95.6 \| \| AHIAL = 15 \| 65.8 \| 68.9 \| 44.7 \| 95.1 \| \| AHIAL = 30 \| 72.7 \| 59.1 \| 63.6 \| 85.2 \| \| SBQ & AL \| SBQ & AL \| SBQ & AL \| SBQ & AL \| SBQ & AL \| \| AHIAL = 10 \| 60.6 \| 50.0 \| 42.4 \| 90.0 \| \| AHIAL = 15 \| 69.6 \| 53.3 \| 56.5 \| 90.0 \| \| AHIAL = 30 \| 80.0 \| 48.8 \| 70.0 \| 79.1 \| ## Combination of SBQ and ApneaLink™Air data The combination of the SBQ ≥ 3 au threshold and AHIAL thresholds of 10, 15 or 30 events/h did not substantially improve NPV or PPV values when compared with those achieved for each tool alone. There was only a slight improvement in PPV for predicting moderate‐to‐severe and severe OSA (Table 3). ## Cancer type and patient age Figure 3 displays PPV and NPV for SBQ ≥ 3 au and AHIAL ≥ 10 events/h for predicting an AHIPSG ≥ 15 events/h, by cancer type and patient age. The lowest PPV values, for both SBQ ≥ 3 au and AHIAL ≥ 10 events/h, occurred in the melanoma and the age ≥ 60 years sub‐groups. FIGURE 3: Bar plots of positive predictive (PPV; red) and negative predictive values (NPV; green) for moderate‐to‐severe OSA (AHIPSG ≥ 15 events/h) when using a STOP‐Bang (SBQ) score ≥3 au across (A) cancer sub‐groups (n = 137; BRC, n = 52; ENDO, n = 35; MEL, n = 50), and (B) age sub‐groups (n = 137; Age < 60, n = 68; Age ≥ 60, n = 69); and for an ApneaLink™Air AHI threshold ≥10 events/h (AHIAL ≥ 10 events/h) across (C) cancer sub‐groups (n = 99; BRC, n = 34; ENDO, n = 26; MEL, n = 39), and (D) age sub‐groups (n = 99; Age < 60, n = 45; Age ≥ 60, n = 54). BRC, breast cancer; ENDO, endometrial cancer; MEL, melanoma ## DISCUSSION Cancer patients often complain of poor sleep and sleep‐related symptoms (e.g., difficulty falling asleep, daytime sleepiness, unrefreshing sleep, snoring, fatigue). 42 We recently published data describing four distinct sleep symptom clusters identified across 318 cancer patients; a minimally symptomatic group ($47.7\%$); insomnia‐predominant group ($24.9\%$); very sleepy with upper airway symptoms ($16.3\%$), and a severely symptomatic group with severe dysfunction ($11.1\%$). 43 Breast cancer patients were more likely to report insomnia related or severe symptoms, whereas melanoma patients were more likely to be minimally symptomatic or sleepy with upper airway symptoms. Endometrial cancer patients were equally distributed across symptom clusters. These sleep symptom clusters overlap with similar symptom clusters reported for OSA patients. 44 Consequently, there is a need to identify whether cancer patient sleep symptomology is related to co‐morbid OSA and is, therefore, specifically treatable. OSA screening may provide a tool for oncologists to identify which patients warrant referral to a sleep physician for evaluation. The purpose of the present methodological study was to validate SBQ and ApneaLink™Air for use as OSA screening methodologies when applied in an oncology clinic setting. There are no previous publications that report SBQ and ApneaLink™Air performance data for cancer clinic cohorts. All previous validation studies for these tools were conducted in sleep/surgical clinics or community settings. 41, 45, 46, 47, 48, 49, 50, 51, 53, 54, 55, 56 The present study demonstrated that, in cancer patients, either SBQ or ApneaLink™Air were modestly effective at excluding patients without severe OSA, with no additional benefit gained when using them in combination. However, both methods were ineffective at identifying patients with moderate‐to‐severe or severe OSA. The prevalence of moderate‐to‐severe OSA in this older, obese, predominantly female cohort ($$n = 143$$) was nearly $50\%$ (71 patients), with 37 patients having severe OSA. This prevalence falls at the higher end of the range reported for the general adult ($21.2\%$–$49.7\%$), 18, 19 and is higher than for the general female population ($13.2\%$–$23.4\%$). 18, 19 Indeeed, prevalence rates arising from PSG data in the present study, approach those reported in sleep‐clinic cohorts, 20, 21, 22, 23 suggesting OSA prevalence in cancer patients may be as high as in the high‐pretest probability environment of a sleep physician referral clinic. Furthermore, given that OSA was not previously clinically recognised in any of the cancer patients included in the present study, these findings emphasise the nature of the unmet need for OSA screening in cancer patients. ## SBQ and ApneaLink™Air data A 2‐step screening model, developed to identify patients with moderate‐to‐severe OSA in a primary care setting using a questionnaire and the ApneaLink™Air device, 36 has been recommended to streamline the utilisation of finite PSG resources. 32 However, in the present study, combining SBQ and ApneaLink™Air data made little difference to screening outcomes (Table 3). ## Patient age and cancer type Both SBQ and ApneaLink™Air tools were both less effective at positively identifying those with moderate‐to‐severe OSA in the melanoma sub‐group (PPV for predicting AHIPSG ≥ 15 events/h was $48.6\%$ and $40.0\%$ for SBQ≥3 au. and AHIAL ≥ 10 events/h, respectively), with ApneaLink™Air being also less effective in the older age group (Figure 3). The reason for these outcomes is not clear from our data set. ## Strengths and limitations This is the first study to document performance characteristics for OSA screening tools, SBQ and ApneaLink™Air device, in an oncology clinic setting. Major strengths include the use of the ‘gold standard PSG’ to confirm OSA status and establish the ‘true’ prevalence of OSA in the study cohort, and experienced technician review of ApneaLink™Air data recording quality. Patients were recruited while attending for post‐treatment follow‐up, and ApneaLink™Air studies were performed unattended at home. Thus, this study reflects the ‘real world’ experience of deploying OSA screening tool(s) in clinical oncology settings. Limitations include a small sample size, a restricted number of cancer types, and potential recruitment bias associated with volunteers self‐selecting for OSA symptoms thus potentially inflating the underlying OSA prevalence in the study cohort. Single‐night studies for PSG and ApneaLink™Air leave our dataset vulnerable to night‐to‐night variability; 59, 60 however, single night PSG studies are the accepted clinical standard, making our study generalisable to actual clinical practice. 32 Since our study featured two female‐associated cancers, our study cohort was predominantly female. Consequently, our findings might not be generalisable to other cancer types or to male‐dominant cancer cohorts. ## CONCLUSION In a cohort of breast, endometrial and melanoma cancer patients, SBQ data effectively identified patients unlikely to have severe OSA, and shared similar performance characteristics to ApneaLink™Air data. Both tools were less effective at positively identifying patients with moderate‐to‐severe or severe OSA. There was no improvement with a 2‐step combined tool approach. We conclude that PSG remains the optimal tool for the positive diagnosis and management of comorbid OSA in cancer patient cohorts. ## AUTHOR CONTRIBUTIONS Harini Subramanian: Data curation (equal); formal analysis (lead); visualization (lead); writing – original draft (equal); writing – review and editing (lead). Veronika Fuchsova: Data curation (equal); formal analysis (lead); investigation (lead). Elisabeth Elder: Funding acquisition (supporting); resources (equal); writing – review and editing (supporting). Alison Brand: Funding acquisition (supporting); resources (equal); writing – review and editing (supporting). Julie Howle: Funding acquisition (supporting); resources (equal); writing – review and editing (supporting). Anna deFazio: Funding acquisition (supporting); resources (equal); writing – review and editing (supporting). Graham J Mann: Funding acquisition (supporting); resources (equal); writing – review and editing (supporting). Terence Amis: Conceptualization (equal); formal analysis (supporting); funding acquisition (lead); methodology (equal); project administration (equal); supervision (equal); visualization (supporting); writing – original draft (equal); writing – review and editing (lead). Kristina Kairaitis: Conceptualization (equal); formal analysis (supporting); funding acquisition (lead); methodology (equal); project administration (lead); supervision (equal); visualization (supporting); writing – original draft (equal); writing – review and editing (lead). ## CONFLICT OF INTEREST The authors explicitly state that there are no conflicts of interest in connection with this article. Anna DeFazio declares receiving research grants and honoraria from AstraZeneca, although this is not in relation to the subject matter or materials discussed in this manuscript. ## ETHICS STATEMENT All study procedures were approved by the Western Sydney Local Health District Ethics Committee, and were undertaken, adhering to the Declaration of Helsinki. ## DATA AVAILABILITY STATEMENT All data collected in this study is primary data. Data will be made available, on acceptance of manuscript for publication. ## References 1. Senaratna C, Perret J, Lodge C. Prevalence of obstructive sleep apnea in the general population: a systematic review. Sleep Med Rev (2017.0) 34 70-81. DOI: 10.1016/J.SMRV.2016.07.002 2. Malhotra A, White DP. Obstructive sleep apnoea. The Lancet (2002.0) 360 237-245. DOI: 10.1016/S0140-6736(02)09464-3 3. Zhou J, Camacho M, Tang X, Kushida C. A review of neurocognitive function and obstructive sleep apnea with or without daytime sleepiness. Sleep Med (2016.0) 23 99-108. DOI: 10.1016/j.sleep.2016.02.008 4. Arita A, Sasanabe R, Hasegawa R. Risk factors for automobile accidents caused by falling asleep while driving in obstructive sleep apnea syndrome. Sleep and Breathing (2015.0) 19 1229-1234. DOI: 10.1007/s11325-015-1145-7 5. Lee W, Lee SA, Ryu HU, Chung YS, Kim WS. Quality of life in patients with obstructive sleep apnea: relationship with daytime sleepiness, sleep quality, depression, and apnea severity. Chronic Respiratory Disease (2015.0) 13 33-39. DOI: 10.1177/1479972315606312 6. Peppard P, Young T, Palta M, Skatrud J. Prospective study of the association between sleep‐disordered breathing and hypertension. New Engl J Med (2009.0) 342 1378-1384. DOI: 10.1056/NEJM200005113421901 7. Lyons OD, Ryan CM. Sleep apnea and stroke. Can J Cardiol (2015.0) 31 918-927. DOI: 10.1016/J.CJCA.2015.03.014 8. Chan M, Wang CY, Seet E. Investigators for the PVC in UOSA (POSA) S. Association of unrecognized obstructive sleep apnea with postoperative cardiovascular events in patients undergoing major noncardiac surgery. JAMA (2019.0) 321 1788-1798. DOI: 10.1001/JAMA.2019.4783 9. Justeau G, Gervès‐Pinquié C, le Vaillant M. Association between nocturnal hypoxemia and cancer incidence in patients investigated for OSA: data from a large multicenter French cohort. Chest (2020.0) 158 2610-2620. DOI: 10.1016/J.CHEST.2020.06.055 10. Marshall NS, Wong KKH, Cullen SRJ, Knuiman MW, Grunstein RR. Sleep apnea and 20‐year follow‐up for all‐cause mortality, stroke, and cancer incidence and mortality in the Busselton health study cohort. J Clin Sleep Med (2014.0) 10 355-362. DOI: 10.5664/JCSM.3600 11. Madut A, Fuchsova V, Man H. Increased prevalence of obstructive sleep apnea in women diagnosed with endometrial or breast cancer. PLoS One (2021.0) 16. DOI: 10.1371/JOURNAL.PONE.0249099 12. Association between sleep‐disordered breathing and breast cancer aggressiveness. PLOS ONE (2018.0) 13. DOI: 10.1371/journal.pone.0207591 13. Martínez‐Garćia MÁ, Martorell‐Calatayud A, Nagore E. Association between sleep disordered breathing and aggressiveness markers of malignant cutaneous melanoma. Eur Respir J (2014.0) 43 1661-1668. DOI: 10.1183/09031936.00115413 14. Martinez‐Garcia MA, Campos‐Rodriguez F, Nagore E. Sleep‐disordered breathing is independently associated with increased aggressiveness of cutaneous melanoma: a multicenter observational study in 443 patients. Chest (2018.0) 154 1348-1358. DOI: 10.1016/J.CHEST.2018.07.015 15. Cabezas E, Pérez‐Warnisher MT, Troncoso MF. Sleep disordered breathing is highly prevalent in patients with lung cancer: results of the sleep apnea in lung cancer study. Respiration (2019.0) 97 119-124. DOI: 10.1159/000492273 16. Dreher M, Krüger S, Schulze‐Olden S. Sleep‐disordered breathing in patients with newly diagnosed lung cancer. BMC Pulmonary Med (2018.0) 18 1-6. DOI: 10.1186/S12890-018-0645-1 17. Association between sleep‐disordered breathing and prostate cancer. Sleep Medicine (2022.0) 91. DOI: 10.1016/j.sleep.2022.02.007 18. Heinzer R, Vat S, Marques‐Vidal P. Prevalence of sleep‐disordered breathing in the general population: the HypnoLaus study. Lancet Respir Med (2015.0) 3 310-318. DOI: 10.1016/S2213-2600(15)00043-0 19. Fietze I, Laharnar N, Obst A. Prevalence and association analysis of obstructive sleep apnea with gender and age differences – results of SHIP‐trend. J Sleep Res (2019.0) 28. DOI: 10.1111/JSR.12770 20. Kuczyński W, Mokros Ł, Stolarz A, Białasiewicz P. The utility of STOP‐BANG questionnaire in the sleep‐lab setting. Sci Rep (2019.0) 9 9. DOI: 10.1038/s41598-019-43199-2 21. Pataka A, Daskalopoulou E, Kalamaras G, Fekete Passa K, Argyropoulou P. Evaluation of five different questionnaires for assessing sleep apnea syndrome in a sleep clinic. Sleep Med (2014.0) 15 776-781. DOI: 10.1016/J.SLEEP.2014.03.012 22. Farney R, Walker B, Farney R, Snow G, Walker J. The STOP‐bang equivalent model and prediction of severity of obstructive sleep apnea: relation to polysomnographic measurements of the apnea/hypopnea index. J Clin Sleep Med (2011.0) 7 459-465. DOI: 10.5664/JCSM.1306 23. Chen H, Lowe A, Bai Y, Hamilton P, Fleetham J, Almeida F. Evaluation of a portable recording device (ApneaLink) for case selection of obstructive sleep apnea. Sleep Breath (2009.0) 13 213-219. DOI: 10.1007/S11325-008-0232-4 24. Peppard P, Young T, Palta M. Longitudinal study of moderate weight change and sleep‐disordered breathing. JAMA (2000.0) 284 3015-3021. DOI: 10.1001/JAMA.284.23.3015 25. Romero‐Corral A, Caples S, Lopez‐Jimenez F, Somers V. Interactions between obesity and obstructive sleep apnea: implications for treatment. Chest (2010.0) 137 711-719. DOI: 10.1378/CHEST.09-0360 26. Young T, Shahar E, Nieto F. Predictors of sleep‐disordered breathing in community‐dwelling adults: the sleep heart health study. Arch Intern Med (2002.0) 162 893-900. DOI: 10.1001/ARCHINTE.162.8.893 27. Howlader N, Noone A, Krapcho M, Garshell J, Miller D, Altekruse S, Kosary C, Yu M, Ruhl J, Tatalovich Z, Mariotto A, Lewis D, Chen H, Feuer E. SEER Cancer Statistics Review, 1975–2012. (2014.0) 28. Sergentanis T, Antoniadis A, Gogas HJ. Obesity and risk of malignant melanoma: a meta‐analysis of cohort and case–control studies. Eur J Cancer (2013.0) 49 642-657. DOI: 10.1016/J.EJCA.2012.08.028 29. Reeves G, Pirie K, Beral V, Green J, Spencer E, Bull D. Cancer incidence and mortality in relation to body mass index in the million women study: cohort study. BMJ (2007.0) 335 1134. DOI: 10.1136/BMJ.39367.495995.AE 30. Young T, Evans L, Finn L, Palta M. Estimation of the clinically diagnosed proportion of sleep apnea syndrome in middle‐aged men and women. Sleep (1997.0) 20 705-706. DOI: 10.1093/SLEEP/20.9.705 31. Kapur V, Strohl K, Redline S, Iber C, O'Connor G, Nieto J. Underdiagnosis of sleep apnea syndrome in U.S. communities. Sleep Breath (2002.0) 6 49-54. DOI: 10.1007/S11325-002-0049-5 32. Douglas J, Chai‐Coetzer C, McEvoy D. Guidelines for sleep studies in adults ‐ a position statement of the Australasian Sleep Association. Sleep Med (2017.0) 36 S2-S22. DOI: 10.1016/J.SLEEP.2017.03.019 33. Chung F, Yegneswaran B, Liao P. STOP questionnaire: a tool to screen patients for obstructive sleep apnea. Anesthesiology (2008.0) 108 812-821. DOI: 10.1097/ALN.0b013e31816d83e4 34. Netzer NC, Stoohs RA, Netzer CM, Clark K, Strohl KP. Using the Berlin questionnaire to identify patients at risk for the sleep apnea syndrome. Ann Intern Med (1999.0) 131 485-491. DOI: 10.7326/0003-4819-131-7-199910050-00002 35. ResMed.ApnealinkPlus/ApnealinkAir clinical guide. (2015.0) 36. Chai‐Coetzer CL, Antic NA, Rowland L. A simplified model of screening questionnaire and home monitoring for obstructive sleep apnoea in primary care. BMJ 66 213-219. DOI: 10.1136/thx.2010.152801 37. Lowery‐Allison AE, Passik SD, Cribbet MR. Sleep problems in breast cancer survivors 1–10 years posttreatment. Palliat Support Care (2018.0) 16 325-334. DOI: 10.1017/S1478951517000311 38. Zhang Y, Liu H, Yang S, Zhang J, Qian L, Chen X. Overweight, obesity and endometrial cancer risk: results from a systematic review and meta‐analysis. Int J Biol Markers (2018.0) 29 e21-e29. DOI: 10.5301/JBM.5000047 39. Berry RB, Budhiraja R, Gottlieb DJ. Rules for scoring respiratory events in sleep: update of the 2007 AASM manual for the scoring of sleep and associated events: deliberations of the sleep apnea definitions task force of the American Academy of sleep medicine. J Clin Sleep Med (2012.0) 8 597-619. DOI: 10.5664/JCSM.2172 40. Nagappa M, Liao P, Wong J. Validation of the STOP‐bang questionnaire as a screening tool for obstructive sleep apnea among different populations: a systematic review and meta‐analysis. PLoS One (2015.0) 10 e0143697. DOI: 10.1371/journal.pone.0143697 41. Liu L, Ancoli‐Israel S. Sleep disturbances in cancer. Psychiatr Ann (2008.0) 38 627-634. PMID: 21243092 42. Kairaitis K, Madut A, Subramanian H. Cancer sleep symptom‐related phenotypic clustering differs across three cancer specific patient cohorts. J Sleep Res (2022.0) 31 e13588. DOI: 10.1111/jsr.13588 43. Ye L, Pien GW, Ratcliffe SJ. The different clinical faces of obstructive sleep apnoea: a cluster analysis. Eur Respir J (2014.0) 44 1600-1607. DOI: 10.1183/09031936.00032314 44. Hug M, Uehli K, Solbach S. Screening for obstructive sleep apnea among hospital outpatients. PLoS One (2018.0) 13. DOI: 10.1371/JOURNAL.PONE.0198315 45. Chen L, Pivetta B, Nagappa M. Validation of the STOP‐bang questionnaire for screening of obstructive sleep apnea in the general population and commercial drivers: a systematic review and meta‐analysis. Sleep and Breathing (2021.0) 25 1741-1751. DOI: 10.1007/s11325-021-02299-y 46. Cho J, Kim H. Validation of ApneaLink™ plus for the diagnosis of sleep apnea. Sleep and Breathing (2017.0) 21 799-807. DOI: 10.1007/S11325-017-1532-3 47. Ragette R, Wang Y, Weinreich G, Teschler H. Diagnostic performance of single airflow channel recording (ApneaLink) in home diagnosis of sleep apnea. Sleep and Breathing (2010.0) 14 109-114. DOI: 10.1007/S11325-009-0290-2/FIGURES/3 48. Chung F, Abdullah HR, Liao P. STOP‐bang questionnaire a practical approach to screen for obstructive sleep apnea. Chest (2016.0) 149 631-638. DOI: 10.1378/chest.15-0903 49. Pataka A, Kotoulas S, Kalamaras G. Gender differences in obstructive sleep apnea: the value of sleep questionnaires with a separate analysis of cardiovascular patients. J Clin Med (2020.0) 9 130. DOI: 10.3390/JCM9010130 50. Pivetta B, Chen L, Nagappa M. Use and performance of the STOP‐bang questionnaire for obstructive sleep apnea screening across geographic regions: a systematic review and meta‐analysis. JAMA Netw Open (2021.0) 4. DOI: 10.1001/jamanetworkopen.2021.1009 51. Erman MK, Stewart D, Einhorn D, Gordon N, Casal E. Validation of the ApneaLink™ for the screening of sleep apnea: a novel and simple single‐channel recording device. J Clin Sleep Med (2007.0) 3 387. PMID: 17694728 52. Nigro CA, Dibur E, Aimaretti S, González S, Rhodius E. Comparison of the automatic analysis versus the manual scoring from ApneaLink™ device for the diagnosis of obstructive sleep apnoea syndrome. Sleep and Breathing (2011.0) 15 679-686. DOI: 10.1007/s11325-010-0421-9 53. Mou J, Pflugeisen B, Crick B. The discriminative power of STOP‐bang as a screening tool for suspected obstructive sleep apnea in clinically referred patients: considering gender differences. Sleep and Breathing (2019.0) 23 65-75. DOI: 10.1007/s11325-018-1658-y 54. Crowley KE, Rajaratnam SMW, Shea SA, Epstein LJ, Czeisler CA, Lockley SW. Evaluation of a single‐channel nasal pressure device to assess obstructive sleep apnea risk in laboratory and home environments. J Clin Sleep Med (2013.0) 09 109-116. DOI: 10.5664/jcsm.2400 55. Cardiorespiratory polygraphy for detection of obstructive sleep apnea in patients with atrial fibrillation. Frontiers in Cardiovascular Medicine (2021.0). DOI: 10.3389/fcvm.2021.758548 56. Trevethan R. Sensitivity, specificity, and predictive values: foundations, pliabilities, and pitfalls in research and practice. Front Public Health (2017.0) 5. DOI: 10.3389/FPUBH.2017.00307 57. Shepertycky MR, Banno K, Kryger MH. Differences between men and women in the clinical presentation of patients diagnosed with obstructive sleep apnea syndrome. Sleep (2005.0) 28 58. Duce B, Milosavljevic J, Hukins C. The 2012 AASM respiratory event criteria increase the incidence of hypopneas in an adult sleep center population. J Clin Sleep Med (2015.0) 11 1425-1431. DOI: 10.5664/JCSM.5280 59. Mosko S, Dickel M, Ashurst J, Mosko S. Night‐to‐night variability in sleep apnea and sleep‐related periodic leg movements in the elderly ~l. Sleep (1988.0) 11 60. Night‐to‐night variability in obstructive sleep apnea severity: relationship to overnight rostral fluid shift. J Clin Sleep Med (2015.0) 11. DOI: 10.5664/jcsm.4462
--- title: 'Residual β-cell function in Brazilian Type 1 diabetes after 3 years of diagnosis: prevalence and association with low presence of nephropathy' authors: - Monica A. L. Gabbay - Felipe Crispim - Sergio A. Dib journal: Diabetology & Metabolic Syndrome year: 2023 pmcid: PMC10026390 doi: 10.1186/s13098-023-01014-z license: CC BY 4.0 --- # Residual β-cell function in Brazilian Type 1 diabetes after 3 years of diagnosis: prevalence and association with low presence of nephropathy ## Abstract ### Background Persistence of β cell-function in Type 1 diabetes (T1D) is associated with glycaemia stability and lower prevalence of microvascular complications. We aimed to assess the prevalence of residual C- peptide secretion in long-term Brazilian childhood onset T1D receiving usual diabetes care and its association to clinical, metabolic variables and microvascular complications. ### Methods A cross-sectional observational study with 138 T1D adults with ≥ 3 years from the diagnosis by routine diabetes care. Clinical, metabolic variables and microvascular complications were compared between positive ultra-sensitive fasting serum C-peptide (FCP +) and negative (FCP-) participants. ### Results T1D studied had ≥ 3 yrs. of diagnosis and $60\%$ had FCP > 1.15 pmol/L. FCP + T1D were older at diagnosis (10 vs 8 y.o; $$p \leq 0.03$$) and had less duration of diabetes (11 vs 15 y.o; $$p \leq 0.002$$). There was no association between the FCP + and other clinical and metabolic variable but there was inversely association with microalbuminuria ($28.6\%$ vs $13.4\%$, $$p \leq 0.03$$), regardless of HbA1c. FCP > 47 pmol/L were associated with nephropathy protection but were not related to others microvascular complications. ### Conclusion Residual insulin secretion is present in $60\%$ of T1D with ≥ 3 years of diagnosis in routine diabetes care. FCP + was positively associated with age of diagnosis and negatively with duration of disease and microalbuminuria, regardless of HbA1c. ## Introduction Type 1 diabetes (T1D) is an autoimmune disease characterized by progressive destruction of β cells that begins long before and continues long after the clinical diagnosis, as believed. It occurs at different rates among individuals, with some demonstrating long residual insulin secretion. There are controversies about the various factors related to the persistence of residual C-peptide secretion in T1D such as genetic, duration of diabetes and age at diagnosis among others. Ethnicity, for instance, might have influenced the rate of β cell loss after diagnosis since Hispanics had shown higher fasting C-peptide at the beginning of Trial Net New Onset Intervention Trials [1]. The importance of residual insulin secretion was already highlighted in the DCCT study (Diabetes Control and Complications Trial) in which participants with stimulated C-peptide higher than 200 nmol/L had better glycemic control, lower risk of hypoglycemia and lower risk of diabetes chronic complications [2, 3]. Classically chronological age, age at T1D diagnosis, duration of T1D, average systolic blood pressure and HbA1c are associated with chronic diabetic complications, but the relation of latter with residual β-cell function has been shown heterogenous results [4–7], besides there is great evidence that C-peptide counteracts the detrimental changes causes by hyperglycemia at the cellular level in animal studies [8]. On the other side, there has no sufficient data about residual β cell function in long duration Brazilian T1D by routine diabetes care as in developing countries with a heterogeneous genetic population. These would help revealing the complex natural history of the disease considering its heterogeneity among different populations. Therefore, the study aimed to assess the prevalence of residual C- peptide secretion, its association with clinical characteristics, and its impact on microvascular complication in Brazilian childhood onset T1D. ## Study participants This is a cross-sectional observational study of 138 Brazilian people with T1D with more than 3 years of duration, selected from electronic medical records between 2014 to 2018 that had realized ultrasensitive C-peptide assay. Inclusion criteria: Individuals recruited from the Diabetes Center—Federal University São Paulo (Southeast Brazil) with Type 1 diabetes (ADA criteria as continuous use of insulin since diagnosis, positive ketones, and presence of pancreatic autoantibodies). Exclusion criteria: Hepatic disease, Chronic renal insufficiency or clinical characteristics suggesting another type of diabetes mellitus. ## Methods The data analyzed were age (years), age at diagnosis of T1D (years), time of diagnosis of diabetes (years), sex, body mass index {BMI (Kg/m2)}, arterial hypertension (SBP > 130 mmHg e DBP > 90 mmHg, or > P95 for age), Total Cholesterol HDL-c (mg/dl), LDL-c (mg/dl), Triglycerides (TG)(mg/dl), TSH (mU/ml) and free T4(ng/ml), HbA1c (HPLC—high-performance liquid chromatography—normal range 20–38 mmol/mol; 3.5 a $5.6\%$), and albumin excretion—Cobas Mira® Roche (AlbUCobas—NV < 17 mg/L or 30 mg/24hs). The ultrasensitive C-peptide assay used was the Mercodia® ELISA (Uppsala, Sweden, cat. No. 10-1141-01) whose detection limit is 1.15 pmol/L, with an inter-and intra-assay coefficient of variation of $5.5\%$ and $3.8\%$, for a C-peptide at 37 pmol / L. Participants whose FCP was greater than 1.15 pmol/L were classified as positive (FCP +) and below this value as negative (FCP-). The presence of microvascular complications was taken derived of data from the electronic medical records and was fundoscopy for retinopathy, DCCT clinical exam protocol for neuropathy, microalbuminuria (in 2 of 3 albumin to creatinine ratios high > 30 mg/ml) for nephropathy and the evaluation of hypoglycemia was performed according to the frequency of file reported hypoglycemia. ## Statistical analyses Normally distributed data were presented as mean SD and variable with skewed distribution were reported as median with interquartile ranges (25th,75th). Categorical variables are expressed as absolute frequency and percentage. For the comparisons of two groups in continuous variables, the t-student test was used depending on the normality assumption (tested by the Anderson–Darling test) and for the others, the Mann–Whitney nonparametric test was used. Fisher’s exact tests were used for categorical variables. Simple and multiple logistic regression models were used to analyze the associations between the outcome variables. Statistics C was used to evaluate the model. We applied a multivariable logistic regression model using complications event as the binary dependent variable and C-peptide as the independent variable. The duration-adjusted comparisons were represented by the P-value and odds ratio of the C-peptide term in the multivariable model. Scatter plots and bar plots were used to illustrate the distribution of variables. The level of significance was 0.05. Two-tailed hypotheses were considered. Software R version 3.6.0 was used to perform all analyzes. ## Results One hundred and thirty-eight participants ($58.7\%$ men) were evaluated and $59.5\%$ of them was FCP + (> 1.15 pmol/L). The means age (sd) at T1D diagnosis was 9.0 ± 0.47 years and by the time of the study was 22.0 ± 0.6 years. Diabetes duration of 12.0 ± 0.5 years and the mean HbA1c was 8.35 ± $1.15\%$ (67 ± 1.6 mmol/mol). Clinical, endocrine, and metabolic characteristics of the participants and the frequency of dyslipidemia and diabetic microvascular complications are shown in Table 1.Table 1Clinical, endocrine, and metabolic characteristics of T1D participantsCharacteristicsTotalFCP−FCP+p valueN (%)13856(40.6)82(59.4)Gender (male)%58.757.159.80.8Age at diagnosis (years)9.0 ± 0.478.70 ± 5.1410.28 ± 5.680.039Age at recruitment (years) mean ± sd22.0 ± 0.623.46 ± 7.2221.84 ± 6.890.11Diabetes duration (years)mean ± sd12.0 ± 0.514.77 ± 6.7111.41 ± 5.680.002Hypothyroidism history (%)16.017.914.60.2BMI (Kg/m2) mean ± sd23.00 ± 0.3023.18 ± 3.2623.52 ± 3.810.59Insulin (IU/kg/day)mean ± sd0.80 ± 0.100.80 ± 0.181.17 ± 2.110.08HbA1c (%) mean ± sd8.35 ± 1.158.55 ± 1.618.73 ± 1.580.42LDL-C (mg/dl) mean ± sd96.0 ± 25.097.4 ± 29.1102.16 ± 29.550.6HDL-C (mg/dl) mean ± sd54.0 ± 13.056.64 ± 13.953.39 ± 13.00.1TG (mg/dl) mean ± sd73.0 ± 46.082.39 ± 41.890.00 ± 61.80.9Dyslipidemia (%)19.021.417.10.6Hypoglycemia (%)18.123.214.60.2Nephropathy (%)19.628.613.40.03Neuropathy (%)2.93.62.41.0Retinopathy (%)3.65.42.40.3FCP− Negative Fast C-peptide, FCP+ Positive Fast C-peptideNegative C-peptide: < 1.15 pmol/L Positive C-peptide: > 1.15 pmol/L. We observed that there was a statistically significant positive association between FCP + and age at T1D diagnosis ($$p \leq 0.039$$) and negative association ($$p \leq 0.002$$) with disease duration. From the regression data, the probability most participants having FCP + with a diabetes duration of 5 years would be about $75\%$, in 10 years $64.5\%$ and in 20 years less than $50\%$ (Fig. 1). Also, these data shown that each time unit increase of diabetes duration(years) correspond to $8\%$ reduction (OR = 0.92) in the probability of having FCP +.Fig. 1Fasting C-peptide levels according to diabetes duration in the T1D participants. The dashed horizontal line across the entire lower portion of the panel displays the limit of detection (1.15 pmol/L) There was no association between the presence of FCP and clinical (age and BMI) and metabolic variables (current HbA1c, LDL-c and TG) by the time of the study. In a regression model adjusted for covariates, we found a lower prevalence ($13.4\%$) of albuminuria among participants with FCP + compared to 27those with FCP- ($28.6\%$, $$p \leq 0.031$$). Simple logistic regression showed that participants without nephropathy were 2.5 times more likely to have FCP + than those with this microangiopathy. ## Discussion This study showed for the first time, that using an ultrasensitive method for serum C-peptide, around $60\%$ of long-term Brazilian T1D with childhood-onset (~ 17 yrs of diabetes duration) had residual insulin secretion. These detectable FCP levels was associated with older age at T1D diagnosis and shorter diabetes duration. Importantly this group with even low levels of residual C-peptide had lower prevalence of microalbuminuria adjusted for diabetes duration and HbA1c levels. The decline of C-Peptide in T1D during the first years (in general three years) as reported in the literature is highly variable [9]. After this period patient become non-C-peptide secretor or low C-peptide secretor, which is more evident and prevalent when using ultra-sensitive C-peptide assays [10]. Most studies evaluating C-peptide in people with long-standing diabetes, included T1Ds with more than 3–5 years of duration, like one from Prof. Greenbaum group (an expert in this area) which use as an eligibility criteria of ages from 6 months to 46 years and diabetes duration of more than 3 years [5]. Others, such as the pioneering study by Denise Faustman [10], stratified by six intervals of disease duration included participants from 0 to more than 40 years duration. This is precisely why it has been studied in recent years, the persistence of prolonged C-peptide production and lower risk of complications, and that people diagnosed in adulthood have more insulin reserve than diagnosed in childhood. [ 11]. And a more recent study considered T1D individuals with diabetes duration with more than 5 years to evaluate C-peptide and complications [12]. Therefore, we selected participants with more than 3 years of diagnosis, but with an average diabetes duration time of 17 years, thus subject to the risk of complications. [ 13], mainly in our country where the T1D glycemic control legacy is not very good. However, the role of diabetes duration on fasting C-peptide in these patients was extensively studied and the probability of being C-peptide negative at 15 years of diagnosis is $45\%$, 20 years is $56\%$, 25 years $66\%$, 30 years is $75\%$ and 40 years is $89\%$. Preliminary studies, from UK Golden Years cohort [14], the 50-year Joslin Medalist study [15], and the Parisian cohort [16] have examined the clinical characteristics of type 1 diabetic patients with long disease duration. Subsequently, Joslin’s group further explored the data to evaluate possible markers of longevity without significant complications, including evaluation of C-peptide and some pancreatic histology. They showed that most of them could produce insulin endogenously ($67\%$ with standard C-peptide assay > 0.03 nmol/L) and confirm the presence of insulin in some available Medalist pancreases. [ 17] It is* important to point out that many Medalists had monogenic diabetes variants that potentially contribute to heterogeneity of beta cell function in this group of patients. In recent years, the development of an ultrasensitive assay for plasma C-peptide detection (change the lower limits of detection of approximately 50 pmol/L to levels as low as 1.15 pmol/L) has made it possible to detect residual secretion of β-cell function in people with T1D, even after decades of disease [18–20]. The percentage of residual FCP that we found is five times higher than the $13.2\%$ found in a group of adults with T1D with a shorter disease time when using a regular C-peptide assay in routine clinics [21]. Studies with similar characteristics to our group, found 55–$66\%$ positivity (lower limit of 1.15 pmol/L for fasting serum C-peptide) and $52\%$ for urinary C- peptide (> 30 pmol/L) [11, 15, 22] while others like participants of Exchange Clinic Network found detectable C-peptide in $29\%$ among 900 participants, suggesting that residual C-peptide secretion is present in almost one out of three T1D individuals, three or more years from diabetes diagnosis [5]. ( Table 2).Table 2Comparation of positive C-peptide among different studies using ultra-sensitive assayGroup(Country)UNIFESP (BR)Wang L, et al. [ 10] (USA)Oram RA, et al. [ 29] (UK)Davis AK et al. [ 5] (USA)Rickels MR et al. [ 27] (USA)N1381827491963Age (Mean or Median) (sd or range)26 (6–52)39(9–85)16(9–23)37.2 ± 18.9 (5–88)18–65DiabetesDuration (yr.)Median (range)17 (3–34)15 (0–73)30 (19–41)13 (3–81) > 2Assaysensitivity detectionMercodia®1.5 pmol/lMercodia®1.5 pmol/lMercodia®1.5 pmol/lTosoh Bioscience®17 pmol/LTosoh Bioscience®7 pmol/LPositive C-peptideFCP > 1.5 pmol/L$55\%$FCP > 1.5 pmol/L$63\%$FCP > 3.3 pmol/l$73\%$Non-FCP > 17 pmol/L$29\%$Peak in the MMT > 7 pmol/L$76\%$A1c (%)8.7 + 1.67.4 + 1.17.9 (7.2–9.0)8.0 + 1.57.6 + 0.7FCP fasting C peptide, MMT Mixed Meal Test According to our analysis, we verified that one of the main factors related to the frequency of FCP considered positive (> 1.15 pmol/L) were age at diabetes diagnosis and the duration of the disease. These two factors, also shown in our population, are in according to other groups where age at the diagnosis was positively associated with C-peptide values [5, 10, 22]. The relationship between age at T1D diagnosis and residual C-peptide can be explained by different insulin profiles found at diagnosis as it has been shown in new onset teenage T1D that still retain approximately $40\%$ of residual insulin-containing islets [21]. In our population each increase of one year of age correspond to $8\%$ reduction (OR = 0.92) in probability of having FCP +. It would be expected better values in our patients than the English and American ones which had longer disease duration, however our patients presented a marginally lower prevalence of residual C-peptide secretion [15, 17]. In addition, detectable C-peptide related to better A1c was found in $38\%$ of children and adolescents after 10 years of diabetes compared to $24\%$ of our patients as they are from public care services. Insufficient metabolic control may have contributed to this result. However, a recent systematic review and meta-analyses of all randomized controlled trials (RCTs) to preserve β-cell function in people with newly diagnosed T1D shown there is a lack of robust evidence that interventions to improve glucose control preserve β-cell function and efforts to treatment algorithms should be a priority [22]. However, one major debate today is what C-peptide level is clinically useful. In part because some studies that correlated residual C-peptide to chronic diabetes complications or hypoglycemia sometimes measure FCP [19, 23, 24] sometimes use stimulated C-peptide [21] besides the sensibility limit of ultra-sensitivity C-peptide assay used [23]. In our T1D participants with a mean 17 years of disease duration, using a C-peptide assay with a limit of sensibility of 1.15 pmol/L we did not find any association between residual C-peptide and lower HbA1c. This agrees with studies with similar T1D populations, while Trial net participants and T1D from the UNITED Team found this association only during the early course of diabetes [23, 25, 26]. Also, a recent work with young adult T1D with 5 years of diabetes duration demonstrated that only those with high levels of residual C-peptide (peak after stimulus test > 4.0 pmol/L or > 1.2 ng/mL) shown β cell responsiveness to hyperglycemia and likely contribute to glycemic control [25]. We have studied β cell responsiveness to sulfonylurea test in a sample of nine patients with FCP + and found that the maximum peak during the test was 0.286 pmol/L (data not shown) which is approximately $93\%$ less than shown in the study above [27]. This might help understanding why we do not get an association between C-peptide levels and HbA1c in our T1D group and why we did no find significant difference in the prevalence of hypoglycemia between our FCP + and FCP- groups. Nevertheless, heterogeneous relationship between the glucagon response to insulin-induced hypoglycemia does exist but again is most evident in T1D with high levels of residual C-peptide [25]. In relation to diabetes chronic complications and residual C-peptide in our study using FCP with limit of sensibility of the assay equal 1.15 pmol/L, we found an inverse relationship between the C-peptide reserve and albuminuria in FCP + T1D patients. The frequency of albuminuria found in our participants was like people with T1D described in the literature with the same time of disease ($19.6\%$ vs $14.8\%$) where $65.2\%$ of microalbuminuric patients had no detectable C-peptide. However, unlike predicted, we found no correlation of C-peptide and retinopathy, perhaps because of the low prevalence found in our participants ($3.9\%$ vs 9–$40\%$) [28]. Also, maybe the low cut-off limit we used to define FCP + patients. The classical work that found C-peptide protection from complications as nephropathy, neuropathy and retinopathy considered FCP levels > 10 pmol/L (with the same C-peptide assay that we used) almost 9 times higher than our C-peptide cut-off. Other study [20] using the same cut-off value of ours (1.15 pmol/L) also did not find relation among residual C-peptide and retinopathy, neuropathy but marginally lower macroalbuminuria in those with detectable levels ($23.4\%$ vs $0\%$, $$p \leq 0.07$$). A larger sample size might have allowed a bigger difference [25]. One point for discussion is why these low levels of C-peptide can decrease the prevalence of nephropathy, regardless to HbA1c level but not the two others microangiopathies (retinopathy and neuropathy). Diabetic nephropathy (DN) is one of the major microvascular complications, present in 20 to $40\%$ of T1D people and is one of the most important causes of kidney failure (KF), but the rate of renal decline varies widely among them. On the other hand, in addition to its well-known role as a biomarker of functional beta-cell mass, the C-peptide is a bioactive molecule with physiological effects on peripheral cell targets and with antioxidant protection on vascular endothelial. Small trials in which C-peptide was given to subjects with T1D with nephropathy or neuropathy showed that C-peptide mitigates renal and neuronal complications [30, 31], while others had shown that C-peptide can suppress various molecular mechanisms involved in the pathophysiology of DN and therefore could prevent the onset and progression of KF [32]. Interestingly almost two decades ago, a double-blind randomized study had demonstrated that administration of biosynthetic human C-peptide plus insulin reduced glomerular permeability and urinary excretion of albumin [33]. In relation to the lower prevalence of DN in our FCP + T1D regardless of the same HbA1c as FCP- ones, we can speculate that the former could have less glycemic variability. Since recent works have been shown that for every 100 pmol/L increases in C-peptide peak the percentage of time spent in the range (70–180 mg/dl) increased by $2.4\%$ with a reduction in time spent at level 1 hyperglycemia (> 180 mg-dl) and level 2 hyperglycemia (250 mg-dl) by $2.6\%$ and $1.3\%$ respectively [34]. Is important to mention that lower time in range was associated with presence of composite microvascular complications in recent study with a group of adults with long T1D duration [34]. When we studied the relationship between C-peptide levels and nephropathy in our T1D group we found out that all patients with fasting C-peptide ≥ 46.9 pmol/L (0.14 ng/ml) were negative for this microangiopathic complication. This number is close to cut-off used in most studies (50 pmol/L but one that considered 10 pmol/) [33] for identify patients at higher risk for complications, more frequent and great glycemic excursions and low 1,5 AG levels. Finally, at this level of glycemic control, we did not find difference on lipid profile between T1D patients with and without residual C-peptide. However, results in this area are heterogenous, some works showing better lipid profile in C-peptide positive and others showing no relationship between unstimulated C-peptide values and lipid parameters in either remitters or non-remitters T1D adult [35]. The present study has some limitations as the small sample size, the cross-sectional data, using FCP although the literature has already shown a good correlation with stimulated C-peptide. Another limitation was the low prevalence of chronic diabetes complications that may have impaired the association studies. The strengths were the assay sensitivity, the heterogenous genetic background of our T1D population and the real-world data from their routine treatment. We can conclude that most of our T1D participants, like American and European data, have residual beta-cell function demonstrated with the use of an ultrasensitive assay after a decade of disease, and this minimal detectable C-peptide appears to protect against albuminuria regardless of HbA1c. Finally, the importance of persistent beta-cell residual function reinforces strategies for its preservation since diagnosis and suggests that a significant percentage of patients with T1D, even after decades of diagnosis, may have benefits in slowing the development of diabetes nephropathy. ## References 1. Tosur M, Cleves MA, Sosenko JM. The effect of ethnicity in the rate of beta-cell functional loss in the first 3 years after type 1 diabetes diagnosis. J Clin Endocrinol Metab (2020.0) 105 e4393-e4406. DOI: 10.1210/clinem/dgaa348 2. Effects of age, duration, and treatment of insulin-dependent diabetes mellitus on residual beta-cell function: observations during eligibility testing for the Diabetes Control and Complications Trial (DCCT). J Clin Endocrinol Metab (1987.0) 65 30-36. DOI: 10.1210/jcem-65-1-30 3. Steffes MW, Sibley S, Jackson M. Beta-cell function and the development of diabetes-related complications in the diabetes control and complications trial. Diabetes Care (2003.0) 26 832-836. DOI: 10.2337/diacare.26.3.832 4. Panero F, Novelli G, Zucco C. Fasting plasma C-peptide and micro-and macrovascular complications in a large clinic-based cohort of Type 1 diabetic patients. Diabetes Care (2009.0) 32 301-305. DOI: 10.2337/dc08-1241 5. Davis AK, DuBose SN, Haller MJ. Prevalence of detectable C-peptide according to age at diagnosis and duration of Type 1 diabetes. Diabetes Care (2015.0) 38 476-481. DOI: 10.2337/dc14-1952 6. Lachin JM, McGee P, Palmer JP. Impact of C-peptide preservation on metabolic and clinical outcomes in the Diabetes Control and Complications Trial. Diabetes (2014.0) 63 739-748. DOI: 10.2337/db13-0881 7. Nakanishi K, Watanabe C. Rate of beta-cell destruction in Type 1 diabetes influences the development of diabetic retinopathy: protective effect of residual beta-cell function for more than 10 years. J Clin Endocrinol Metab (2008.0) 93 4759-4766. DOI: 10.1210/jc.2008-1209 8. Ryk A, Tosiewicz A, Michalak A. Biological activity of C-peptide in microvascular complications of Typ1 diabetes—time for translational studies or back to the basics?. Int J Mol Sc (2020.0) 21 9723. DOI: 10.3390/ijms21249723 9. Palmer JP. C-peptide in the natural history of Type 1 Diabetes. Diabetes Metab Res Ver (2009.0) 25 325-328. DOI: 10.1002/dmrr.943 10. Wang L, Lovejoy NF, Faustman DL. Persistence of prolonged C-peptide production in type 1 diabetes as measured with an ultrasensitive C-peptide assay. Diabetes Care (2012.0) 35 465-470. DOI: 10.2337/dc11-1236 11. Oram RA, Jones AG, Besser REJ. The majority of patients with long-duration type 1 diabetes are insulin microsecretors and have functioning beta cells. Diabetologia (2014.0) 57 187-191. DOI: 10.1007/s00125-013-3067-x 12. Vollenbrock CE, Mul D, Dekker P, Birnie E, Vries-Velraeds MMC. Fasting and meal -stimulated serum C-peptide in long-standing type 1 diabetes mellitus. Diabet Med (2023.0) 40 e15012. DOI: 10.1111/dme.15012 13. 13.Hao W, Gitelman S, DiMeglio LA, et al. Type 1 Diabetes TrialNet Study Group. Fall in C-peptide during first 4 years from diagnosis of type 1 diabetes: variable relation to age, HbA1c, and insulin dose. Diabetes Care. 2016; 39:1664–70. 14. Bain SC, Gill GV, Dyer PH. Characteristics of type 1 diabetes of over 50 years duration (the Golden Years Cohort). Diabet Med (2003.0) 20 808-811. DOI: 10.1046/j.1464-5491.2003.01029.x 15. Keenan HA, Costacou T, Sun JK. Clinical factors associated with resistance to microvascular complications in diabetic patients of extreme disease duration: the 50-year Medalist study. Diabetes Care (2007.0) 30 1995-1997. DOI: 10.2337/dc06-2222 16. Feldman-Billard S, Limon S, Morin Y, Altman JJ. Type 1 diabetes with no diabetic complications, sixty-two years later. J Diabetes Complications (2001.0) 15 285-286. DOI: 10.1016/S1056-8727(01)00153-2 17. Keenan HA, Sun JK, Levine J, Doria A, Aiello LP, Eisenbarth G, Bonner-Weir S, King GL. Residual insulin production and pancreatic -cell turnover after 50 years of diabetes: Joslin Medalist Study. Diabetes (2010.0) 59 2846-2853. DOI: 10.2337/db10-0676 18. Greenbaum CJ, Harrison LC. Guidelines for intervention trials in subjects with newly diagnosed type 1 diabetes. Diabetes (2003.0) 52 1059-1065. DOI: 10.2337/diabetes.52.5.1059 19. Kuhtreiber WM, Washer SLL, Hsu E. Low levels of C-peptide have clinical significance for established Type 1 diabetes. Diabet Med (2015.0) 32 1346-1353. DOI: 10.1111/dme.12850 20. Fotinopoulos E, Clarke CAL, Pattenden RJ. Impact of routine clinic measurement of serum C-peptide in people with a clinician-diagnosis of type 1 diabetes. Diabet Med (2021.0) 38 e14449. PMID: 33131101 21. Leete P, Willcox A, Krogvold L. Differential insulinic profiles determine the extent of β-cell destruction and the age at onset of type 1 Diabetes. Diabetes (2016.0) 65 1362-1369. DOI: 10.2337/db15-1615 22. Narendran P, Tomlinson C, Beese S. A systematic review and meta-analysis of interventions to preserve insulin -secreting β-cell function in people with type 1 diabetes: results from intervention studies aimed at improving glucose control. Diabet Med (2022.0) 39 e14730. DOI: 10.1111/dme.14730 23. Williams KV, Becker DJ, Orchard TJ. Persistent C-peptide levels and microvascular complications in childhood onset type 1 diabetes of long duration. J Diabetes Complicat (2019.0) 33 657-661. DOI: 10.1016/j.jdiacomp.2019.05.019 24. Cheng J, Yin M, Tang X. Residual β-cell function after 10 years of autoimmune type 1 diabetes: prevalence, possible determinants, and implications for metabolism. Ann Transl Med (2021.0) 9 650. DOI: 10.21037/atm-20-7471 25. Marren SM, Hammersley S, McDonald TJ. Persistent C-peptide is associated with reduced hypoglycaemia but not HbA1c in adults with longstanding Type 1 diabetes: evidence for lack of intensive treatment in UK clinical practice?. Diabet Med (2019.0) 36 1092-1099. DOI: 10.1111/dme.13960 26. Gubitosi-Klug RA, Braffett BH, Hitt S. Residual β cell function in long-term type 1 diabetes associates with reduced incidence of hypoglycemia. J Clin Invest (2021.0) 131 e143011. DOI: 10.1172/JCI143011 27. Rickels MR, Evans-Molina C, Bahnson HT. High residual C-peptide likely, contributes to glycemic control in type 1 diabetes. J Clin Invest (2020.0) 130 1850-1862. DOI: 10.1172/JCI134057 28. Tomislav B, Tomic M, Vuckovic-Rebrina S. Preserved C-peptide secretion in patients with type 1 diabetes and incipient chronic complications is associated with lower serum resistin and higher uric acid levels. J Diab Metab Dis. (2020.0) 19 1185-1189. DOI: 10.1007/s40200-020-00620-2 29. Oram RA, McDonald TJ, Shields BM. Most people with long-duration type 1 diabetes in a large population-based study are insulin microsecretors. Diabetes Care (2015.0) 38 323-328. DOI: 10.2337/dc14-0871 30. Johansson BL, Borg K, Fernqvist-Forbes E. Beneficial effects of C-peptide on incipient nephropathy and neuropathy in patients with Type 1 diabetes mellitus. Diabet Med (2000.0) 17 181-189. DOI: 10.1046/j.1464-5491.2000.00274.x 31. Brunskill NJ. C-peptide and diabetic kidney disease. J Intern Med (2017.0) 281 41-51. DOI: 10.1111/joim.12548 32. Johansson BL, Kernell A, Sjöberg S. Influence of combined C-peptide and insulin administration on renal function and metabolic control in diabetes type 1. J Clin Endocrinol Metab (1993.0) 77 976-981. PMID: 8408474 33. Malahi AE, Van Elsen M, Charleer S. Relationship between Time in Range, Glycemic variability, HbA1c, and complications in adults with Type 1 diabetes. J Clin Endocrinol Metab (2022.0) 107 e570-e581. DOI: 10.1210/clinem/dgab688 34. Zhang L, Xu Y, Jiang X. Impact of flash glucose monitoring on glycemic control varies with the age and residual β-cell function of patients with type 1 diabetes mellitus. J Diabetes Investig (2022.0) 13 552-559. DOI: 10.1111/jdi.13693 35. Nwosu BU, Parajuli S, Khatri K. Partial clinical remission reduces Lipid-based cardiovascular risk in adult patients with Type 1 diabetes. Front Endocrinol (2021.0) 12 706555. DOI: 10.3389/fendo.2021.705565
--- title: 'Analysis of urinary C–C motif chemokine ligand 14 (CCL14) and first-generation urinary biomarkers for predicting renal recovery from acute kidney injury: a prospective exploratory study' authors: - Ben-Shu Qian - Hui-Miao Jia - Yi-Bing Weng - Xin-Cheng Li - Chao-Dong Chen - Fang-Xing Guo - Yu-Zhen Han - Li-Feng Huang - Yue Zheng - Wen-Xiong Li journal: Journal of Intensive Care year: 2023 pmcid: PMC10026399 doi: 10.1186/s40560-023-00659-2 license: CC BY 4.0 --- # Analysis of urinary C–C motif chemokine ligand 14 (CCL14) and first-generation urinary biomarkers for predicting renal recovery from acute kidney injury: a prospective exploratory study ## Abstract ### Background Acute kidney injury (AKI) is a frequent syndrome in the intensive care unit (ICU). AKI patients with kidney function recovery have better short-term and long-term prognoses compared with those with non-recovery. Numerous studies focus on biomarkers to distinguish them. To better understand the predictive performance of urinary biomarkers of renal recovery in patients with AKI, we evaluated C–C motif chemokine ligand 14 (CCL14) and two first-generation biomarkers (cell cycle arrest biomarkers and neutrophil gelatinase-associated lipocalin) in two ICU settings. ### Methods We performed a prospective study to analyze urinary biomarkers for predicting renal recovery from AKI. Patients who developed AKI after ICU admission were enrolled and urinary biomarkers including tissue inhibitor of metalloproteinase-2 (TIMP-2), insulin-like growth factor-binding protein 7 (IGFBP7), CCL14, and neutrophil gelatinase-associated lipocalin (NGAL) were detected on the day of AKI diagnosis. The primary endpoint was non-recovery from AKI within 7 days. The individual discriminative ability of CCL14, [TIMP-2] × [IGFBP7] and NGAL to predict renal non-recovery were evaluated by the area under receiver operating characteristics curve (AUC). ### Results Of 164 AKI patients, 64 ($39.0\%$) failed to recover from AKI onset. CCL14 showed a fair prediction ability for renal non-recovery with an AUC of 0.71 ($95\%$ CI 0.63–0.77, $p \leq 0.001$). [ TIMP-2] × [IGFBP7] showed the best prediction for renal non-recovery with an AUC of 0.78 ($95\%$ CI 0.71–0.84, $p \leq 0.001$). However, NGAL had no use in predicting non-recovery with an AUC of 0.53 ($95\%$ CI 0.45–0.60, $$p \leq 0.562$$). A two-parameter model (non-renal SOFA score and AKI stage) predicted renal non-recovery with an AUC of 0.77 ($95\%$ CI 0.77–0.83, $$p \leq 0.004$$). When [TIMP-2] × [IGFBP7] was combined with the clinical factors, the AUC was significantly improved to 0.82 ($95\%$ CI 0.74–0.87, $$p \leq 0.049$$). ### Conclusions Urinary CCL14 and [TIMP-2] × [IGFBP7] were fair predictors of renal non-recovery from AKI. Combing urinary [TIMP-2] × [IGFBP7] with a clinical model consisting of non-renal SOFA score and AKI stage enhanced the predictive power for renal non-recovery. Urinary CCL14 showed no significant advantage in predicting renal non-recovery compared to [TIMP-2] × [IGFBP7]. ### Supplementary Information The online version contains supplementary material available at 10.1186/s40560-023-00659-2. ## Background Acute kidney injury (AKI) is a frequent complication of critical illness, resulting in increased short-term and long-term mortality, significant healthcare costs, and higher risks of chronic kidney disease (CKD) and end-stage renal disease [1–4]. Moreover, many studies indicated that the pattern of AKI recovery affected the prognosis and outcomes [5, 6]. Preventing the non-recovery of renal function should become the therapeutic target of AKI. Therefore, an early biomarker for AKI recovery is needed. Among AKI biomarkers, urine neutrophil gelatinase-associated lipocalin (NGAL) and the recent combination of urine tissue inhibitor of metalloproteinases-2 and insulin-like growth factor-binding protein 7 ([TIMP-2] × [IGFBP7]) are two first-generation biomarkers that can be used to detect kidney damage and predict AKI before serum creatinine [7–9]. However, only a few studies have assessed the performance of [TIMP-2][IGFBP7] as prognosis markers for non-recovery within 48 h or at discharge in patients following cardiac surgery or patients at surgical ICU [10, 11]. Meanwhile, urine C–C motif chemokine ligand 14 (CCL14) was recently reported to have a good even excellent performance in predicting persistent KDIGO stage 3 AKI, with areas under the receiver operating characteristic curves (AUCs) from 0.81 to 0.93 [12–14]. CCL14 is a kind of chemokine released from tubular epithelial cells after injury. CCL14 binds with the C–C chemokine receptors type 1, C–C chemokine receptors type 5, and C–C chemokine receptors type 3 on monocytes and T cells [12, 15]. Renal monocyte recruitment and activation affected the mechanisms of inflammation and fibrosis in kidney tissue damage [16]. Previous work has shown that CCL14 is one of inflammatory markers mediating the risk of progression to end-stage renal disease in diabetes [17]. Hence, CCL14 as an independent predictor of renal recovery from AKI is biologically plausible. However, no study has explored the predictive role of urine CCL14 for renal non-recovery from AKI. Now, we for the first time report an exploratory comparison of urine CCL14 and first-generation urinary biomarkers in predicting non-recovery in critically ill patients with AKI. ## Study design and ethics The study is a prospective exploratory study designed to assess the predictive value of urinary biomarkers for renal non-recovery from AKI. The study conformed to the provisions of the Declaration of Helsinki. Ethical approval was obtained from the Human Ethics Committee of Beijing Chao-yang Hospital, Capital Medical University (ethics number 2018-117). Written informed consent was obtained from patients or their delegates. Study design and manuscript preparation followed the Standards for Reporting Diagnostic Accuracy (STARD) statement [18]. ## Participants The present study was performed in two ICUs at different Chinese tertiary hospitals. We screened patients admitted to the ICUs from October 2020 to May 2022. Patients with new-onset AKI were prospectively and consecutively enrolled. Exclusion criteria included: [1] age < 18 years; [2] established AKI before ICU admission; and [3] failure to obtain adequate urine samples. All enrolled patients adhered to the same management principles as follows: the KDIGO bundle consisting of optimization of volume status, maintenance of perfusion pressure, discontinuation of nephrotoxic drugs and prevention of hyperglycemia [19]; active treatment of primary disease and complications; and the same treatment principles using antibiotics, nutritional metabolism, and organ support. Furthermore, angiotensin-converting enzyme inhibitors/angiotensin receptor blockers and nonsteroidal anti-inflammatory drugs would be discontinued [20]. ## Biomarker measurements Urine samples were collected as soon as AKI was diagnosed. After centrifugation at 3000 rpm for 10 min at 4 °C, the supernatant urine was stored and frozen at − 80 °C until analyzed. The [TIMP-2] × [IGFBP7] was measured using a commercially available NephroCheck Test (Astute Medical, San Diego, CA, USA). NGAL and CCL14 were measured by enzyme-linked immunosorbent assay (ab119600 (NGAL); ab272201 (CCL14), Abcam, UK). The technicians measuring biomarkers were blinded to the clinical data and the physicians responsible were blinded to the results of biomarkers test. ## Outcomes and definitions The diagnosis of AKI was based on changes in the serum creatinine (SCr) or urine output proposed by the KDIGO guidelines, meeting any of the following: [1] increase in SCr ≥ 0.3 mg/dl (≥ 26.5 µmol/L) within 48 h; [2] increase in SCr to ≥ 1.5 times baseline, which was known or suspected to have occurred within 7 days in the past; [3] urine output < 0.5 mL/kg/h for more than 6 h [21]. Baseline creatinine levels were obtained as follows: if more than 5 values were obtained within 6 months to 7 days prior to admission, the median of all values available was used. Otherwise, the lowest value during the 7 days before admission was used. Assuming a baseline glomerular filtration rate (GFR) of 75 mL/min/1.73 m2, the missing baseline creatinine was estimated using the Modification of Diet in Renal Disease formula* [22, 23]. CKD was defined as an estimated GFR of less than 60 mL/min/1.73 m2 for at least 3 months according to the National Kidney Foundation [24]. The primary endpoint was a failure to recover from AKI within 7 days. We defined renal recovery as the lack of any stage of AKI according to either SCr or urine output criteria. For example, patients with AKI stage 2 had to have a decrease in SCr to below $150\%$ of baseline and be absent in the phase of oliguria (urine output < 0.5 mL/kg/h) for more than 6 h. Patients needing kidney replacement therapy (KRT) on day 7 and those who died after AKI within 7 days were considered renal non-recovery, as renal reversal without survival is rare [5]. For patients diagnosed with both SCr and urine output criteria, the AKI stage was recorded as the more severe one. The secondary endpoints were the initiation of KRT in the ICU stay, hospital mortality, and 30-day mortality. The KRT was initiated if patients met at least one of the indications (Additional file 1: Table S1) [25]. ## Sample size calculation The formula calculating the sample size for a cohort study was used 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}$$n=\frac{{\left({Z}_{\alpha }\sqrt{2\overline{pq} }+{Z}_{\beta }\sqrt{{p}_{0}{q}_{0}+{p}_{1}{q}_{1}}\right)}^{2}}{{\left({p}_{1}-{p}_{0}\right)}^{2}},$$\end{document}n=Zα2pq¯+Zβp0q0+p1q12p1-p02,where Z was a statistical value, p1 and p0 represented the expected incidence of the exposure group and the non-exposure group, respectively, q0 = 1 − p0, q1 = 1 − p1, ‾p was the average of the two incidences, $q = 1$ − p, α = 0.05 and the power (1 − β) was $90\%$. Based on our pretest results, the incidence of renal non-recovery was 0.61 for the exposed group (urine CCL14 levels above the threshold) and 0.23 for the non-exposed group (urine CCL14 levels above the threshold). According to the formula mentioned above, the sample size calculated was 68. The same formula was used to calculate sample sizes for [TIMP-2] × [IGFBP7] and NGAL, which were 36 and 154, respectively. Selecting the largest one of the three, the sample size for this study was 154. Considering the loss rate to follow-up (about $5\%$), the estimated total sample size was 154 + (154 × $5\%$) = 162. ## Data collection The ICU critical care platform in the hospital assisted in prospective collection of clinical data, including patient demographics, prior health history, diagnosis, comorbidities, mechanical ventilation, and use of vasopressors. Patient severity was estimated by Acute Physiology and Chronic Health Evaluation (APACHE II) and Sequential Organ Failure Assessment (SOFA) scores on the day of AKI diagnosis (day 0). Serum creatinine was detected and recorded at ICU admission and every 12 h thereafter until day 7 after AKI. Urine output was measured hourly from the catheter in the ICU period and recorded in the ICU critical care platform. Moreover, death during hospital stay and 30 days after AKI establishment were followed up. ## Statistical analysis SPSS Statistics 24 and MedCalc software were performed for statistical analysis. Categorical variables were described as percentiles, and continuous variables were described as mean ± standard deviation (SD) or median (Q1, Q3). Continuous data between the two groups (recovery group and non-recovery group) were compared using t tests or Mann–Whitney U tests, and categorical variables were compared using Chi-square tests or Fisher’s exact tests. We assessed correlations using Spearman’s rank correlation coefficients. For all analyses, a two-sided $p \leq 0.05$ was considered statistically significant. The predictive discrimination of each marker and related model was assessed using the area under the receiver-operating characteristic (ROC) curve (AUC). The following values were utilized to describe the AUC: 0.90–1.0, excellent; 0.80–0.89, good; 0.70–0.79, fair; 0.60–0.69, poor; and 0.50–0.59, useless. We used the Youden index to determine the optimal cutoff for calculations of specificity, and sensitivity. Confidence intervals for the AUCs and pairwise comparisons of AUCs were calculated by Delong’s method. The associations of clinical variables on day 0 with renal non-recovery were evaluated by multivariate logistic regression analysis using a stepwise forward-selection procedure. According to the previous review, clinical parameters associated with renal non-recovery such as age, CKD, comorbidity, illness severity, medical admission, and severity of AKI were included in univariate analysis (t tests or Chi-square tests) [7]. Clinical parameters with $p \leq 0.15$ in the univariate analysis were included in the multivariate logistic regression analysis. Ordinal variables were directly entered into regression analysis. Continuous variables were transformed into categorical variables and regarded as dummy variables. Variables with $p \leq 0.05$ in the multivariate logistic regression were independent risk factors for renal non-recovery. The sample size of logistic regression analysis met at least 10 events per variable [26]. The contribution of biomarkers to clinical prediction was further investigated by net reclassification improvement (NRI) and integrated discrimination improvement (IDI) methods. ## Overall patient characteristics and outcomes During the study period, 1365 critically ill patients were screened, and 181 ($13.3\%$) of them developed AKI. 164 patients with AKI were ultimately enrolled after excluding 17 ineligible patients. All of the patients were Asian. 100 ($61.0\%$) patients had renal recovery from AKI occurrence within 7 days, and 64 ($39.0\%$) patients encountered renal non-recovery (Fig. 1).Fig. 1Study flow diagram. ICU intensive care unit, AKI acute kidney injury Baseline clinical characteristics and outcomes are shown in Table 1. There were no significant differences in demographic characteristics, comorbidities, and nephrotoxic drug use. However, patients who failed to recover had remarkably higher APACHE II scores and non-renal SOFA scores than those who successfully recovered. Besides, in terms of kidney features, the distribution of serum creatinine, eGFR, urine output, and AKI stage at enrollment showed significant differences between patients with and without renal recovery. In the non-recovery group, a more proportion of patients were postoperative and emergency sources. Table 1Patient characteristics between patients with and without renal recoveryRecovery ($$n = 100$$)Non-recovery ($$n = 64$$)p valueAge (years)60 [46, 72]56 [43, 66]0.259Male gender65 (65.0)37 (57.8)0.410Body mass index (kg/m2)22.9 (20.6, 25.4)23.9 (20.4, 27.0)0.162APACHE II score13.0 [11, 17]17 [12, 22]0.014Non-renal SOFA score4 [2, 7]6 [3, 10]0.026Hemoglobin (g/L)90.0 (82, 102.5)91.0 (77, 109.0)0.499Chronic comorbidities Diabetes18 (18.0)12 (18.8)1.000 Hypertension33 (33.0)24 (37.5)0.615 COPD5 (5.0)3 (4.7)1.000 Coronary artery disease16 (16.0)7 (10.9)0.490 CKD6 (6.0)7 (10.9)0.374 Chronic liver disease33 (33.0)25 (39.1)0.504ACEI/ARB3 (3.0)6 (9.4)0.156Reason for ICU admission Surgical62 (62.0)50 (78.1)0.030 Emergency24 (24.0)5 (7.8)0.008 Medical14 (14.0)9 (14.1)0.725Mechanical ventilation84 (84.0)51 (79.7)0.532PaO2/FiO2302.5 (192.0, 420.0)288.3 (204.3, 413.7)0.562Sepsis27 (27.0)16 (25.0)0.776Vasopressors22 (22.0)10 (15.6)0.420Nephrotoxic drugs use11 (11.0)8 (12.5)0.770Baseline creatinine(µmol/L)65.0 (53.9, 76.3)65.4 (53.0, 76.0)0.915Serum creatinine diagnosing AKI (µmol/L)113.0 (90.9, 144.0)135.6 (105.1, 205.1)0.001eGFRa (mL/min/1.73 m2)58.6 ± 20.746.1 ± 22.90.001UO 24 h after diagnosing AKI (mL/kg/h)0.42 (0.36, 0.47)0.27 (0.25, 0.36)0.002AKI classification Stage 173 (73.0)18 (28.1)< 0.001 Stage 222 (22.0)32 (50.0) Stage 35 (5.0)14 (21.9)Outcomes KRT in ICU9 [9]19 (30.6)0.001 Hospital mortality12 [12]13 (20.3)0.183 30-Day mortality11 [11]11 (17.2)0.348Values are median (Q1, Q3), mean ± SD or n (%). Nephrotoxic drug primarily includes vancomycin, aminoglycosides, rifampicin, amphotericin B, immunosuppressants and chemotherapy. aGlomerular filtration rate was estimated by the Modification of Diet in Renal Disease formulaAPACHE II Acute Physiology and Chronic Health Evaluation, SOFA Sequential Organ Failure Assessment, COPD chronic obstructive pulmonary disease, CKD chronic kidney disease, ACEI angiotensin-converting enzyme inhibitors, ARB angiotensin receptor blocker, ICU intensive care unit, eGFR estimated glomerular filtration rate, UO urine output, AKI acute kidney injury, KRT kidney replacement therapy 19 ($30.6\%$) patients in renal non-recovery patients received KRT and 9 ($9.0\%$) in renal recovery patients. Both hospital mortality and 30-day mortality were similar between the two subsets ($12.0\%$ vs. $20.3\%$, $$p \leq 0.183$$, and $11.0\%$ vs. $17.2\%$, $$p \leq 0.348$$, respectively). ## Relationship between biomarker levels at enrollment and renal recovery We compared urinary biomarker levels between the recovery and non-recovery groups (Fig. 2). Significant differences were observed in urinary biomarkers levels of [TIMP-2][IGFBP7] and CCL14. The patients with renal recovery showed levels of [TIMP-2] × [IGFBP7] and CCL14 were 0.12 (0.05, 0.37) [(ng/mL)$\frac{2}{1000}$] and 245.77 (80.15, 760.13) pg/mL, respectively. However, patients who failed to recover showed higher concentrations of [TIMP-2] × [IGFBP7] and CCL14 which were 1.09 (0.30, 1.28) [(ng/mL)$\frac{2}{1000}$] and 963.01 (359.91, 1531.04) pg/mL, respectively. Unfortunately, there was no significant difference in NGAL levels between the recovery group and non-recovery group ($$p \leq 0.548$$) and multivariate logistic regression analysis of biomarkers revealed that elevated urine NGAL was not an independent risk factor for renal recovery. Fig. 2Urinary biomarkers levels on day 0 stratified by renal recovery. Day 0 means the day of AKI diagnosis. TIMP-2 tissue inhibitor of metalloproteinases-2, IGFBP-7 insulin-like growth factor-binding protein 7, CCL14 C–C motif chemokine ligand 14, NGAL neutrophil gelatinase-associated lipocalin ## Urinary biomarkers and prediction of renal recovery [TIMP-2] × [IGFBP7] alone predicted renal non-recovery from AKI with an AUC of 0.78 ($95\%$ CI 0.71–0.84, $p \leq 0.001$). Its optimal threshold value was 0.72 [(ng/mL)$\frac{2}{1000}$] with a sensitivity of $65.6\%$ and a specificity of $85.0\%$ for predicting renal non-recovery. CCL14 alone predicted renal non-recovery from AKI with a lower AUC of 0.71 ($95\%$ CI 0.63–0.77, $p \leq 0.001$). Its optimal cutoff value was 625.69 pg/mL, with a sensitivity of $65.6\%$ and a specificity of $72.0\%$. NGAL alone, however, had no predictive value for renal non-recovery with an AUC of 0.53 ($95\%$ CI 0.45–0.60, $$p \leq 0.562$$). For the multiple biomarkers model, combing CCL14 with [TIMP-2] × [IGFBP7] failed to help improve predictive performance, with a lower AUC than the [TIMP-2] × [IGFBP7] alone (Table 2).Table 2Predictive accuracy of urinary biomarkers on day 0 for renal non-recoveryUrinary BiomarkersAUCp valueCutoffSensitivity (%)Specificity (%)($95\%$ CI)($95\%$ CI)($95\%$ CI)[TIMP-2] × [IGFBP7]0.78< 0.0010.72 (ng/mL)$\frac{2}{100065.685.0}$(0.71–0.84)(52.7–77.1)(76.5–91.4)CCL140.71< 0.001625.69 (pg/mL)65.672.0(0.63–0.77)(52.7–77.1)(62.1–80.5)NGAL0.530.56237.69 (pg/mL)34.481.0(0.45–0.60)(22.9–47.3)(71.9–88.2)[TIMP-2] × [IGFBP7] + CCL140.77< 0.0010.22490.651.0(0.70–0.83)(80.7–96.5)40.8–61.1Day 0 means the day of AKI diagnosis. AUC area under the receiver operating characteristic, CI confidence interval, TIMP-2 tissue inhibitor of metalloproteinases-2, IGFBP-7 insulin-like growth factor-binding protein 7, CCL14 C–C motif chemokine ligand 14, NGAL neutrophil gelatinase-associated lipocalin According to previous studies, the widely used cutoff values for [TIMP-2] × [IGFBP7] to diagnose AKI were 0.3 ng/mL$\frac{2}{1000}$ for high sensitivity and 2.0 ng/mL$\frac{2}{1000}$ for high specificity [27, 28]. We tried to apply those cutoff values to predict renal non-recovery. When the cutoff was 2.0 ng/mL$\frac{2}{1000}$, no ability to predict renal non-recovery was observed. The sensitivity and negative predictive value (NPV) improved when the cutoff was modified to 0.3 ng/mL$\frac{2}{1000}$ (Additional file 1: Table S2). We also explored whether urinary biomarkers could predict the incidence of secondary endpoints (Additional file 1: Table S3). Only urinary CCL14 predicted the initiation of KRT in ICU (AUC = 0.70, $$p \leq 0.001$$). None of the three biomarkers had the ability to predict 30 days and in-hospital mortality. ## Clinical risk prediction models for renal recovery The univariate analyses showed that the APACHE II score, non-renal SOFA score, surgical and emergency reasons for ICU admission, serum creatinine, eGFR, urine output, and AKI stage might be risk factors for renal non-recovery (Table 1). There was a linear correlation between the APACHE II score and the non-renal SOFA score ($r = 0.352$, $p \leq 0.001$). According to a previous study, the non-renal SOFA score had better predictive value; therefore, the non-renal SOFA score was included in the multivariate logistic regression analysis [29]. Furthermore, urine output, eGFR, and serum creatinine levels were significantly associated with the AKI stage ($r = 0.448$, $$p \leq 0.002$$; r = − 0575, $p \leq 0.001$; and $r = 0.6$, $p \leq 0.001$, respectively) and determined AKI stage in clinical practice, so the AKI stage were included in multivariate analysis. The multivariate analyses showed the non-renal SOFA score and AKI stage were independent risk factors for renal non-recovery (Additional file 1: Table S4). We used any one of the factors or both two factors to construct clinical models for comparisons to find the best prediction model. The clinical risk prediction model joining the non-renal SOFA score with the AKI stage demonstrated the best AUC of 0.77 ($95\%$ CI 0.77–0.83, $$p \leq 0.004$$) for predicting renal non-recovery (Table 3).Table 3Stepwise analysis for prediction of non-recovery from AKICharacteristicModelAUC ($95\%$ CI)p valueNon-renal SOFA scoreA0.60 (0.52–0.68)0.037AKI Stage 2B0.74 (0.66–0.80)< 0.001Stage 3A + B0.77 (0.70–0.83)0.004CI confidence interval, AUC area under the receiver operating characteristic, SOFA Sequential Organ Failure Assessment, AKI acute kidney injury ## Combining clinical and urinary biomarkers data To determine the contributions of biomarkers when added to the clinical model, we compared the AUCs for models with and without biomarkers (Fig. 3). When [TIMP-2] × [IGFBP7] was combined with the clinical prediction model to predict renal non-recovery, the power was significantly improved, resulting in the best predictive AUC of 0.82 ($95\%$ CI 0.74–0.87, $$p \leq 0.049$$). When CCL14 was combined with the clinical prediction model, the power (AUC = 0.80, $95\%$ CI 0.73–0.86, $$p \leq 0.125$$) failed to enhance. Fig. 3ROC curves of clinical model and corresponding urinary biomarkers model on day 0 for predicting renal non-recovery. Day 0 means the day of AKI diagnosis. aCompared with clinical model. Clinical model included non-renal SOFA score and AKI stage. AUC area under the receiver operating characteristics curve, CCL14 C–C motif chemokine ligand 14, TIMP-2 tissue inhibitor of metalloproteinases-2, IGFBP-7 insulin-like growth factor-binding protein 7 We also assessed the capability of [TIMP-2][IGFBP7] and CCL14 to reclassify the degree of risk of recovery and non-recovery. Multivariate logistic regression analysis was used to calculate the probability of renal non-recovery based on models without and with biomarkers. Patients were stratified into three prespecified groups of “low,” “intermediate,” and “high” probability groups based on prediction models without biomarkers using cutoffs of < $30\%$, 30–$60\%$, and > $60\%$, respectively. Then, we compared the proportions of reclassified patients across these three groups when biomarkers were introduced into models. For [TIMP-2] × [IGFBP7], $10.7\%$ of patients were correctly reclassified into risk prediction categories by the biomarker-introduced model compared with the clinical model alone ($$p \leq 0.042$$). The IDI of [TIMP-2][IGFBP7] for renal non-recovery prediction was $8.0\%$ ($$p \leq 0.003$$). However, the addition of CCL14 to the clinical model was unable to achieve significant improvement in predicting renal function reversal. Adding the two biomarkers at the same time to the clinical model also failed to improve renal non-recovery prediction (Table 4).Table 4NRI and IDI for assessing the contributions of different biomarkers for non-recovery prediction when combing with clinical modelModelsNRI (%)p valueIDI (%)p value[TIMP-2] × [IGFBP7] + clinical model vs. clinical model10.70.0428.00.003CCL14 + clinical model vs. clinical model3.80.5683.20.033CCL14 + [TIMP-2] × [IGFBP7] + clinical model vs. clinical model9.70.2319.20.007NRI net reclassification improvement, IDI integrated discrimination improvement, TIMP-2 tissue inhibitor of metalloproteinases-2, IGFBP-7 insulin-like growth factor-binding protein 7, CCL14 C–C motif chemokine ligand 14 ## Sensitivity analysis Of 164 patients, 91 ($55.5\%$) were diagnosed with AKI stage 1 and 73 ($44.5\%$) were diagnosed with AKI at stages 2–3. We repeated the risk prediction analyses after removing patients with AKI stage 1. Urine [TIMP-2] × [IGFBP7] had a good predictive value in patients with AKI stages 2–3 and Urine CCL14 had a fair predictive value in patients with AKI stages 2–3 (Additional file 1: Table S5). ## Key findings We performed a prospective study to explore the relationship between urinary biomarkers and renal non-recovery in critically ill patients with AKI and analyze the accuracy of a novel urinary biomarker (CCL14) and first-generation urinary biomarkers ([TIMP-2] × [IGFBP7] and NGAL) for predicting non-recovery. In this cohort, we found that both CCL14 and [TIMP-2] × [IGFBP7] levels were higher in patients who failed to recover from AKI. Higher CCL14 and [TIMP-2] × [IGFBP7] levels were independently associated with renal non-recovery. However, only the addition of [TIMP-2] × [IGFBP7] to the clinical model significantly improved predictive performance for renal non-recovery. Urine CCL14 failed to exceed [TIMP-2] × [IGFBP7] in the prediction of renal non-recovery. Urine NGAL may not be promising in predicting renal recovery in critically ill patients with AKI. ## Relationship to previous studies To our knowledge, this is the first investigation of urine CCL14 in critically ill patients as a predictor of non-recovery from AKI. The previous study mainly concentrated on persistent AKI lasting more than 72 h, and the outcomes were also correlated with higher ill severity scores (non‑renal APACHE III score) and AKI stage at enrollment [12, 13]. Frustratingly, the AUC of CCL14 in the present study was lower than in previous studies. This may be related to the exclusion of patients with AKI stage 1 in the RUBY and SAPPHIRE studies. Functional AKI, meaning no damage or stress to the kidney, may occur more frequently in our setting and thus may explain the underperformance of the urine CCL14. Nevertheless, $28.1\%$ of patients with AKI stage 1 in our study indeed experienced a renal non-recovery within 7 days, and our results provide a lesson for this population. Larger cohort studies are needed to validate the value of urine CCL14 in predicting renal recovery in AKI stage 1, or we can follow the example of Koyner et al. and find an appropriately raised standardized cut-off to achieve high specificity in identifying patients at high risk of renal non-recovery [30]. Moreover, our analysis essentially replicates the performance of urine [TIMP-2] × [IGFBP7] with the results of prior studies. When it was added to the clinical model consisting of non-renal SOFA score and AKI stage, the performance of predicting renal non-recovery was improved (AUC increased to 0.82) and further confirmed by NRI and IDI analysis. These results support the use of urine [TIMP-2][IGFBP7] for stratification of AKI patients in the ICU. Currently, there is no definitive cutoff value for urinary CCL14 in the prediction of renal non-recovery. For urine CCL14, its cutoff value for predicting renal non-recovery is lower than previously reported values in the present study [14, 30]. The racial disparity may be linked to the threshold. For example, Stanley et al. observed variation in urine biomarkers of lupus nephritis across ethnicities [31]. All included patients in this study were Asian, which differed from European and American in the previous study. Another reason may be the selection of biomarker test kit. The NEPHROCLEAR CCL14 Test used in the previous study was different from the enzyme-linked immunosorbent assay kit used in this study. It is necessary for us to design a trial to explore racial disparity in urinary CCL14 using the same test kit in the future. Urine NGAL has been proven to have good utility for predicting short-term and long-term kidney function reversal [32–34]. Yet, no significant difference in urine NGAL levels existed between the recovery and non-recovery groups in the present study (Fig. 2). We speculate that this is because NGAL is released by kidney epithelial cells and activated neutrophils during systemic inflammation [34], whereas the mentioned studies either included more than half of septic or infective patients or enrolled septic AKI participants. It is reasonable to measure inflammation-related AKI by inflammatory indicators, but in some studies similar to the present cohort, involving a majority of patients who underwent surgery before AKI, urine NGAL may be useless predictors for renal recovery from AKI [11, 35]. On the other hand, an earlier study has revealed that urine NGAL can be stored stably at − 80 °C for 6 months [36]. However, our study lasted more than one year, resulting in longer term storage of urinary samples. The instability of Urine NGAL during long-term storage may impact the discrimination of renal recovery [37]. ## Implications of study findings First, biomarkers are earlier than the elevated serum creatinine and oliguria, allowing a window of time when interventions might be able to prevent further injury. Second, our findings may have implications for design of intervention studies for AKI patients with high risk of renal non-recovery. Given the excellent performance of urine [TIMP-2] × [IGFBP7] in predicting AKI at high-risk populations, Zarbock et al. considered urine [TIMP-2] × [IGFBP7] ≥ 0.3 as inclusion criteria for intervention trial of the KDIGO bundle in a multicenter-center study enrolling 278 patients undergoing cardiac surgery and found a reduced occurrence of moderate and severe AKI in the intervention group compared with the control group [38]. Similarly, biomarker-directed intervention trials could be designed to confirm the benefit of the bundle interventions in patients with high risk of renal non-recovery. Third, our findings may have implications for clinical management of patients with AKI. Individuals with a high likelihood of recovery identified by biomarkers can have a regular dose of medication and no need for invasive monitoring. Therefore, early transfer out of the ICU is possible to reduce ICU-related complications, such as delirium, which was associated with greater mortality within 30 days of discharge [39]. ## Strengths and limitations Our study has several strengths. CKD is a risk factor for AKI, this study included not only new-onset AKI patients but also AKI patients with worsening preexisting CKD. Second, we tested multiple urinary biomarkers, which allowed us to visually compare a novel biomarker (CCL14) with first-generation markers. Third, the added value of [TIMP-2] × [IGFBP7] in renal non-recovery prediction was consistently found in multiple analyses, increasing our findings’ robustness. Our study has, however, limitations. Although regression analysis showed two clinical parameters were independent risk factors for renal non-recovery in our study, the sample size was relatively small. We did not assess the predictive accuracy of urinary biomarkers in validation cohorts. Therefore, further studies will be needed to validate it. Moreover, this study included critically ill patients from two ICUs, which limits the generalizability of the findings. It was, however, performed in two tertiary hospitals, suggesting some degree of external validity for similar hospitals. In addition, we tested urinary biomarkers only on the day of AKI diagnosis, thus were unable to compare the kinetics of the three urinary biomarkers. We assessed the short-term prognosis but ignored the long-term prognosis, it would be better if we had explored the relationship between urinary biomarkers and the long-term prognosis of AKI. ## Conclusion Urine CCL14 and [TIMP-2][IGFBP7] were fair predictors of renal non-recovery from AKI. Combing urine [TIMP-2] × [IGFBP7] with a clinical model of non-renal SOFA score and AKI stage enhanced the predictive performance for renal non-recovery. Urine CCL14 showed no significant advantage in predicting renal non-recovery compared to [TIMP-2] × [IGFBP7]. ## Supplementary Information Additional file 1. Table S1. Indications for initiation of KRT in patients with acute kidney injury. Table S2. Predictive accuracy of [TIMP-2][IGFBP7] for renal non-recovery at different cutoff values. Table S3. Predictions of secondary outcome of urinary biomarkers on day 0. Table S4. Multivariable regression analysis of clinical variables. Table S5. Urinary biomarkers on day 0 for predicting renal non-recovery in patients with AKI stage 2-3. ## References 1. Collister D, Pannu N, Ye F, James M, Hemmelgarn B, Chui B, Manns B, Klarenbach S. Health care costs associated with AKI. Clin J Am Soc Nephrol (2017) 12 1733-1743. DOI: 10.2215/CJN.00950117 2. Noble RA, Lucas BJ, Selby NM. Long-term outcomes in patients with acute kidney injury. Clin J Am Soc Nephrol (2020) 15 423-429. DOI: 10.2215/CJN.10410919 3. James MT, Bhatt M, Pannu N, Tonelli M. Long-term outcomes of acute kidney injury and strategies for improved care. Nat Rev Nephrol (2020) 16 193-205. DOI: 10.1038/s41581-019-0247-z 4. Vijayan A, Abdel-Rahman EM, Liu KD, Goldstein SL, Agarwal A, Okusa MD, Cerda J. Recovery after critical illness and acute kidney injury. Clin J Am Soc Nephrol (2021) 16 1601-1609. DOI: 10.2215/CJN.19601220 5. Kellum JA, Sileanu FE, Bihorac A, Hoste EAJ, Chawla LS. Recovery after acute kidney injury. Am J Respir Crit Care Med (2017) 6 784-791. DOI: 10.1164/rccm.201604-0799OC 6. Pannu N, James M, Hemmelgarn B, Klarenbach S. Association between AKI, recovery of renal function, and long-term outcomes after hospital discharge. Clin J Am Soc Nephrol (2013) 8 194-202. DOI: 10.2215/CJN.06480612 7. Forni LG, Darmon M, Ostermann M, Oudemans-van Straaten HM, Pettilä V, Prowle JR, Schetz M, Joannidis M. Renal recovery after acute kidney injury. Intensive Care Med (2017) 43 855-866. DOI: 10.1007/s00134-017-4809-x 8. Gaião SM, Paiva JAOdC. Biomarkers of renal recovery after acute kidney injury. Rev Bras Ter Intensiva. (2017) 29 373-381. DOI: 10.5935/0103-507X.20170051 9. Kulvichit W, Kellum JA, Srisawat N. Biomarkers in acute kidney injury. Crit Care Clin (2021) 37 385-398. DOI: 10.1016/j.ccc.2020.11.012 10. Dewitte A, Joannes-Boyau O, Sidobre C, Fleureau C, Bats ML, Derache P, Leuillet S, Ripoche J, Combe C, Ouattara A. Kinetic eGFR and novel AKI biomarkers to predict renal recovery. Clin J Am Soc Nephrol (2015) 10 1900-1910. DOI: 10.2215/CJN.12651214 11. Meersch M, Schmidt C, Van Aken H, Martens S, Rossaint J, Singbartl K, Görlich D, Kellum JA, Zarbock A. Urinary TIMP-2 and IGFBP7 as early biomarkers of acute kidney injury and renal recovery following cardiac surgery. PLoS ONE (2014) 9 e93460. DOI: 10.1371/journal.pone.0093460 12. Hoste E, Bihorac A, Al-Khafaji A, Ortega LM, Ostermann M, Haase M, Zacharowski K, Wunderink R, Heung M, Lissauer M. Identification and validation of biomarkers of persistent acute kidney injury: the RUBY study. Intensive Care Med (2020) 46 943-953. DOI: 10.1007/s00134-019-05919-0 13. Bagshaw SM, Al-Khafaji A, Artigas A, Davison D, Haase M, Lissauer M, Zacharowski K, Chawla LS, Kwan T, Kampf JP. External validation of urinary C-C motif chemokine ligand 14 (CCL14) for prediction of persistent acute kidney injury. Crit Care (2021) 25 185. DOI: 10.1186/s13054-021-03618-1 14. Massoth C, Küllmar M, Enders D, Kellum JA, Forni LG, Meersch M, Zarbock A. Comparison of C-C motif chemokine ligand 14 with other biomarkers for adverse kidney events after cardiac surgery. J Thorac Cardiovasc Surg (2021) 10 S0022-5223 15. Detheux M, Ständker L, Vakili J, Münch J, Forssmann U, Adermann K, Pöhlmann S, Vassart G, Kirchhoff F, Parmentier M. Natural proteolytic processing of hemofiltrate CC chemokine 1 generates a potent CC chemokine receptor (CCR)1 and CCR5 agonist with anti-HIV properties. J Exp Med (2000) 192 1501-1508. DOI: 10.1084/jem.192.10.1501 16. Meng X-M, Tang PM-K, Li J, Lan HY. Macrophage phenotype in kidney injury and repair. Kidney Dis (2015) 1 138-146. DOI: 10.1159/000431214 17. Niewczas MA, Pavkov ME, Skupien J, Smiles A, Md Dom ZI, Wilson JM, Park J, Nair V, Schlafly A, Saulnier P-J. A signature of circulating inflammatory proteins and development of end-stage renal disease in diabetes. Nat Med (2019) 25 805-813. DOI: 10.1038/s41591-019-0415-5 18. Bossuyt PM, Reitsma JB, Bruns DE, Gatsonis CA, Glasziou PP, Irwig L, Lijmer JG, Moher D, Rennie D, de Vet HCW. STARD 2015: an updated list of essential items for reporting diagnostic accuracy studies. BMJ (2015) 351 h5527. DOI: 10.1136/bmj.h5527 19. Meersch M, Schmidt C, Hoffmeier A, Van Aken H, Wempe C, Gerss J, Zarbock A. Prevention of cardiac surgery-associated AKI by implementing the KDIGO guidelines in high risk patients identified by biomarkers: the PrevAKI randomized controlled trial. Intensive Care Med (2017) 43 1551-1561. DOI: 10.1007/s00134-016-4670-3 20. Moore PK, Hsu RK, Liu KD. Management of acute kidney injury: core curriculum 2018. Am J Kidney Dis (2018) 72 136-148. DOI: 10.1053/j.ajkd.2017.11.021 21. Kellum JA, Lameire N. Diagnosis, evaluation, and management of acute kidney injury: a KDIGO summary (Part 1). Crit Care (2013) 17 204. DOI: 10.1186/cc11454 22. Mårtensson J, Jonsson N, Glassford NJ, Bell M, Martling C-R, Bellomo R, Larsson A. Plasma endostatin may improve acute kidney injury risk prediction in critically ill patients. Ann Intensive Care (2016) 6 6. DOI: 10.1186/s13613-016-0108-x 23. Kashani K, Al-Khafaji A, Ardiles T, Artigas A, Bagshaw SM, Bell M, Bihorac A, Birkhahn R, Cely CM, Chawla LS. Discovery and validation of cell cycle arrest biomarkers in human acute kidney injury. Crit Care (2013) 17 R25. DOI: 10.1186/cc12503 24. Webster AC, Nagler EV, Morton RL, Masson P. Chronic kidney disease. Lancet (2017) 389 1238-1252. DOI: 10.1016/S0140-6736(16)32064-5 25. Tandukar S, Palevsky PM. Continuous renal replacement therapy: who, when, why, and how. Chest (2019) 155 626-638. DOI: 10.1016/j.chest.2018.09.004 26. Peduzzi P, Concato J, Kemper E, Holford TR, Feinstein AR. A simulation study of the number of events per variable in logistic regression analysis. J Clin Epidemiol (1996) 49 1373-1379. DOI: 10.1016/S0895-4356(96)00236-3 27. Fan W, Ankawi G, Zhang J, Digvijay K, Giavarina D, Yin Y, Ronco C. Current understanding and future directions in the application of TIMP-2 and IGFBP7 in AKI clinical practice. Clin Chem Lab Med (2019) 57 567-576. DOI: 10.1515/cclm-2018-0776 28. Hoste EA, McCullough PA, Kashani K, Chawla LS, Joannidis M, Shaw AD, Feldkamp T, Uettwiller-Geiger DL, McCarthy P, Shi J. Derivation and validation of cutoffs for clinical use of cell cycle arrest biomarkers. Nephrol Dial Transplant (2014) 29 2054-2061. DOI: 10.1093/ndt/gfu292 29. Jia HM, Cheng L, Weng YB, Wang JY, Jiang YJ, Zheng X, Xin X, Guo SY, Chen CD, Guo FX. cell cycle arrest biomarkers for predicting renal recovery from acute kidney injury: a prospective validation study. Ann Intensive Care (2022) 12 14. DOI: 10.1186/s13613-022-00989-8 30. Koyner JL, Chawla LS, Bihorac A, Gunnerson KJ, Schroeder R, Demirjian S, Hodgson L, Frey JA, Wilber ST, Kampf JP. Performance of a standardized clinical assay for urinary C-C Motif Chemokine Ligand 14 (CCL14) for persistent severe acute kidney injury. Kidney360. (2022) 3 1158-1168. DOI: 10.34067/KID.0008002021 31. Stanley S, Vanarsa K, Soliman S, Habazi D, Pedroza C, Gidley G, Zhang T, Mohan S, Der E, Suryawanshi H. Comprehensive aptamer-based screening identifies a spectrum of urinary biomarkers of lupus nephritis across ethnicities. Nat Commun (2020) 11 2197. DOI: 10.1038/s41467-020-15986-3 32. Moon SJ, Park HB, Yoon SY, Lee SC. Urinary biomarkers for early detection of recovery in patients with acute kidney injury. J Korean Med Sci (2013) 28 1181-1186. DOI: 10.3346/jkms.2013.28.8.1181 33. Thomsen J, Sprogøe U, Toft P. Urine neutrophil gelatinase-associated lipocalin and urine output as predictors of the successful discontinuation of continuous renal replacement therapy in critically ill patients with acute kidney injury. BMC Nephrol (2020) 21 375. DOI: 10.1186/s12882-020-02035-w 34. Mårtensson J, Bell M, Oldner A, Xu S, Venge P, Martling C-R. Neutrophil gelatinase-associated lipocalin in adult septic patients with and without acute kidney injury. Intensive Care Med (2010) 36 1333-1340. DOI: 10.1007/s00134-010-1887-4 35. Zeng XF, Li JM, Tan Y, Wang ZF, He Y, Chang J, Zhang H, Zhao H, Bai X, Xie F. Performance of urinary NGAL and L-FABP in predicting acute kidney injury and subsequent renal recovery: a cohort study based on major surgeries. Clin Chem Lab Med (2014) 52 671-678. DOI: 10.1515/cclm-2013-0823 36. van de Vrie M, Deegens JK, van der Vlag J, Hilbrands LB. Effect of long-term storage of urine samples on measurement of kidney injury molecule 1 (KIM-1) and neutrophil gelatinase-associated lipocalin (NGAL). Am J Kidney Dis (2014) 63 573-576. DOI: 10.1053/j.ajkd.2013.10.010 37. Haase-Fielitz A, Haase M, Bellomo R. Instability of urinary NGAL during long-term storage. Am J Kidney Dis (2009) 53 564-565. DOI: 10.1053/j.ajkd.2009.01.009 38. Zarbock A, Küllmar M, Ostermann M, Lucchese G, Baig K, Cennamo A, Rajani R, McCorkell S, Arndt C, Wulf H. Prevention of cardiac surgery-associated acute kidney injury by implementing the KDIGO Guidelines in high-risk patients identified by biomarkers: the PrevAKI-multicenter randomized controlled trial. Anesth Analg (2021) 133 292-302. DOI: 10.1213/ANE.0000000000005458 39. Devlin JW, Needham DM. Long-term outcomes after delirium in the ICU: addressing gaps in our knowledge. Am J Respir Crit Care Med (2021) 204 383-385. DOI: 10.1164/rccm.202104-0910ED
--- title: 'Community coalition efforts to prevent childhood obesity: two-year results of the Shape Up Under 5 study' authors: - Christina D. Economos - Larissa Calancie - Ariella R. Korn - Steven Allender - Julia M. Appel - Peter Bakun - Erin Hennessy - Peter S. Hovmand - Matt Kasman - Melanie Nichols - Mark C. Pachucki - Boyd A. Swinburn - Alison Tovar - Ross A. Hammond journal: BMC Public Health year: 2023 pmcid: PMC10026415 doi: 10.1186/s12889-023-15288-5 license: CC BY 4.0 --- # Community coalition efforts to prevent childhood obesity: two-year results of the Shape Up Under 5 study ## Abstract ### Background Cross-sector collaborations and coalitions are promising approaches for childhood obesity prevention, yet there is little empirical evidence about how they affect change. We hypothesized that changes in knowledge of, and engagement with, childhood obesity prevention among coalition members can diffuse through social networks to influence policies, systems, and environments. ### Methods We studied a community coalition ($$n = 16$$, Shape Up Under 5 “SUU5 Committee”) focused on early childhood obesity prevention in Somerville, MA from 2015–17. Knowledge, engagement, and social network data were collected from Committee members and their network contacts ($$n = 193$$) at five timepoints over two years. Policy, systems, and environment data were collected from the SUU5 Committee. Data were collected via the validated COMPACT Stakeholder-driven Community Diffusion survey and analyzed using regression models and social network analysis. ### Results Over 2 years, knowledge of ($$p \leq 0.0002$$), and engagement with ($$p \leq 0.03$$), childhood obesity prevention increased significantly among the SUU5 Committee. Knowledge increased among the Committee’s social network ($$p \leq 0.001$$). Significant changes in policies, systems, and environments that support childhood obesity prevention were seen from baseline to 24 months ($$p \leq 0.003$$). ### Conclusion SUU5 had positive effects on “upstream” drivers of early childhood obesity by increasing knowledge and engagement. These changes partially diffused through networks and may have changed “midstream” community policies, systems, and environments. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12889-023-15288-5. ## Background Almost $14\%$ of 2–5 years old children have obesity in the United States (US) [1]. Children with obesity in early childhood are likely to have obesity as adolescents and adults, increasing their risk for diabetes, hypertension, cardiovascular disease, and certain cancers [2, 3]. Moreover, obesity in this age group suggests the presence of unhealthy eating and physical activity behaviors that could be modified with early interventions [4]. Obesity rates are significantly higher among Hispanic and African American children compared to White children, highlighting health disparities that can develop early in life [5, 6]. Obesity in early childhood is influenced by many interacting factors, including families’ ability to access and afford healthy foods, caregivers’ child-feeding beliefs, access to safe places for active play, childcare center policies and practices, and children’s changing taste preferences [7, 8]. A number of evidence-based obesity prevention policies and interventions have been developed and are readily available for adoption [9, 10]. A systematic review of interventions aiming to promote healthy weights in children 2–5 years old found that interventions that included the following three components were more successful: encouraging parents to praise healthy behaviors in their children, educating parents about the importance of reducing screen time for their children, and engaging healthcare providers in intervention content delivery [11, 12]. However, parents, healthcare providers, and early childcare education professionals report barriers in adopting evidence-based obesity prevention strategies without environmental and policy changes within the broader system [13–15]. Engaging cross-sector collaborations, which include community coalitions and partnerships across multiple sectors (e.g., government, healthcare, community-based organizations) [16–18] are promising approaches for implementing childhood obesity prevention strategies and interventions in community settings [19, 20]. A review of interventions involving community coalitions found elements of Community-based Participatory Research were present in all 13 included studies, and five reported improvements in anthropometric, behavioral, or policy outcomes [20]. The review found successful studies demonstrated the importance of community engagement in community-based childhood obesity prevention research [20]. While engaging cross-sector collaborations and coalitions are promising approaches for addressing early childhood obesity in communities, there is little empirical evidence about how these collaborations work to effect change. Practical, evidence-based theories for how to work across sectors to change policies, systems, and environments (PSE) to reduce early childhood obesity are needed. We previously developed the Stakeholder-driven Community Diffusion (SDCD) theory (Fig. 1), which posits that a multi-sector group can diffuse knowledge and engagement across both existing and new relationship networks, thus facilitating changes in policies, systems, and environments that influence obesity [21–23]. SDCD was informed by Community-based Participatory Research [24], Community Coalition Action Theory [25], Diffusion of Innovations [26], and systems science [27]. In SDCD, knowledge and engagement are critical upstream mechanisms that provide a foundation and motivation for creating long-term, widespread, sustainable change within communities [21, 28]. SDCD was developed as part of the COMPACT study (R01HL115485) that integrated multiple systems science methods to study and support dynamics within community coalitions that fostered the adoption, implementation, and dissemination of evidence-based childhood obesity prevention strategies across multiple community sectors [28]. The goal of SDCD-informed interventions are to catalyze whole-of-community change with multi-level, multicomponent obesity prevention strategies that are implemented in a variety of settings across a community [29].Fig. 1Stakeholder-driven Community Diffusion theory-informed intervention [21] Previous research has provided suggestive evidence that similar interventions in other settings have had measurable impact, and has examined the SDCD causal mechanism with retrospective analyses [30]. The purpose of this study is to conduct a prospective analysis of an SDCD-driven intervention, testing the following three hypotheses: 1) knowledge of and engagement with early childhood obesity prevention efforts will increase among Committee members; 2) knowledge and engagement will increase among Committee members’ networks; and 3) policies, systems, and environments in the community where the Committee works will become more supportive of early childhood obesity prevention efforts over the 2-year study period. ## Shape up under 5 study Shape Up Under 5 (SUU5) was a two-year study (2015–2017) designed to test the SDCD theory as a systems approach to early childhood obesity prevention in Somerville, MA [21]. At the time of the study, the prevalence of obesity among children under five years old participating in the Massachusetts Pediatric Nutrition Surveillance System was $14.4\%$, which was slightly higher than the national obesity prevalence ($13.4\%$) [31]. Central to SUU5 was the formation of a 16-person coalition (“SUU5 Committee”) composed of early childhood education and care, healthcare, parks and recreation, local public health department, and public school system representatives [21]. The Committee met 16 times in total – every 4 to 6 weeks – facilitated by a research team from Tufts University and advised by Community-based System Dynamics (CBSD) experts. CBSD is a method for engaging community members in developing a dynamic model of a complex problem and potential solutions; CBSD is rooted in CBPR and system dynamics [32]. CBSD projects often utilize group model building (GMB), which are structured activities that engage groups to build models of a dynamic problem in order to create a shared understanding of system structures that drive that problem over time and build consensus on how to address the problem [32, 33]. SUU5 employed CBSD and GMB activities because they are promising practices for building a shared vision of how to move forward to address complex social challenges while building systems thinking capabilities among participants [32, 34]. The intervention had three phases: 1) reviewing evidence-based strategies for addressing early childhood obesity; 2) GMB to develop a shared understanding of the systems driving early childhood obesity locally and to prioritize actions to address the problem grounded in community context; and 3) stakeholders taking action to address early childhood obesity in their community, with support from the research team [21]. A detailed description of GMB activities and their effects on participants, and the structure and change of the SUU5 social networks over time are presented elsewhere [35, 36]. Briefly, SUU5 Committee members decided to prioritize early childhood obesity prevention, using evidence-based policy and environmental change strategies, within their own organizations and [35] and the network’s evolving membership provided access to a wide range of resources, ideas, and an ability to broadly disseminate intervention messages [36]. ## Data collection – knowledge and engagement We collected data about perceived knowledge and engagement from SUU5 Committee members. Knowledge and engagement data using the COMPACT SDCD survey; details on the development and validation of the can be found elsewhere [36–38]. The survey was administered at baseline and then every six months online with a total of five measurements over the two-year study period [21]. Knowledge was defined as stakeholders’ perceived understanding of various factors related to early childhood obesity prevention efforts, and comprised of 18 items on a 5-point Likert scale across five domains: (i) the problem of early childhood obesity, (ii) modifiable intervention factors, (iii) their own role and the role of others, (iv) how to intervene sustainably, and (v) available resources [37]. Engagement was defined as a latent construct broadly representing stakeholders’ enthusiasm and agency for addressing early childhood obesity. The engagement scale included 25 items on the same 5-point response scale across the following five domains: (i) dialogue and mutual learning, (ii) flexibility, (iii) influence and power, (iv) leadership and stewardship, and (v) trust [37]. ## Data collection – social networks and diffusion of knowledge and engagement We also surveyed committee members’ early childhood obesity prevention social contacts, and collected the same responses regarding perceived knowledge and engagement to test the diffusion mechanism posited in the SDCD theory. At each SDCD survey timepoint the Committee identified up to 20 people with whom they discussed early childhood obesity prevention in the past three months [37]; nominated discussion partners are referred to as “first-degree alters” [39]. To measure knowledge and engagement, first-degree alters were invited to complete the COMPACT SDCD survey including network data at each round of data collection after they were nominated by a Committee member [36]. ## Data collection – policies, systems, and environments As proposed in the SDCD theory, knowledge and engagement spreads through social networks and promotes the implementation of PSE changes [21]. PSE change was assessed via an online survey of respondents’ perception of policy, systems, and environmental changes in Somerville at three times: baseline, 12 months, and 24 months [21]. The PSE survey was informed by existing surveys assessing the nutrition and physical activity environment in early childhood education settings and the research team’s knowledge of environments in Somerville [40–42]. PSE changes that were captured in the survey included: (i) Somerville’s commitment to different aspects of the built environment (e.g., workplace accommodations for breastfeeding); (ii) current use of public places or programs by Somerville residents (e.g., public parks and swimming pools); (iii) availability of family health materials at workplaces; and (iv) training and support for health promoting activities. There were three survey modules. All Committee members were given module A ($$n = 16$$), which asked questions about the built environment and workplace prioritization of early childhood obesity prevention. Committee members who provided direct service to children and/or families (e.g., pediatrician) were given modules A and B ($$n = 12$$), which included questions about the type and frequency of distribution of resources for clients. Committee members who worked in a childcare setting (e.g., center-based childcare provider) were given modules A, B, and C, which included questions about food and menu policies in childcare settings ($$n = 4$$). We decided which Committee members would receive which modules (i.e., either A only; A and B; or A, B, and C) a priori, based on their occupation at the beginning of the study. The instructions in the survey also suggested that respondents review relevant materials (e.g., food menus, staff manuals, parent handbooks, or other policy documents or guidelines) before they completed the survey. Additional information can be found in Appel 2019, and request for further information about this survey can be directed to the corresponding author. Response options were presented as Likert scales where the scales aligned with the question format (e.g., “In general, how would you rate Somerville’s involvement in the following topics related to prenatal and early childhood health?”). The responses for that are on a 5-point Likert scale. We classified “I do not know” responses as missing. ## Data collection – community campaign reach Through the regular Committee meetings using GMB and other structured activities, the SUU5 Committee determined there was a need for coordinated messaging about early childhood obesity prevention strategies that target multiple audiences (e.g., parents, caregivers, and health service providers) and that resonated with the diverse population in Somerville. To fill this need, the Committee and members of the research team at Tufts University created an evidence-informed community communications campaign called Small Steps: Eat, Play, Sleep (adapted from the 9–5-2–1-0 campaign to include recommendations for children under the age of five [43, 44]). The campaign included posters, three different brochures with age-appropriate recommendations and information for children in three age ranges (birth – nine months, nine months – three years, and three – five years) informational videos, activities for an annual Mayor’s Wellness Challenge and a tip-sheet for health and childcare providers. The poster and brochures were made available in four languages: English, Spanish, Portuguese, and Haitian Creole [21, 35], at least one of which is spoken by about $85\%$ of Somerville residents. The informational videos were made available in English and Spanish. The Committee and research team worked together to ensure that the physical materials were displayed in multiple locations throughout the city, and that the digital resources were available on city websites [45]. To estimate the reach of the campaign, we reviewed logs that the Committee filled out describing where they placed the materials and an estimate of how many people might have seen them and we surveyed first-degree alters nominated by the SUU5 Committee. The reach survey asked participants if they saw campaign materials, what materials they saw, and where the materials were located. ## Data analysis – knowledge and engagement For knowledge and engagement, we calculated domain-specific scores by summing responses within domains using a 5-point Likert scale from strongly disagree [1] to strongly agree [5], dividing by the number of items, and then scaled from 0 to 1 to enable direct comparisons between domains. We also calculated composite scores for knowledge and engagement by averaging domain scores. Scores were calculated for each of the five survey rounds for Committee members and first-degree alters. To examine knowledge and engagement changes over time, we used mixed effects regression models with random intercept and trend allowing individual subject effects to vary over time. Pairwise comparisons of mean knowledge and engagement scores by survey round were made using ANOVA with the Tukey method for type 1 error adjustment. All analyses were conducted with SAS 9.4. ## Data analysis – social networks and diffusion of knowledge and engagement We constructed a “network neighborhood” variable for each Committee member’s cluster of first-degree alters by taking the average knowledge and engagement score of the cluster of alters named by the Committee member [46]. Network neighborhood scores for both knowledge and engagement were calculated at each survey round. All first-degree alters were carried forward to the next round, creating a cumulative network. For any first-degree alter missing a survey round, last observation carried forward was used as a conservative approach to impute missing knowledge and engagement scores. Network diagrams with knowledge and engagement scores were created using the R-igraph package implemented in RStudio (v1.2.5019) [46]. Analysis of potential diffusion patterns beyond the first-degree alters (e.g., further into the community networks) was conducted separately using simulation approaches and is reported in a related paper [47]. ## Data analysis – policies, systems, and environments PSE scores were calculated by summing the raw score for all completed sections and dividing it by the total possible score for the completed sections. Change in PSE scores was analyzed using the same methods as knowledge and engagement change. In addition, the relationship of PSE scores to knowledge and engagement score was analyzed using linear regression, regressing PSE score on time and knowledge or engagement score, adjusting for years of Committee member experience. ## Results The Committee was highly educated ($75\%$ graduate degree), mostly white ($94\%$) and Hispanic ($6\%$), and predominately female ($88\%$). The mean age was 50 years old (range: 29 to 70 years old), and years of experience in their occupations ranged from 4 to 44 years (mean = 20 years). First degree alters were demographically similar; $62\%$ held graduate degrees, $89\%$ were white, $10\%$ were Hispanic, and $85\%$ were female. The mean age was 43 (range: 21 to 74 years old) and occupational experience ranged from 0 to 37 years (mean = 13 years). ## Hypothesis 1: Knowledge and engagement change among committee members Overall knowledge scores were higher at every follow-up (6, 12, 18 and 24 months), compared to baseline, although the difference was not significant after 6 months (Table 1). Increases in knowledge of available resources was significantly higher at every round compared to baseline. Knowledge of the problem, intervention factors, and sustainability were significantly higher than baseline at some, but not all, rounds. Overall engagement increased from baseline to 24 months (β = 0.011 (0.005), $$p \leq 0.03$$). Within engagement domains, influence and power increased (β = 0.029 (0.011), $$p \leq 0.01$$) and flexibility increased between baseline and 12 months (mean difference = 0.089, CI = 0.00087, 0.1776).Table 1Composite- and domain-level knowledge and engagement scores among Shape Up Under 5 Committee members ($$n = 16$$) and first-degree alters over the two-year pilot study period. Mean (SD) knowledge and engagement scores range from 0 (low) to 1 (high). The number of first-degree alters who were surveyed expanded at each measurement time as new alters were nominated by committee membersMeasurement timeCommittee members ($$n = 16$$)Baseline6 months12 months18 months24 monthsβ (SE)p-valueKnowledge0.67 (0.07)0.73 (0.09)0.73 (0.08)0.78 (0.08)0.77 (0.09)0.027 (0.006)0.0002 Problem0.75 (0.11)0.80 (0.10)0.79 (0.12)0.82 (0.11)0.83 (0.10)0.021 (0.007)0.007 Intervention factors0.73 (0.11)0.78 (0.15)0.80 (0.120.80 (0.10)0.78 (0.14)0.014 (0.005)0.020 Roles0.60 (0.14)0.59 (0.19)0.64 (0.12)0.68 (0.15)0.70 (0.13)0.029 (0.011)0.010 Sustainability0.80 (0.13)0.86 (0.11)0.84 (0.11)0.87 (0.10)0.85 (0.10)0.010 (0.008)0.200 Resources0.46 (0.11)0.58 (0.19)0.59 (0.16)0.72 (0.12)0.67 (0.17)0.058 (0.011)0.0002Engagement0.73 (0.14)0.75 (0.12)0.80 (0.12)0.79 (0.11)0.76 (0.13)0.011 (0.005)0.030 Dialogue & mutual learning0.82 (0.12)0.84 (0.14)0.85 (0.12)0.88 (0.11)0.82 (0.11)0.006 (0.006)0.300 Flexibility0.76 (0.16)0.77 (0.15)0.84 (0.13)0.83 (0.12)0.78 (0.16)0.014 (0.009)0.200 Influence & power0.55 (0.30)0.58 (0.24)0.67 (0.23)0.68 (0.21)0.64 (0.29)0.029 (0.011)0.010 Leadership & stewardship0.77 (0.13)0.75 (0.18)0.78 (0.12)0.76 (0.10)0.74 (0.10)-0.002 (0.006)0.600 Trust0.78 (0.15)0.81 (0.11)0.86 (0.14)0.82 (0.11)0.82 (0.11)0.009 (0.007)0.200First degree altersn = 65n = 133n = 165n = 186n = 193Knowledge0.70 (0.10)0.72 (0.10)0.72 (0.09)0.73 (0.11)0.74 (0.11)0.008 (0.002)0.001 Problem0.72 (0.12)0.77 (0.11)0.78 (0.11)0.79 (0.12)0.80 (0.12)0.012 (0.003)0.0001 Intervention factors0.76 (0.11)0.79 (0.13)0.78 (0.12)0.78 (0.13)0.78 (0.13)-0.001 (0.003)0.843 Roles0.60 (0.14)0.66 (0.16)0.63 (0.17)0.64 (0.17)0.67 (0.18)0.007 (0.004)0.090 Sustainability0.82 (0.11)0.80 (0.13)0.83 (0.13)0.82 (0.12)0.83 (0.14)0.004 (0.003)0.153 Resources0.58 (0.15)0.56 (0.15)0.60 (0.16)0.63 (0.20)0.63 (0.22)0.018 (0.005)0.0003Engagement0.76 (0.07)0.80 (0.09)0.77 (0.10)0.78 (0.10)0.78 (0.10)-0.001 (0.002)0.662 Dialogue & mutual learning0.84 (0.08)0.88 (0.11)0.86 (0.11)0.87 (0.13)0.87 (0.12)0.001 (0.003)0.747 Flexibility0.73 (0.11)0.83 (0.11)0.79 (0.15)0.80 (0.13)0.80 (0.14)0.003 (0.003)0.245 Influence & power0.62 (0.16)0.65 (0.18)0.60 (0.19)0.61 (0.22)0.60 (0.22)-0.013 (0.005)0.004 Leadership & stewardship0.79 (0.09)0.82 (0.11)0.78 (0.11)0.79 (0.11)0.80 (0.11)0.002 (0.002)0.427 Trust0.80 (0.10)0.80 (0.11)0.80 (0.12)0.82 (0.13)0.83 (0.14)0.009 (0.003)0.004Betas estimated using a mixed effects regression model with random intercept and trend allowing individual subject effects to vary over timeChanges between baseline and a follow-up time point with significant p-values ($p \leq 0.05$) are bolded$p \leq 0.05$, change from baseline using ANOVA with the Tukey method for type 1 error ## Hypothesis 2: Knowledge and engagement change among first-degree alters Overall knowledge increased among first-degree alters (β = 0.008 (SE = 0.002), $$p \leq 0.001$$) (Table 1); specifically, knowledge of the problem of, and knowledge of resources for early childhood obesity increased significantly. Overall engagement did not significantly increase among first-degree alters. Influence and power significantly decreased during the study period and perceived trust significantly increased (β = 0.009 (SE 0.003), $$p \leq 0.004$$). Knowledge and engagement scores varied within neighborhood networks over time (Fig. 2 and Supplemental Table 1). We did not conduct hypothesis testing on changes within neighborhood networks due to the small size of several networks. The baseline network included 90 stakeholders and 131 ties, and the end of the study network included 217 stakeholders and 356 ties (Fig. 2).Fig. 2Network diagram showing levels of knowledge (top) and engagement (bottom) among network neighborhoods at baseline (left) and the end of the two-year pilot study (right) ## Hypothesis 3: Change in policies, systems, and environments Committee members reported improvements in policies, systems, and environments from baseline to 24 months (difference between means: 0.084; $95\%$ CI: 0.028, 0.140; $$p \leq 0.003$$) (Fig. 3). The PSE survey type (A, A + B or A + B + C) did not affect whether Committee members perceived improvements in PSE. Changes in Committee members’ engagement scores were significantly associated with PSE score changes (correlation coefficient = 0.56, $$p \leq 0.037$$).Fig. 3Box plot showing changes in Committee member ($$n = 16$$) reported policies, systems, and environments supportive of early childhood obesity prevention from baseline to the end of the two-year pilot study ## Reach of an evidence-based community communication campaign A total of 55 first-degree alters completed the campaign reach survey, $73\%$ of whom reported seeing Small Steps: Eat Play Sleep campaign materials. The most frequently viewed campaign assets were the posters ($83\%$), followed by brochures ($75\%$), Mayor’s Wellness Challenge materials ($38\%$), PowerPoint presentations (i.e., tip sheets for health and childcare providers) ($13\%$) and other materials ($8\%$). The most common setting to encounter campaign materials was the organization in which the respondent worked ($58\%$), a school ($38\%$), city of Somerville building ($33\%$), library ($20\%$), healthcare office ($18\%$), playground ($15\%$), WIC office ($13\%$), childcare center ($13\%$), farmers’ market ($10\%$), and playgroup ($10\%$). ## Discussion We found that the SUU5 committee significantly increased knowledge of and engagement with early childhood obesity prevention and that knowledge of the topic increased within the members’ professional networks. With support from the research team, the committee designed, created, and disseminated materials for a community-wide communications campaign that provided consistent, evidence-based early childhood obesity prevention recommendations [35]. Committee members reported significant improvements in policies, systems, and environments in policies, systems, and environments that support early childhood obesity prevention at the conclusion of the two-year study. Changes in Committee members’ engagement were significantly associated with reported PSE improvements. Our upstream intervention was designed to catalyze whole-of-community change by facilitating changes in knowledge, engagement, and relationships across sectors, and by using systems thinking methods to spark new “big picture” ideas for addressing childhood obesity and related disparities. A review of whole-of-community obesity prevention studies found that most studies reported improvements in health outcomes (e.g., reductions in BMI), behaviors (e.g., reduced sugar-sweetened beverage consumption, increased physical activity), and/or psychosocial outcomes (e.g., reduced depressive symptoms) [29], and another study found that whole-of-community obesity prevention approaches may be cost effective [48]. Importantly, a review of whole-of-community obesity prevention studies found that low socioeconomic position groups benefited as much or more compared to high socioeconomic position groups, suggesting that this approach is well-suited for addressing persistent weight-related disparities in developed countries [49]. The SUU5 Committee chose to create a communications campaign that was designed to promote consistent, evidence-based messages about early childhood obesity prevention to caregivers in the many settings where children spend their time. Social marketing is established as part of an ecological approach to childhood obesity prevention and can bolster the effects of other programs and policies operating in a community [50, 51]. The campaign developed by the Committee was an important demonstration that the SDCD-intervention could yield cross-sector collaboration towards evidence-based action that spanned multiple community settings. Members of this research group used data presented in this report as input into an agent-based model that characterizes the interpersonal interactions that collectively comprise diffusion of knowledge and engagement over time [47]. They found that the model could reproduce patterns of knowledge and engagement diffusion presented in this report, providing confidence in the model’s explanatory power. Next, the researchers used the model for two types of extrapolation beyond the data presented here: they projected increases in both knowledge and engagement among the broader community (i.e., including those who were not study participants), and explored several counterfactual scenarios to gain insight into how different implementation strategies might have affected change in community knowledge and engagement. The study conducted by Kasman and colleagues offers an exciting example of how mathematical models can expand the scope of research questions that can be answered using data from a community-based intervention. Based in part on the data and analyses in this report, a future iteration of the agent-based model used by Kasman and colleagues could be extended beyond “upstream” changes in knowledge and engagement to explore the causal mechanisms that might translate these to changes in PSE. ## Strengths Engagement was fostered by using CBSD, GMB, and GMB-like activities during Committee meetings, and a strength of this approach includes facilitation techniques that promote group participation and power leveling (e.g., requesting each participant speaks at least once during activities, participants sharing one idea at a time, preventing dominant voices overtaking conversation) [32]. Engagement was highest among Committee members after meetings where they worked on the campaign design and dissemination plans showing the power of engaging leaders in obesity prevention design [35]. As engagement increased over the course of the intervention, increases in PSE accelerated as well. This finding may reflect Committee members’ perceptions that their communication campaign improved PSEs in Somerville, and/or Committee members’ engagement with the topic of early childhood obesity may have translated into additional improvements in PSEs that support early childhood obesity prevention. ## Limitations The study design did not include a comparison group so we cannot rule out whether other factors beyond the SDCD theory-informed intervention influenced observed changes in knowledge and engagement. Future studies should include a comparison group (e.g., quasi-experimental design with a wait-list comparison group) to isolate the effects of the intervention on participants. The Committee and first degree alters were predominantly highly educated, white, and female, reflecting demographic patterns in the broader fields of public health and nutrition [52, 53]. Efforts to diversify the fields [54] and the stakeholder groups this research group partners with are underway. Missing data from first-degree alters made it difficult to fully capture potential diffusion of knowledge and engagement throughout the community. We addressed this limitation by carrying first-degree alters forward to subsequent waves of data collection and by imputation, allowing alters to remain part of the network from their first nomination. If last observation carried forward was not possible (e.g., if respondent was missing initial survey) mean imputation was used replacing the missing value with the mean of remaining values. Our study underscores a need to for new, creative methods for collecting social network data that provides a more complete network sample. Researchers are experimenting with apps, administrative data, and records to collect network data without asking participants to complete lengthy surveys [55]. While the PSE survey assessed Committee members’ perceptions, more objective measures of PSE change would strengthen future studies. Existing data capturing PSE and child health outcomes could be utilized to assess the effects of SDCD theory-informed interventions and similar coalition-led initiatives. ## Implications These findings highlight a potential mechanism through which community coalitions and similar groups affect positive change related to children’s health in their communities. ## Conclusion SUU5’s SDCD-informed intervention increased knowledge of, and engagement with, early childhood obesity prevention among a multi-sector group of stakeholders, and improvements in those constructs diffused into some professional network clusters. Increases in engagement among the group were also associated with improvements in perception of policies, systems, and engagement related to early childhood obesity prevention in the community. The SDCD theory [21, 22], data collection instruments [20, 38], effects on community-based networks [36, 56], agent-based modeling [47], and interventions advances the science of effective obesity prevention research in communities [35, 57]. ## Supplementary Information Additional file 1: Supplemental Table 1. Knowledge scores among Shape Up Under 5 neighborhood networks at each data collection point over the two-year pilot study. Mean (SD) knowledge and engagement scores range from 0 (low) to 1 (high). Neighborhood networks sizes (n) increase cumulatively over time. Supplemental Table 2. Engagement scores among Shape Up Under 5 neighborhood networks at each data collection point over the two-year pilot study. Mean (SD) knowledge and engagement scores range from 0 (low) to 1 (high). Neighborhood networks sizes (n) increase cumulatively over time. ## References 1. Skinner AC, Ravanbakht SN, Skelton JA, Perrin EM, Armstrong SC. Prevalence of obesity and severe obesity in US children, 1999–2016. Pediatr Am Acad Pediatr (2018.0) 141 1999-2016 2. Kelsey MM, Zaepfel A, Bjornstad P, Nadeau KJ. Age-related consequences of childhood obesity. Gerontology. S. Karger AG (2014.0) 60 222-8 3. Geserick M, Vogel M, Gausche R, Lipek T, Spielau U, Keller E. Acceleration of BMI in Early Childhood and Risk of Sustained Obesity. N Engl J Med (2018.0) 379 1303-1312. PMID: 30281992 4. Baidal JAW, Taveras EM. Childhood Obesity: Shifting the Focus to Early Prevention. Arch Pediatr Adolesc Med (2012.0) 166 1179-1181. PMID: 23108905 5. 5.Dixon B, Peña MM, Taveras EM. Lifecourse approach to racial/ethnic disparities in childhood obesity. In: Advances in Nutrition. Oxford Academic; 2012. p. 73–82. 6. Isong IA, Rao SR, Bind M-A, Avendaño M, Kawachi I, Richmond TK. Racial and Ethnic Disparities in Early Childhood Obesity. Pediatrics (2018.0) 141 e20170865. PMID: 29269386 7. Maziak W, Ward KD, Stockton MB. Childhood obesity: are we missing the big picture?. Obes Rev (2008.0) 9 35-42. PMID: 18154601 8. Savage JS, Fisher JO, Birch LL. Parental Influence on Eating Behavior: Conception to Adolescence. J Law Med Ethics (2007.0) 35 22. PMID: 17341215 9. 9.Scott-Sheldon LA, Hedges LV, Cyr C, Young-Hyman D, Khan LK, Magnus M, et al. Childhood obesity evidence base project: a systematic review and meta-analysis of a new taxonomy of intervention components to improve weight status in children 2–5 years of age, 2005–2019. 2020;16(S2). 10. Jackson JK, Jones J, Nguyen H, Davies I, Lum M, Grady A. Obesity Prevention within the Early Childhood Education and Care Setting: A Systematic Review of Dietary Behavior and Physical Activity Policies and Guidelines in High Income Countries. Int J Environ Res Public Health (2021.0) 18 1-21 11. Economos CD, Haire-Joshu D. Preventing Obesity in 2–5-Year Olds: A Pathway to Advancing Intervention Research. Child Obes (2020.0) 16 S259-S261. PMID: 32936040 12. Scott-Sheldon LAJ, Hedges LV, Cyr C, Young-Hyman D, Khan LK, Magnus M. Childhood obesity evidence base project: a systematic review and meta-analysis of a new taxonomy of intervention components to improve weight status in children 2–5 years of age, 2005–2019. Child Obes (2020.0) 16 S2-21 13. Nanney MS, LaRowe TL, Davey C, Frost N, Arcan C, O’Meara J. Obesity prevention in early child care settings: a bistate (Minnesota and Wisconsin) Assessment of best practices, implementation difficulty, and barriers. Heal Educ Behav (2017.0) 44 23 14. Sonneville K, La Pelle N, Taveras E, Gillman M, Prosser L. Economic and other barriers to adopting recommendations to prevent childhood obesity: results of a focus group study with parents. BMC Pediatr (2009.0) 21 9 15. Pagnini D, King L, Booth S, Booth M. The weight of opinion on childhood obesity: recognizing complexity and supporting collaborative action. Int J Pediatr Obes (2009.0) 4 233-241. PMID: 19922037 16. Bryson JM, Crosby BC, Stone MM. The Design and Implementation of Cross-Sector Collaborations: Propositions from the Literature. Public Adm Rev (2006.0) 66 44-55 17. Butterfoss F, Kegler M, DiClemente RJ, Crosby RA, Kegler MC. Toward a comprehensive understanding of community coalitions: moving from practice to theory. Emerging theories in health promotion practice and research: strategies for improving public health (pp. 194-227) (2002.0) 157-93 18. Mattessich PW, Rausch EJ. Cross-sector collaboration to improve community health: a view of the current landscape. Health Aff (2014.0) 33 1968-1974 19. Hennessy E, Economos CD, Hammond RA, Sweeney LB, Brukilacchio L, Chomitz VR. Integrating complex systems methods to advance obesity prevention intervention research. Heal Educ Behav (2020.0) 47 213-223 20. Korn AR, Hennessy E, Tovar A, Finn C, Hammond RA, Economos CD. Engaging coalitions in community-based childhood obesity prevention interventions: a mixed methods assessment. Child Obes (2018.0) 14 537-552. PMID: 30188181 21. Appel JM, Fullerton K, Hennessy E, Korn AR, Tovar A, Allender S. Design and methods of Shape Up Under 5: Integration of systems science and community-engaged research to prevent early childhood obesity. PLoS ONE (2019.0) 14 e0220169. PMID: 31369570 22. Kasman M, Hammond RA, Heuberger B, Mack-Crane A, Purcell R, Economos C. Activating a Community: An Agent‐Based Model of Romp & Chomp, a Whole‐of‐Community Childhood Obesity Intervention. Obesity (2019.0) 27 oby.22553 23. Hennessy E, Calancie L, Christina ED. Using models to understand community interventions for improving public health and food systems. Food Syst Model Tools Assess Sustain Food Agric (2022.0) 1 283-307 24. 24.Minkler M, Wallerstein N, editors. Community-based participatory research for health: from process to outcomes. 2nd ed. Jossey-Bass: San Francisco; 2011. 25. Butterfoss F, Kegler M, DiClemente RJ, Crosby RA, Kegler MC. The community coalition action theory. Emerging theories in health promotion practice and research: strategies for improving public health (2009.0) 157-193 26. Rogers EM. Diffusion of innovations (2010.0) 27. Luke DA, Stamatakis KA. Systems science methods in public health: dynamics, networks, and agents. Annu Rev Public Health (2012.0) 33 357-376. PMID: 22224885 28. Economos CD, Hammond RA. Designing effective and sustainable multifaceted interventions for obesity prevention and healthy communities. Obesity (2017.0) 25 1155-1156. PMID: 28653500 29. Ewart-Pierce E, Mejía Ruiz MJ, Gittelsohn J. "Whole-of-Community" Obesity Prevention: A Review of Challenges and Opportunities in Multilevel. Multicomponent Interventions Curr Obes Rep (2016.0) 5 361-374. PMID: 27379620 30. Kasman M, Hammond RA, Heuberger B, Mack-Crane A, Purcell R, Economos C. Activating a Community: An Agent-Based Model of Romp & Chomp, a Whole-of-Community Childhood Obesity Intervention. Obesity (2019.0) 27 oby.22553 31. 31.Massachusetts Department of Public Health, Bureau of Family Health and Nutrition, Nutrition Division. 2010 Pediatric Data Report: Pediatric Nutrition Surveillance System (PedNSS). Centers for Disease Control and Prevention (CDC) / Massachusetts Women, Infants And Children (WIC) Nutrition Program; 2012. 32. Hovmand P. Community based system dynamics (2014.0) 33. Hovmand PS, Andersen DF, Rouwette E, Richardson GP, Rux K, Calhoun A. Group Model-Building ‘Scripts’ as a Collaborative Planning Tool. Syst Res Behav Sci (2012.0) 29 179-193 34. Vennix JAM. Group model building: facilitating team learning using system dynamics (1996.0) 35. Calancie L, Fullerton K, Appel JM, Korn AR, Hennessy E, Hovmand P. Implementing group model building with the shape up under 5 community committee working to prevent early childhood obesity in somerville. Massachusetts J Public Heal Manag Pract (2022.0) 28 E43-55 36. Korn A, Hammond R, Hennessy E, Must A, Pachucki M, Economos C. Evolution of a coalition network during a whole-of-community intervention to prevent early childhood obesity. Child Obes (2021.0) 17 379-390. PMID: 33761266 37. Korn AR, Hennessy E, Hammond RA, Allender S, Gillman MW, Kasman M. Development and testing of a novel survey to assess stakeholder-driven community diffusion of childhood obesity prevention efforts. BMC Public Health (2018.0) 18 1-1 38. Korn AR, Appel J, Hammond RA, Hennessy E, Mâsse LC, Must A. Validation and refinement of the stakeholder-driven community diffusion survey for childhood obesity prevention. Implement Sci (2021.0) 16 91. PMID: 34627319 39. Adams J. Gathering social network data (2019.0) 40. Benjamin SE, Ammerman A, Sommers J, Dodds J, Neelon B, Ward DS. Nutrition and Physical Activity Self-assessment for Child Care (NAP SACC): Results from a Pilot Intervention. J Nutr Educ Behav (2007.0) 39 142-149. PMID: 17493564 41. Henderson KE, Grode GM, Middleton AE, Kenney EL, Falbe J, Schwartz MB. Validity of a measure to assess the child-care nutrition and physical activity environment. J Am Diet Assoc (2011.0) 111 1306-1313. PMID: 21872693 42. 42.Healthy Kids, Healthy Future. Available from: https://healthykidshealthyfuture.org/. [Cited 23 Oct 2021]. 43. Rogers VW, Hart PH, Motyka E, Rines EN, Vine J, Deatrick DA. Impact of Let’s Go! 5-2-1-0: A Community-Based, Multisetting Childhood Obesity Prevention Program. J Pediatr Psychol (2013.0) 38 1010-1020. PMID: 23933841 44. 44.Department of Health and Human Services. City of Somerville - Shape Up Somerville. Available from: https://www.somervillema.gov/departments/health-and-human-services/shape-somerville. [Cited 4 Sept 2019]. 45. 45.City of Somerville Health and Human Services Department. The well-being of Somerville report. Cambridge Health Alliance; 2017. 46. Csardi G, Nepusz T. The igraph software package for complex network research. InterJournal, complex Syst (2006.0) 1695 1-9 47. 47.Kasman M, Hammond RA, Mack-Crane A, Purcell R, Korn AR, Appel JM, et al. Using agent-based modeling to extrapolate community-wide impact from a stakeholder-driven childhood obesity prevention intervention: Shape Up Under 5. Child Obes. 19(2):130–8. 48. Flego A, Keating C, Moodie M. Cost-effectiveness of whole-of-community obesity prevention programs: an overview of the evidence. Expert Rev Pharmacoecon Outcomes Res (2014.0) 14 719-727. PMID: 25209305 49. Boelsen-Robinson T, Peeters A, Beauchamp A, Chung A, Gearon E, Backholer K. A systematic review of the effectiveness of whole-of-community interventions by socioeconomic position. Obes Rev (2015.0) 16 806-816. PMID: 26214391 50. Evans W, Christoffel K, Necheles J, Becker A. Social marketing as a childhood obesity prevention strategy. Obesity (Silver Spring) (2010.0) 18 S23. PMID: 20107457 51. Gracia-Marco L, Vicente-Rodríguez G, Borys J, Le Bodo Y, Pettigrew S, Moreno L. Contribution of social marketing strategies to community-based obesity prevention programmes in children. Int J Obes (2011.0) 35 472-479 52. Jones JA, Banks L, Plotkin I, Chanthavongsa S, Walker N. Profile of the Public Health Workforce: Registered TRAIN Learners in the United States. Am J Public Health (2015.0) 105 e30. PMID: 25689192 53. Hooker RS, Williams JH, Papneja J, Sen N, Hogan P. Dietetics Supply and Demand: 2010–2020. J Acad Nutr Diet (2012.0) 112 S75-91. PMID: 22709865 54. Coronado F, Beck AJ, Shah G, Young JL, Sellers K, Leider JP. Understanding the Dynamics of Diversity in the Public Health Workforce. J Public Health Manag Pract (2020.0) 26 389. PMID: 31688743 55. Levy DE, Pachucki MC, O’Malley AJ, Porneala B, Yaqubi A, Thorndike AN. Social connections and the healthfulness of food choices in an employee population. Nat Hum Behav 2021 510 (2021.0) 5 1349-57 56. Moore TR, Pachucki MC, Calancie L, Korn AR, Hennessy E, Economos CD. Coalition-Committees as Network Interventions: Baseline Network Composition in Context of Childhood Obesity Prevention Interventions. Syst 2021 (2021.0) 9 66 57. Calancie L, Nappi D, Appel J, Hennessy E, Korn AR, Mitchell J. Implementing and evaluating a stakeholder-driven community diffusion-informed early childhood intervention to prevent obesity, Cuyahoga County, Ohio, 2018–2020. Prev Chronic Dis (2022.0) 19 E03. PMID: 35050848
--- title: 'How European Fans in Training (EuroFIT), a lifestyle change program for men delivered in football clubs, achieved its effect: a mixed methods process evaluation embedded in a randomised controlled trial' authors: - Christopher Bunn - Victoria Palmer - Nai Rui Chng - Eivind Andersen - Cindy M. Gray - Kate Hunt - Judith G. M. Jelsma - Heather Morgan - Maria Nijhuis-van der Sanden - Hugo V. Pereira - Matthew Philpott - Glyn C. Roberts - John Rooksby - Øystein B. Røynesdal - Marlene N. Silva - Marit Sørensen - Pedro J. Teixeira - Theo van Achterberg - Irene van de Glind - Willem van Mechelen - Femke van Nassau - Hidde P. van der Ploeg - Sally Wyke journal: BMC Public Health year: 2023 pmcid: PMC10026416 doi: 10.1186/s12889-023-15419-y license: CC BY 4.0 --- # How European Fans in Training (EuroFIT), a lifestyle change program for men delivered in football clubs, achieved its effect: a mixed methods process evaluation embedded in a randomised controlled trial ## Abstract ### Background A randomised trial of European Fans in Training (EuroFIT), a 12-week healthy lifestyle program delivered in 15 professional football clubs in the Netherlands, Norway, Portugal, and the United Kingdom, successfully increased physical activity and improved diet but did not reduce sedentary time. To guide future implementation, this paper investigates how those effects were achieved. We ask: 1) how was EuroFIT implemented? 2) what were the processes through which outcomes were achieved? ### Methods We analysed qualitative data implementation notes, observations of 29 of 180 weekly EuroFIT deliveries, semi-structured interviews with 16 coaches and 15 club representatives, and 30 focus group discussions with participants (15 post-program and 15 after 12 months). We descriptively analysed quantitative data on recruitment, attendance at sessions and logs of use of the technologies and survey data on the views of participants at baseline, post program and after 12 months. We used a triangulation protocol to investigate agreement between data from difference sources, organised around meeting 15 objectives within the two research questions. ### Results We successfully recruited clubs, coaches and men to EuroFIT though the draw of the football club seemed stronger in the UK and Portugal. Advertising that emphasized getting fitter, club-based deliveries, and not ‘standing out’ worked and attendance and fidelity were good, so that coaches in all countries were able to deliver EuroFIT flexibly as intended. Coaches in all 15 clubs facilitated the use of behaviour change techniques and interaction between men, which together enhanced motivation. Participants found it harder to change sedentary time than physical activity and diet. Fitting changes into daily routines, planning for setbacks and recognising the personal benefit of behaviour change were important to maintain changes. Bespoke technologies were valued, but technological hitches frustrated participants. ### Conclusion EuroFIT was delivered as planned by trained club coaches working flexibly in all countries. It worked as expected to attract men and support initiation and maintenance of changes in physical activity and diet but the use of bespoke, unstable, technologies was frustrating. Future deliveries should eliminate the focus on sedentary time and should use only proven technologies to support self-monitoring and social interaction. ### Trial registration ISRCTN81935608, registered $\frac{16}{06}$/2015. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12889-023-15419-y. ## Introduction Low levels of physical activity and high levels spent in sedentary time are contributing to the increasing global prevalence of non-communicable diseases such as cardiovascular disease, type 2 diabetes and some cancers, [1] and innovative programs addressing both physical activity and sedentary time are urgently needed. Although global estimates suggest men report being more physically active than women [2], men engage less with traditional lifestyle change programs [3]. One approach to attracting men to lifestyle change is to offer programs in partnership with professional sports clubs, which is expected to attract men through the draw of the football club. [ 4–8] The first large-scale randomised controlled trial (RCT) to demonstrate the success of this approach was the Scottish Football Fans in Training (FFIT) weight management and healthy lifestyle program. [ 4] Delivered by trained coaches in professional football clubs, FFIT was designed to attract overweight men and enable them to lose weight through improvements in physical activity and diet [9]. FFIT was shown to be cost effective, with improvements in diet, physical activity and weight maintained to 12 months and partially maintained 42 months after baseline [10]. FFIT has been scaled-up through a single-license franchise model in over 70 UK professional football clubs, and scaled-out into football and other sporting contexts in Australia, Canada, New Zealand, England and other European countries [11]. One adaptation of FFIT was the European Fans in Training (EuroFIT) program [8]. EuroFIT extended the innovation from Scotland to other European countries but shifted the focus from body weight loss to improving physical activity and reducing sedentary behaviour (although men were encouraged to change their diet and lose weight if they wanted to [12]). A four-country pragmatic randomised controlled trial (RCT) demonstrated that EuroFIT was effective in increasing objectively measured physical activity, and improving diet, weight, wellbeing, self-esteem, vitality and biomarkers of cardiometabolic health, but not in reducing sedentary time 12 months after baseline [8]. Before recommending the widespread roll-out of EuroFIT, we wanted to investigate in more detail how it was implemented and whether it worked as we expected it to do. We thought this more detailed information could support future deliveries and add insight into how lifestyle change programs delivered in professional sports settings can work to both attract men and enable positive public health outcomes. Thus the aim of this paper is to examine the process through which EuroFIT achieved its effects. We do this by reporting a mixed-method process evaluation [13] designed to allow us to explore the multiple pathways and dynamics at work in how EuroFIT operated and achieved its outcomes [14, 15]. In line with our published protocol, we ask, 1) how was implementation achieved in football clubs and countries, and what was delivered? and 2) what were the processes through which the EuroFIT program achieved outcomes? [ 16] In answering these questions we pay particular attention to the mechanisms through which outcomes were achieved, thus allowing us to interrogate and evaluate the theory of change (presented as a logic model [16]) for the EuroFIT program. We also consider any differences in country context that might have been at work. Before describing our methods, we outline the EuroFIT program and the theories underpinning it (full description of these have been published elsewhere [12, 16]). ## The EuroFIT program: an overview The EuroFIT program was designed to support men to: become more physically active and less sedentary; improve their diets; and maintain these changes over the long term. It was delivered over 12, weekly, 90-min sessions that combined ‘classroom’ discussion, including ‘tools’ for behaviour change, with group-based physical activity. All sessions were led by football club community coaches. Full details of EuroFIT are published including a description of the program in the TIDieR template [16] and a summary is published in our trial report [8]. Here we describe in more detail the program theory of change (the rationale underpinning program assumptions), detailing the resources needed to deliver EuroFIT and the processes through which it was expected to work, in attracting men, and in initiating and maintaining change (Fig. 1). We also describe in more detail the theories underpinning program development and the bespoke technologies developed to support the program. Fig. 1EuroFIT Theory of Change Relational, physical and financial resources are needed to implement EuroFIT, including professional football clubs with sufficient commitment to try delivering the program and to recruit coaches willing to take part in training. The program also requires access to club facilities and materials for the program, and funding to pay for materials such as manuals and t-shirts (Fig. 1). Funding for the deliveries of EuroFIT reported in this paper was provided through the research grant. EuroFIT drew on theories derived from various schools of thought in psychology and sociology. From psychology, EuroFIT drew explicitly on contemporary theories of motivation specifically, self-determination theory [17] and achievement goal theory [18]. The goal was to facilitate participants to develop internalised, autonomous motivation that is both self-relevant and self-referenced for becoming more active, sitting less and eating a healthier diet. These theories emphasise the need to ground change in autonomous forms of motivation and to do so in a mastery-oriented, warm and supportive environment as opposed to externally imposed forms of motivation which tell men what they ‘ought’ to do. The theories also emphasise setting task-oriented, self-referenced and personally meaningful goals, rather than externally or ego-oriented goals to encourage ‘healthier living’. The change processes encouraged by these theories were operationalised in relation to initiating and maintaining change (Fig. 1) and taught the use of a set of behaviour change techniques (BCTs) [19, 20], presented to participants as a ‘toolbox’ from which they could choose the ‘tools’ they felt to be appropriate for them. These included self-monitoring, goal setting, problem solving, action planning and social support to both initiate and support change in the long term. Coaches facilitated interaction between men in sessions to help vicarious learning and to enhance motivation through enjoyment. From sociological theory, EuroFIT – like FFIT—drew on accounts of gender, culture and performance. EuroFIT is responsive to masculine identities and practices [21], seeking to work with and not against them. For example, we expected that EuroFIT would attract men because it was male-only and based in professional football clubs, a context which is recognised as symbolically valuable to many men (Fig. 1). By avoiding prescriptive content and offering science-led options and tools for implementing them; and by delivering the program in an interactive format that encouraged mutual learning, sociability and enjoyment [22], we expected men to feel valued and challenged in a way that was congruent with their masculine identity to support them to both initiate and maintain change. We also expected EuroFIT to provide an energising environment in which participants were free to construct and rehearse new performances of the self, supported by their peers [23]. We developed bespoke technologies to support the delivery of EuroFIT. Participants were provided with a novel pocket-worn device (the ‘SitFIT’), about the size of a matchbox, to self-monitor their daily step count (for physical activity) and time spent upright (for sedentary behaviour) [24, 25]. The SitFIT can display data from the last seven days and can be connected to computers (PC or Mac) to provide a more detailed historical record of these data through the ‘MatchFIT’ app. Alongside self-monitoring, MatchFIT’s primary purpose was to encourage between-session social support and group participation in physical activity via a team-based collective step challenge [12, 16]. Each EuroFIT group use the app to represent their football club in a weekly ‘game’ against a computer-simulated team. The groups competed to achieve a team-based average step count that exceeded the opposing team’s score, which was based on an algorithm designed to challenge the team to exceed the average they achieved in the previous week’s game. Groups were also encouraged to keep in touch with commonly used social media platforms such as WhatsApp or Facebook. All materials, including SitFIT and MatchFIT were offered in the appropriate language for each country. For the duration of the research, the country-based research teams provided implementation support for football clubs. This included funding to deliver the program, guidance on recruiting participants, provision of detailed manuals for coaches, a two-day training program for club coaches and support for trouble shooting problems during delivery. Full details of the implementation process and support offered to clubs are already published [26]. ## Process evaluation design The mixed methods process evaluation reported here was embedded in the EuroFIT RCT (Trial registration ISRCTN81935608, registered $\frac{16}{06}$/2015) [8]. Following UK Medical Research Council guidance, we investigated program implementation (how EuroFIT delivery was achieved and what was actually delivered), mechanisms of impact (the processes through which EuroFIT affected outcomes) and context (the broad cultural context of the country and specific cultural context of the football club in which EuroFIT was delivered). Our published protocol details eight research methods structured around delivering 18 research objectives organised around these three domains [16]. ## Changes to protocol In conducting analyses for this paper, we made some minor changes to methods described in the protocol. The protocol described eight methods, two of which (structured telephone questionnaire with participants opting out of the study and structured questionnaires to coaches on training and on program delivery) did not yield sufficient data to support mixed methods analysis and are not included here. The protocol did not describe two other methods which did prove useful in analysis. First, in recruiting football clubs and supporting clubs in their initial set up of the program, research teams in each country wrote implementation notes that were discussed at weekly team meetings. Both notes and discussion were useful in examining club recruitment. Second, to recruit participants to EuroFIT, clubs used social and other media adverts with a link to a web-based form for potential participants to express their interest in taking part. Data from this form were useful in examining participant recruitment. The protocol detailed 18 objectives, but we address only 15 in this paper. We have already reported participant demographic and health risk profile, [8] and characteristics of clubs taking part in EuroFIT [27]. Club level barriers and facilitating factors are described in another paper currently under review. ## Data collection Table 1 shows the eight methods used, organised around delivering the 15 objectives addressed in this paper, which in turn, were organised around the research questions. It also notes which aspect of the EuroFIT Theory of Change (resource, attract men, initiate change, maintain change) the mixed methods analysis can shed light on. Table 1Research objectives, Theory of Change element considered and data collection methods for the EuroFIT process evaluationStudy team implementation notesExpressions of interest and recruitment via a web-based formBaseline, post program and 12 M questionnaires EuroFIT participants($$n = 500$$)Participant attendance sheets for each session ($$n = 360$$)Coach questionnaires: post training and post program($$n = 30$$)Participants’ SitFIT and MatchFIT usage logsObservation of sessions ($$n = 30$$)Interviews with club representatives ($$n = 15$$) and coaches ($$n = 15$$) plus interviews with club representatives of interested clubs not able to take partPost program and 12 M focus group discussions with EuroFIT participants ($$n = 30$$)ConvergenceAgreement (A), Partial Agreement (PA), Silence (S), Dissonance (D), Not Applicable (N/A)Q1. How was implementation achieved in football clubs and countries and what was delivered? 1. Sources and procedures for recruitment of clubs and reported decision making in clubs in relation to participating in EuroFITXXA Element of theory of change considered—resources 2. Sources and procedures for recruitment of coaches to deliver the EuroFIT program in participating clubsXNA Element of theory of change considered—resources 3. Experiences of coach training and its usefulness in program deliveryXNA Element of theory of change considered—resources 4. Sources and procedures for recruitment of participants – how men were attractedXNA Element of theory of change considered—attract men 5. Participation in the EuroFIT program including number of sessions attended and the reported extent to which SitFIT and MatchFIT were usedXXXPA Element of theory of change considered—initiate change 6. The number of sessions and key elements of the EuroFIT program that were delivered by coachesXNA Element of theory of change considered—initiate and maintain change 7. The extent to which coaches delivered the EuroFIT program according to the coach manual and trainingXXA Element of theory of change considered—initiate and maintain changeQ2. What were the processes through which the EuroFIT program affected outcomes? 8. Participants’ reported reasons for joining, or continuing with the EuroFIT programXXA, D in relation to 'men like me' and 'club importance' Element of theory of change considered—attract men 9. Interaction between men and between men and coaches during the programXNA Element of theory of change considered—attract men and initiate change 10. How coaches used the coach manual and associated materials to deliver the EuroFIT programXXPA Element of theory of change considered—initiate and maintain change 11. Coaches’ views and experiences of the EuroFIT program and materials and in particular, which elements of the program were viewed as helpful and unhelpful in supporting participants to make lifestyle changesXNA Element of theory of change considered—initiate and maintain change 12. Participants’ views and experiences of the EuroFIT program and materials, which elements of the program were viewed as helpful and unhelpful in supporting them to make changes and how the environment coaches created influenced participant responses. Pay particular attention to the use of the toolbox of behaviour change techniques including SitFIT, MatchFIT, goal setting and self-monitoringXXPA, D in relation to objectively measured sedentary time Element of theory of change considered—maintain change 13. Participants’ experiences of maintaining (or not) any lifestyle changes made as a result of the program 12 months after baselineXNA Element of theory of change considered—initiate and maintain change 14. Participants’ views of which aspects of the program that were helpful and which less so for supporting long-term changeXXA Element of theory of change considered—initiate and maintain change 15. The characteristics of coaches that delivered EuroFIT in relation to background, demographic characteristics, skills, and experiencesXXNA Element of theory of change considered—resources ## Research team implementation notes Throughout the recruitment and delivery phases, the EuroFIT research team kept detailed notes on implementation. These notes were recorded in the minutes of regular meetings and provided data on issues that were faced. ## Expressions of interest and recruitment Expressions of interest were captured via a web form, which provided data on the numbers of men interested in participating in EuroFIT. Potential participants were screened by telephone for eligibility from this database before recruitment began, as described elsewhere. [ 8, 12] ## Participant questionnaires All 1113 RCT participants (560 allocated to intervention and 554 to comparison group) were asked to complete self-report questionnaires at baseline, when the intervention group had finished the program, (post-program) and 12 months after it began (12-months). The intervention group participants were also asked to complete additional self-report questions about their experiences of the program both post-program and at 12-months. The pre-program questionnaire asked about motivations for joining EuroFIT. Post-program questionnaires asked which components and tools of the EuroFIT program men received, used or did not use, and which they found useful. The 12-month questionnaire was designed to assess participants’ views on the ongoing usefulness of the components and tools in EuroFIT. ## Participant attendance Club coaches were asked to keep a record of attendance at each of the 12 EuroFIT sessions, via an online database provided by the research team. ## Participants’ SitFIT and MatchFIT usage logs Participants registered on the MatchFIT app and then SitFIT and MatchFIT usage data were collected remotely from users on a dedicated server. Data generated included logs of data uploads, error reports, user clicks, logins and logouts. Each item of logged data included a timestamp, information about the web browser and device type being used, and (except for pre-registration and pre-login activities in MatchFIT) a unique SitFIT identifier. ## Observation of sessions Observations of EuroFIT sessions were conducted in each of the 15 clubs by members of the local research teams, aiming for two sessions per club. We aimed to observe session 4, which covers multiple topics and activities, and one of session 3 or 5–12 in each club. We avoided sessions 1 and 2 to allow groups time to ‘form’ without feelings of being studies. Data were written down in note form using a standardised proforma, and written up in the style of thick descriptions [28, 29]. They focused on how participants interacted with one another and with coaches who led sessions, how they responded to different elements of the program and any other features of the interaction which may have influenced program effects. Notes of session 4 were also used to assess whether key components of the session were delivered as intended in a measure of fidelity. ## Semi-structured interviews with coaches and football club representatives We conducted semi-structured interviews with 16 coaches who delivered EuroFIT and with 15 football club representatives involved in managing EuroFIT when the program had been delivered. Interviews covered what they thought of the program, their experiences of what worked or did not work when launching EuroFIT, barriers to and facilitators of implementation, whether or not they saw a future for the program in the club, and what would be needed in the future to enable EuroFIT to continue at their club. Interviews were conducted by trained members of the research team, and were audio recoded and transcribed verbatim. ## Focus group discussions with participants Focus groups were conducted in each football club with a sample of participants who attended six or more EuroFIT sessions at two time points, when the program ended and 12 months after the program started. We tried to recruit the same men at each time point. We prompted discussion on what men thought of the EuroFIT program sessions, any perceived impacts on their lives, which elements of the program they found helpful or unhelpful and ways in which the program might be improved. Focus groups were conducted by trained moderators who were assisted by note takers, and were audio recorded and transcribed verbatim. ## Data analysis Our approach to data analysis is described in detail in our protocol paper [16]. Quantitative data were summarised using SPSS (v21) and reported descriptively. Qualitative data (study team notes, observations, interviews, focus groups) were analysed following a framework approach [16, 30], which included the development and testing of a thematic framework through on-going discussion in the research team. The thematic framework was applied separately by researchers in each country in local languages and supported by Nvivo 11, MAXQDA and AtlasTi (depending on site). Data from each of the four sites were then summarised by theme in English by local research teams, with example data extracts translated into English. These summaries were then compared systematically using framework approach matrices by three researchers (CB, NRC and VJP) based on the research questions and research objectives. Qualitative analysts from across the four research teams discussed data extensively and checked interpretations in multiple online and offline fora throughout the analysis process. Finally, to compare findings from different data sources, we were guided by the mixed methods ‘triangulation protocol’ [31], assessing agreement and dissonance across the datasets and also to identify areas of ‘silence’ i.e. where a given dataset has nothing to contribute, summarised in a ‘convergence matrix’ organised by objective. This final stage allowed us to examine the extent to which the data confirmed, were ambivalent toward or contradicted the causal assumptions in the Theory of change as well as any potential differences in delivery between countries. ## Informed consent and ethical approvals When participants agreed to join the study, they completed and signed an informed consent form, in the presence of a trained researcher. Participants received information sheets explaining the study, its aims and procedures, and were encouraged to read and ask any questions they had. They were asked to provide consent for each type of data collection they were involved in. All participants gave their written consent for the use of de-identified data collected during the study to be used in publications and outputs. Ethical approvals for the RCT and the process evaluation were obtained from appropriate country-specific ethics committees (Ethics Committee of the VU University Medical Center [2015.184]; Regional Committees for Medical and Health Research Ethics, Norway [$\frac{2015}{1862}$]; Ethics Council of the Faculty of Human Kinetics, University of Lisbon [CEFMH $\frac{36}{2015}$]; and Ethics Committee at the University of Glasgow College of Medicine, Veterinary and Life Sciences [200140174]). All methods were deployed in this study were performed in accordance with relevant guidelines and regulations. ## Results A summary of response rates by method can be found in Table 2. We present results in relation to the research objectives, organised around the two research questions: 1) how was implementation achieved in football clubs and countries, and what was delivered? 2) what were the processes through which the EuroFIT program affected outcomes? In each section, we also report on whether aspects of the theory of change were supported and any differences between countries which might be indicative of country-specific contexts to affect outcomes and the final column of Table 1 shows results of data triangulation in relation to agreement, dissonance and areas of silence across the datasets in relation to each objective. Table 2Response rates for survey and qualitative methodsCountry/MethodBaseline Survey(study pop)Post-Program Survey(study pop)12 Month Survey(study pop)Observations (total delivered)Coach Interviews(N sites)Club representatives (N sites)Post-Program FGD* n(N sites)12 Month FGD** n(N sites)Participant attendance logsNL275 [280]249 [280]240 [280]8 [48]5 [4]5 [4]28 [4]29 [4]96 [96]NO255 [260]200 [260]174 [260]6 [36]3 [3]3 [3]18 [3]20 [3]71 [72]PO232 [233]194 [233]201 [233]6 [36]3 [3]3 [3]24 [3]21 [3]72 [72]UK334 [340]245 [340]253 [340]9 [60]5 [5]4 [5]36 [5]37 [5]117 [120]Total1096 [1113]921 [1113]908 [1113]29 [180]16 [15]15 [15]106 [15]107 [15]356 [360]NL The Netherlands, NO Norway, PO Portugal, UK United KingdomFGD Focus Group Discussions Qualitative data* extracts are labelled to indicate participant ID (P (participant); C (coach); and (club representative)), club (NOR1-3, UK1-5, POR1-3, NL1-4) and data source (PPFGD [post-program focus group], 12MFGD [12-month focus group], OBS [observations], INT [interview]. Additional examples from the qualitative data can be found in Additional File 1, which presents qualitative extracts in relation to the theory of change in more detail. ## How was implementation achieved, what was delivered and by whom? Recruitment of clubs and coaches and experiences of coach training (objective 1, 2, 3 and 15 Table 1; element of theory of change—resources). Using research team members’ professional networks, we successfully recruited 15 clubs from four countries to deliver the EuroFIT program. Study implementation notes recorded that three clubs pulled out between initial approach and commencement of training, one in Norway and two in the Netherlands. Despite research funding paying for delivery, financial difficulties and changes in club priorities regarding community engagement as reasons for withdrawal. Using research team members’ networks, it proved possible to replace these clubs within the study timeline. Analysis of study implementation notes and interviews showed agreement that the clubs co-operated well with the research team throughout the study. All 15 participating clubs sent at least two coaches for two-days of training to deliver EuroFIT program, which was conducted in each country by members of the research team who had been involved in the development of the program. Most of the coaches were male ($\frac{21}{30}$, $70\%$) and aged between 19 and 56. Most were experienced football coaches, and some had additional health and fitness qualifications. In Norway and some clubs in the Netherlands some coaches were not club staff but were employed specifically to deliver EuroFIT on a sessional basis. Analysis of interview data showed that coaches were generally positive about the training they received and felt it prepared them to use the tools they needed to be able to deliver the EuroFIT program. For example, one said:“I liked the fact that we received a lot of relevant information. It gave me the confidence I needed. I wouldn’t have been the same with a manual only. I liked having the information up front, so I felt I knew what I was saying. It was good. Was positive. A good combination of theory and practice.” [ C-NL1-INT] However, analysis of study implementation notes showed a high turnover of coaching staff at some clubs in the UK with some of the coaches originally trained to deliver to program leaving their posts before delivery or part-way through the program. As a result, the research team had to develop bespoke interim training (which took one instead of two days) for new UK-based coaches. In relation to the theory of change, the relational resources described as necessary for program delivery, were available in all countries, as were the financial resources (because they were covered by the research grant). We recruited clubs through professional networks, although additional efforts were needed to replace clubs that dropped out of the study in the Netherlands and Norway. Attracting men to the program (objective 4, Table 1; element of theory of change – attract men). To recruit participants, football clubs e-mailed invitations to fans and used club websites, social media posts, features in the local press, and match-day leafleting/announcements to advertise the program [8]. Recruitment materials emphasised that the program took place in the football club, asking, “Do you want to become more active with [club name]? “ and explained that “EuroFIT aims to help you increase your physical activity levels and to be less sedentary. The program will offer you a toolbox of skills and techniques for living a healthy life and a chance to get fitter and feel better in yourself.” Materials also highlighted that the program would be interactive, led by club coaches, and that they would be with like-minded people, saying; “Each session includes physical activity led by [CLUB] coaches at [YOUR STADIUM/TRAINING GROUND] as well as focusing on how to incorporate new skills and techniques into your everyday life. The sessions will also offer you the opportunity to meet like-minded people and share tips and advice.” Data from expressions of interest forms showed that advertisements proved attractive to men wanting to join the program particularly in the UK and Portugal. The mean number of men per club expressing an interest in joining was 200.2 (155–272) in the UK, 121.0 (87–187) in the Netherlands, 155.3 (128–205) in Norway, and 441.0 (333–616) in Portugal. In every club, expressions of interest exceeded the number of places available ($$n = 80$$), with the Portuguese clubs attracting particularly high levels of interest. In relation to the theory of change, the approach to attracting men which emphasized gaining fitness, being based in the club, and reassuring men they would not ‘stand out’ – i.e. they would be with like-minded men—proved successful in all countries, but especially in Portugal and the UK. Participation in the EuroFIT program and the reported use of SitFIT and MatchFIT (objective 5, Table 1; element of theory of change – initiate change). Coach-reported mean attendance at the 12 EuroFIT sessions was comparable across the four countries: 8.7 sessions (SD 3.3) out of 12 sessions in the UK, 9.8 (SD 2.5) in the Netherlands, 8.1 (SD 3.0) in Norway and 8.7 (SD 3.2) in Portugal. Analysis of post program questionnaires showed, as reported previously [8], $65.0\%$ of participants said they used the SitFIT ‘a great deal’ (score 4 on a scale of 0–4) and only $36.8\%$ reported they used MatchFIT ‘a great deal’. SitFIT and MatchFIT usage logs suggest that $\frac{341}{560}$ ($60.8\%$) successfully uploaded SitFIT data to the server (suggesting agreement with self-reported data) and $\frac{359}{560}$ ($64.1\%$) registered for MatchFIT. In relation to the theory of change, we presumed that regular attendance at the program would be necessary to initiate change, and that use of SitFIT and MatchFIT would enhance engagement. The data suggest a good level of attendance and use of SitFIT as expected. Number and key elements of the sessions delivered and extent to which coaches delivered EuroFIT as intended (objectives 6 and 7, Table 1; element of theory of change – initiate and maintain change). As reported previously, [8] fidelity of delivery was good: we observed deliveries of the fourth of 12 sessions in $\frac{14}{15}$ clubs and in these, coaches delivered 221 of 252 ($88\%$) key tasks. Analysis of data from interviews with coaches agreed with those from observations. Coach training had encouraged flexibility in session delivery to suit the coaches’ own styles and the needs of the different groups, whilst retaining the core elements of the program and adhering to the principles of valuing men, supporting autonomy and positive interaction. Coaches from the Netherlands, UK and Norway, in particular described how they made adaptations to suit the needs of their group and/or the facilities available at their club (particularly for physical activity). In the Netherlands and UK, coaches reported changing the physical activity sessions to involve higher intensity activities (usually football instead of walking football), as they felt men were expecting more intense physical activity from the outset. However, some coaches also described omitting content. For example:“I did not do the thing, filling the glasses you know [part of a suggested activity to enhance visual representation of sugar sweetened beverages in a food-related session]. I looked at the group and knew they would not be interested in these glasses and discussing how many lumps of sugar would be in them, well you know. This pouring, it doesn’t fit with this group.” [ C-NL1-INT] In relation to the theory of change, we presumed that fidelity in delivery would be important to initiate and maintain change and thus to achieve outcomes. Based on the analysis we can say that the program was largely delivered as intended because most of the key elements were delivered in the observed sessions and coaches described situations in which they were adapted the content to deliver flexibly as intended. ## What were the processes through which the EuroFIT program affected outcomes? Participants’ reported reasons for joining, continuing with or opting out of the EuroFIT program (objective 8, Table 1; element of theory of change – attract men). We have reported above that advertisements proved attractive to men wanting to join EuroFIT, particularly in the UK and Portugal. Figure 2 shows that the most cited reasons in baseline questionnaires were: to get fitter ($91.3\%$), to lose weight ($87.3\%$) and to improve lifestyle ($74.5\%$). Data from focus group discussions were consistent with this. For example, one Dutch participant said:‘So yes, I had purely looked at that, I just had something like well, there was no guarantee that you would lose weight, but you are going to exercise more, eat differently so you will get rid of some kilos. Well, that was the reason for me to start.’ [ P-NL1-PPFGD]Fig. 2Self-reported reasons for joining EuroFIT The importance of the club was reported as a factor by more men in the UK ($40.6\%$) and Portugal ($46.8\%$), than in the Netherlands ($24.3\%$) and Norway ($26.2\%$). This did not resonate with the post-program focus groups, which all mentioned the importance of the club as a draw. For example, in the Netherlands, one man explained how getting privileged access to valued parts of their club attracted him:It is still your team where you’ve been going to for years and normally you don’t get to go onto that football field. And now you are allowed to enter the field and train some more and the support you get. That’s just really important.’ [ P-NL1-PPFGD] In the baseline questionnaires, participants in Norway ($46.9\%$) and the UK ($34.7\%$) were more likely to report that being with men ‘like them’ was a reason for joining than participants in the Netherlands ($12.9\%$) and Portugal ($15.0\%$). Again, there was some dissonance between questionnaire and focus group data, with examples of wanting to be with men ‘like them’ emerging in focus groups in the UK, Netherlands, and Portugal. For example, in a post program discussion in one UK club, participants agreed why they were attracted to EuroFIT:P6:Meet similar people as well. I:Yeah. P6:The same age group, same interests and, you know, perhaps eating too many beefburgers and having too many pints [of beer], um, being able to relate and move forward together as a group. [ P-UK5-PPFGD]. In the Netherlands, the importance of the program being men only was emphasised: ‘I was thinking about that actually …. That it was only men [–-] If there had been women I wouldn’t have joined.’ [ P-NL3-PPFGD]. In Portugal too, one man suggested: ‘I liked the idea of being with other men of my age and condition’ [P-PT1-PFGD]. In relation to the theory of change, the assumption of multiple motivations for joining the program proved correct, with men in all countries commonly citing 'weight loss', 'desire to improve lifestyle' and 'getting fitter' as reasons to join and there was agreement between data sets (see also Additional File 1). Assumptions in the theory of change about the appeal of the club were confirmed in the UK and Portugal with less evidence of its importance in the Netherlands and Norway. Similarly, assumptions about the appeal of being with others ‘like them’ were confirmed in the UK and Norway, with less evidence in the Netherlands and Portugal. Interaction between men, and between men and coaches, during the program (objective 9, Table 1; element of theory of change – attract men and initiate change). We expected that interactions between men, and between coaches and men, would be important for group cohesion and success. The theory of change (Fig. 1) suggested that interactions with the coaches would make men feel valued and would provide support in ways consistent with masculine identities (for example, by using ‘football talk’). Session observations suggested that coaches in all four countries did encourage men to feel that their participation in EuroFIT was valued. This was made most visible through the informal ‘small talk’ (often about football) that occurred between men and coaches before the sessions began, which demonstrated a commitment to building relationships. For example, an observation in the Netherlands recorded that:The coach then asks: “Have you guys been to the match?” “ Yes,” said one of the men, “it was spectacular.” “ I know right?” said the coach, “I saw you there!” “ Yes, 4–3 hey?!”’ ( OBS-NL1) The way coaches valued participants was also made evident through ‘grand gestures.’ For example, in one club in the UK, a coach gave a teddy bear to a man who had just had a child. Participants in all four countries also commented on these exchanges and the welcoming nature of coaches (see Additional File 1). The theory of change also suggested that encouraging and enabling interactions between men themselves would mean that sharing experiences would support change and that participants would enjoy the sessions and thus keep coming. In all four countries observations demonstrated that coaches encouraged interactions and men enjoyed being at the sessions. In the classroom-based sessions men shared stories of their successes, which were met with positive feedback from other men. They also reflected on more challenging experiences they had faced, when other men offered support in return: A man who has had a visitor from [another] office tells the group of the struggle he had:“I can’t get into my rhythm, you know, what I want to do. I have to take him to dinner and things. I’ve been good, but it’s still tougher than it should be. ”Other men sympathise and a discussion about how he might cope with such responsibilities in the future is started. Suggestions from other men include: planning the restaurants that he will take his visitor to in advance and pre-choosing his meals to avoid making impulsive decisions; tell visitors that you are eating healthily; eat carefully when you do have freedom to choose e.g., breakfast, snacks and packed lunch; and to make sure that you have at least one nice dinner. [ OBS-UK3] Similar supportive interactions were observed in the physical activity parts of the sessions. Although participants were often keen to do ‘better’ than other clubs in the program, there was no evidence of competition between men in the same group where interactions were supportive. Men’s enjoyment of the program was evidenced by observations that men’s interactions were often interspersed with jokes and banter, which contributed to the supportive environment, the fostering of relatedness and enjoyment of sessions. For example, observation of a physical activity session in the Netherlands recorded:‘Now the men have to walk on their toes and make themselves as long as possible with their arms in the air. The string of men walk along the lines on their toes with their hands in the air. “ Oh beautiful, Swan Lake!”, says the man with the bruised ribs from the sidelines.’ ( OBS-NL3) In summary, in relation to the theory of change, the observations of sessions confirmed the assumptions we made about the importance of positive interactions in delivering outcomes (see also Additional File 1). Coaches’ interactions with men made them feel valued, and men’s supportive, humorous interactions with each other made the program both enjoyable and potentially motivational. Coaches’ use of the program manual and experience of delivering the program (objective 10 and 11, Table 1; element of theory of change – initiate and maintain change). Our theory of change made the coaches’ role explicit only in relation to how they interacted with men and how they encouraged interactions. Implicit to the theory of change, however, is that coaches, through following the program manual and being well prepared for sessions, would be crucial in providing an excellent motivational environment to support change [18] which included teaching participants the skills embedded in the ‘toolbox’ of BCTs. In interviews, coaches in all countries reported that they found the manual essential for delivering the program. For example, a coach in the Netherlands said: “The coach manual was very handy. It was really nice to have these clear instructions” [C-NL4-INT]. There was agreement between the observations and interviews that coaches did teach participants how to use of the ‘toolbox’ of BCTs. For example, a coach in Norway said:Along with the other tools, then the SMART-goals have also been very good indeed, and those you link with the SitFIT and MatchFIT, and that is like, that you get to link those tools together has been a very good method [C-NOR2-INT] Many of the coaches played an important role in the repetition and practice of BCTs, and prompted men to set optimal challenges for themselves, building on their own progress, as one coach from the UK explained:Sometimes, the guys trying to make big targets, we'd try and calm them down. Which, again, the [training emphasised] you need to do that, erm, which, which definitely is good advice. Erm, so yeah, I think goal setting is very important. The guys did stick to it. [ C-UK1-INT] However, other coach interviews showed that it could be hard to explain goal setting to participants, as one Dutch coach explained:“I think they also found it difficult, like how to choose something [a SMART goal] that is specific enough and that I can potentially accomplish.” [ C-NL2-INT] The repetition of goal setting each week could be experienced as boring, as a coach from another Dutch club explained, suggesting that goal setting may be important in initiating but not maintaining changes:“.. so at a certain moment, some of the men – not all of them—would go like ‘Yeah, I know by now!’ and they no longer felt like evaluating progress over and over again.” [ C-NL4-INT] Coaches clearly understood that their role was to facilitate change having created an enabling motivational environment and taught the flexible use of BCTs. This is illustrated by an extract from a coach in Norway, who said:I like very much that they [participants] set their own goals and that we get to choose the tools we need ourselves. We try, we give them many options to take those tools that fit into their everyday life and that we do not say that so and so and so, and if you do not do this, you fail. That you, that we make them aware, and that making your own choices is done every day, that we take those choices unconsciously in a way. [ C-NOR2-INT] Consistent with our theory of change, analysis of observations of sessions and interviews with coaches confirmed that coaches used the manual as expected, creating an enabling environment for behaviour change by teaching BCTs. Participants’ views and experiences of the EuroFIT program and materials (objective 12, Table 1; element of theory of change – initiate and maintain change). As we have seen, our theory of change placed explicit emphasis on learning to use a ‘toolbox’ of BCTs to promote an agency- and competency-based approach to initiating and maintaining changes in physical activity, sedentary and dietary behaviours. Specifically, the ‘toolbox’ focused on: goal setting, problem solving, action planning, self-monitoring of behaviour and outcomes, and social support. Data from focus groups showed that, in all four countries, men engaged with the toolbox of BCTs. For example:I think the great thing about the program was that it wasn't, you [didn’t have to] to make a massive amount of effort to make an improvement. It was small improvements, and you could measure them. And so, what I think was a great thing was that you learned about walking a little bit more, standing up a little bit more, just doing incremental things made you fitter. [ P-UK1-PPFGD] Responses to post-program questionnaires supported these findings, with participants reporting high levels of engagement with SitFIT (self-monitoring) devices, SMART goal setting, planning for making cumulative small changes to their everyday lives and seeking social support (see Fig. 3).Fig. 3Self-reported use of BCTs at post-program Objective measurements of physical activity indicated that after completing EuroFIT, on average, men were walking an additional 1208 steps/day ($95\%$ CI: 869 to 1546, $p \leq 0$·001) over and above the baseline average of 8,372 steps/day [8]. This finding is consistent with post-program focus group data, in which EuroFIT participants offered reoccurring narratives of being more active. For example, one said:Now I developed an excel [file] to track my routines and progress regarding my almost daily walks and runs and my weight. I really like to fulfil it and think of ways to upgrade it… I put time, intensity [into it]… look (Participant shows interviewers some printed files)’ [P -POR2-PPFGD] Self-monitoring was one of the main BCTs that men reported using. Men reported keeping the SitFIT device in their pockets and constantly checking their progress throughout the day or after certain activities (e.g., climbing the stairs). Across the four countries, participants commonly reported routinely using their SitFIT for monitoring their physical activity, as this interaction from a post program focus group in the UK demonstrates:P2:I think the SitFIT is quite a big thing actually for me, you know? Err, I go out for a walk and [my wife] says have you got your SitFIT, have you got your SitFIT?P6:Yeah, I get that. P2:You have? Yeah, yeah, I’ve got it, I’ve got…I know I’ve got it. P5:Or you’re gutted if you go out and you’ve forgotten it. P3:You do, it’s the worst feeling. P5:You can’t get those steps back. P7:Worst feeling, yeah. [ UK3-PPFGD]. As described, we developed the SitFIT (and some aspects of the MatchFIT app) to support self-monitoring. Whilst most felt positive about the SitFIT and in some cases found sharing their steps either through MatchFIT, Facebook or WhatsApp motivating, one man in the UK described how he felt that sharing his steps on MatchFIT might let the other men in his group down if he had not achieved what he had hoped (men in this group were taking a lot of steps each day):I think the thing as well that I noticed was I’d look at it and if it’d have, like our average would be like eighteen thousand steps, I’d look at it and go, I’m only on sixteen thousand, if I put mine, it’ll bring everybody down […] I felt bad if I plugged it in and then…then the [MatchFIT average] figure was less after I plugged in. [ P-UK3-PPFGD] While men spoke positively about the SitFIT, they also described challenges using the device. These were practical (e.g., difficulty attaching device to clothing), technical (e.g., difficulty synchronising data) or mechanical (e.g., steps not registering correctly). While self-monitoring appeared to be a useful BCT for increasing physical activity (steps), men reported that they found decreasing their sedentary time more difficult. Post-program focus group data program suggested that, while participants internalised the message to stand more, they found it harder to use self-monitoring and goal setting to change their sitting behaviour:P5:I’ll look at it [upright time on SitFIT] and I’ll see what it is but I don’t kind of work towards it, do you know what I mean?P2:No. P4:It’s not…you can’t really quantify what you’re doing so…[UK3-PPFGD]. As part of the toolbox, EuroFIT aimed to encourage social support and relatedness through peer interaction. Participants responded well to encouragement to use social media for social support outside of EuroFIT sessions, using both WhatsApp and Facebook to support one another to be active, both during the program and up to 12 months afterwards (see Addditional File 1). Another way that men were encouraged to engage with EuroFIT between sessions was through the MatchFIT app. As measured by server log data, the overall uptake (participants creating an account and using the app at least once) of MatchFIT across all countries was $63.5\%$. Uptake varied significantly across countries: in the UK it was $37.2\%$, in the Netherlands $65.2\%$, in Norway $78.7\%$, and in Portugal $82.8\%$. MatchFIT was introduced in week 4 of the program. About half of those who registered had stopped using the app by week 10, and, by week 12, only a third were still using it. Qualitative data from participants suggested that, while the idea of MatchFIT was appealing in all countries, particularly in the Netherlands, the practical challenges of connecting the SitFIT and uploading data limited wider engagement (see Additional File 1). Some participants also commented that playing against an algorithm, rather than other EuroFIT teams, was demotivating. In relation to our theory of change, men’s experiences of the program, taken from post program questionnaires and focus groups, suggests that the program worked as intended for them in initiating changes to their lifestyles (particularly physical activity) and that this was consistent across countries. Self-monitoring through the SitFIT worked well, although this was less successful for sedentary behaviour than for walking and men found making changes to sedentary behaviour difficult. The motivation strategies were effective, self-referenced goal setting enabled small cumulative changes in physical activity, and relatedness through social support was important for embedding change and MatchFIT demonstrated potential to encourage interaction in between sessions but technical difficulties uploading data made its use frustrating. Participants’ experiences of maintaining (or not) any lifestyle changes and what was helpful in maintaining change (objectives 13 and 14, Table 1; element of theory of change – maintaining change). EuroFIT was designed to support men to make changes that they would be able to sustain beyond its 12-week delivery. As noted above, practising BCTs was helpful to participants in initiating and maintaining changes through the program, and men were encouraged to make small changes that could be accommodated within their daily routines. In the 12-month questionnaire, $71\%$ of the intervention group said they became more active by making small changes to everyday life ‘often’ or ‘a great deal’. Focus group data from all countries confirmed this, with examples such as the following offered by participants:I’ve got to enjoy it, my routine of walking around for the [news]paper in the morning and stuff like that, and I’ve kept it going and it’s just because it gave me the push that I needed, basically. [ P-UK4-12MFGD] The same pattern of convergence between qualitative and quantitative data was observed in relation to the routinisation of dietary changes, in all countries. Some described such routinisation in very direct terms e.g., “I don't make an effort… this new way of behaving is like the normal me.” [ P-POR3-12MFGD]. Our theory of change (Fig. 1) suggested that recognising the personal benefit of behaviour change would help men to maintain changes. Men in all countries reported feeling fitter and losing weight, and some had noticed energy improvements since participating in EuroFIT, often reflecting on how these changes had influenced their daily lives:At my home I think at least that previously when I confidently sat in the favourite chair with the remote control, then I think it is much nicer to see myself confident where I am active and that you do something with yourself to be more present at home both energy-wise and with the kids and, yeah – day-to-day stuff, really [P-NOR1-12MFGD] We have reported elsewhere that EuroFIT participants, on average, reduced their sedentary time post-program (i.e. at 12 week measures) by an estimated 14.4 min/day ($95\%$ CI: -25·1 to -3·8, $$p \leq 0$$·008), but that this reduction was not maintained at 12 months [8]. This finding from the RCT converged with men’s accounts in the 12-month focus groups, which described finding it difficult to integrate changes to sedentary behaviour into their daily lives. For example, one Portuguese participant said, “Although I do some PA, changing sedentary behaviour is hard… I cannot say honestly that I accomplished it.” [ P-POR1-12MFGD]. As part of EuroFIT, participants were taught how to maintain new behaviours in the face of adversity, through activities which focussed on avoiding and overcoming setbacks. Participants from all four countries described how they implemented these aspects of EuroFIT, as this example from Portugal illustrates:This weekend it was my son’s birthday, we had a big party and I had everything that I liked… Coca-Cola, beer, lots of food… I can’t even recall what exactly…Sunday I started the day by running 5 km – That’s the way to go” [P-POR3-12MFGD] However, some participants expressed unhappiness about extent to which they had been able to keep changes going. For example, one man said:“I feel guilty, I must be the worst participant ever. I think I lost about 12 pounds, but look at me now” [he had regained the weight] [P-NL1-12MFGD] Such participants reported facing challenges relating to illness and injury, work patterns and culture, falling back into old habits, and life events such as divorce or bereavement. Poorer winter weather was also discussed as a de-motivating factor to continue exercise in all countries. The theory of change also suggested that the program would nurture social support beyond the 12 weekly sessions. This was reported as happening in some clubs in Portugal, the Netherlands and the UK where groups were set up via WhatsApp or Facebook to continue to walk or play walking or five-a side football. Where this happened, it was seen as motivational was seen as important program, as in this example in the UK:P3:I mean the whole thing when we did the EuroFIT, [..] Our group was brilliant, wasn't it? how we got on well. And it kind of continued. It's as though you handpicked them, you know what I mean? The group. P2:I could say a few things, yes. Since we’ve stopped, actually, the program, I was playing regularly walking football. I really enjoyed that. There is such a nice group of people, I would like to call them friends, and we really were so, you know, we were getting together, playing football together, and as [name] said just recently, which was really important for me, was we were not as competitive as the others could be, so that was pure joy to play and really get physically active, and I can tell you we were really sweating a lot, you know? [ UK4-12MFGD]. In Portugal, the Netherlands and the UK men who continued to do physical activity together as a group described a real strong sense of camaraderie and motivation. Men in Norway, where distances were further, did not meet up and that many of them now were walking alone (because of geographical distance and no organised sessions) and this made it hard to maintain their activity levels. They missed the group, the people and the weekly sessions they had together, and said that these things would have made it easier to maintain the activity long term. As one said ‘The mile is long if you are by yourself’ [P-NOR3-12MFGD]. In summary, men’s experience of the program suggests it largely worked as described by the theory of change, at least for those who maintained change – in that they recognised the personal benefits brought by relatively small, but cumulative, routinised, lifestyle changes, although not everyone had been able to keep those up. There was also evidence that men had learnt to avoid setbacks and plan for difficult encounters, although again not all participants were able to do this because of illness or other life events. Groups formed to continue exercise after the program were highly valued where they happened and were seen as motivational but, in their absence, as in Norway, it was harder to motivate ongoing physical activity. ## Discussion Our analysis shows that EuroFIT was largely implemented as planned in professional football clubs in four countries, that coaches delivered the program as intended and that the program worked largely as expected for participants. The research project paid for deliveries of EuroFIT within the trial context which facilitated recruitment of professional football clubs to deliver the program, although the fact that we had to identify additional clubs when three dropped out, suggests that it might be more difficult to recruit clubs if funding for deliveries was not readily available. Coaches responded well to the training provided in EuroFIT session delivery, although high turnover of staff was a problem in some clubs. Recruitment strategies proved successful in all countries, but especially Portugal and the UK and it is possible that the ‘draw’ of the football club was not so clear amongst men in the Netherlands and Norway. The UK has a long history of community development activity in professional football clubs [32], so football fans could be used to being invited to programs. Club-based activities and programs targeting adults have only recently been instigated in Portugal, although the love of football may be even stronger so the opportunity to take part in club-based activity highly motivational. Attendance at program sessions, e assumed would be necessary to initiate change, was good. Fidelity, assumed necessary for successful outcomes, was also good; and coaches responded flexibly to participants’ perceived needs, a feature of the program we think it essential to retain in subsequent implementation [26]. Interactions between men, and between coaches and men, were also confirmed as important elements of success and as intended the environment was seen to be supportive and motivational. EuroFIT was one of the first studies to deliberately engage motivation theory behaviour change strategies into the protocol. From both self-determination and achievement goal theories, we incorporated BCTs associated with autonomy, relatedness, competence, mastery and self-referenced goal strategies and coaches were very successful in using those strategies to foster behaviour change. Self-monitoring using the SitFIT and the setting of self-referenced goals were seen as particularly important by participants and coaches and reinforcement of BCTs was also important to maintain changes through the program. MatchFIT, on the other hand, did not seem to play an active role in encouraging maintenance of behaviour change in most countries, perhaps because of the technical difficulties encountered in its use, and unlike in the FFIT program[23] few narratives of identity change were presented by participants. EuroFIT was not successful in supporting men to make long-term improvements to their sedentary behaviour. While there is good evidence for different approaches to, and the benefits of, becoming more active [1], less is known about how to intervene successfully to decrease sedentary time, which is often confused with physical inactivity [33]. A systematic review reported that interventions designed to increase physical activity are less successful in changing sedentary behaviour, but when interventions focus on sedentary behaviour specifically, reductions are possible [34]. Despite best efforts at the program design stage, it seems that EuroFIT may have fallen into the same trap: sedentary time did not change in the long term, and participant narratives, combined with objective outcomes, suggest that increasing physical activity (rather than reducing sedentary time) was a dominant focus for most. It is possible that the twin focus on physical activity and sedentary time limited the effort participants put into most unfamiliar of these, sedentary time. It could be that in multi-outcome focussed programs like EuroFIT different tools are needed to support changing sedentary behaviour. That said, the mechanisms through which changes in physical activity were likely produced – i.e. goal setting, self-monitoring, action planning, social support and problem solving – are consistent with both the theory of change and with literature relating to behaviour change for physical activity in men [35]. In this way, our study adds to this body of evidence and reinforces the utility of these BCTs for interventions targeting increases in physical activity. Specifically, it confirms positive findings from other interventions targeting physical activity set in professional sports contexts [4–6, 36–38]. A novel feature of EuroFIT was its use of bespoke technologies. The SitFIT activity tracker [25] was widely appreciated by participants, but in some instances they also reported challenges in using it (e.g., practical, technical, and mechanical). These findings highlight both the potential of bespoke devices and the risk using them entails. Intervention designers considering deploying bespoke devices therefore need to consider whether the benefits offered by such devices outweigh the risk they potentially pose for intervention implementation. There are many more commercial but affordable physical activity self-monitoring devices available than when EuroFIT was developed and using one of these is probably a better option. Our findings in relation to the MatchFIT app suggest that it did not provide the platform for social support that it was intended to at least partly because of technical difficulties in uploading data. This finding is consistent with the ‘ManUp’ study [39] and with HockeyFIT [40], both of which sought to promote social support for physical activity and lifestyle change through app-based social support, which reported low levels of social engagement and rapid drop off [39]. Our study suggests that it is possible to implement EuroFIT in different club settings. It also indicates ways in which clubs tailor EuroFIT to their own contexts and characteristics. Future research and practice relating to the delivery of health interventions in professional sports settings should ensure programs are designed with flexibility in mind, to ensure that variations in resource availability do not present barriers to implementation, although the key elements of how the program functions should be maintained. Given the significant public health benefits that have been demonstrated by such interventions [4, 10, 41], and the contributions they can make to population health [42, 43], care and attention needs to be paid to the implementation contexts with the potential to maximise these benefits [44]. In addition, as we have previously suggested [26], an important requisite for future roll-out of EuroFIT would be a strong delivery partner organisation to ensure financial and human resources, while ensuring continued quality of delivery in clubs. A final point of discussion raised by our study relates to harnessing the appeal of professional sports club settings and in being with other like them to attract participants to health interventions. As we have reported, our findings generally support the body of literature which argues that professional sports club settings are a powerful motivator for intervention participants, and perhaps particularly men [4, 5, 36, 37, 45, 46]. However, the baseline survey findings from Norway and the Netherlands suggest that it cannot be taken for granted at least for local clubs. Researchers intending to harness the appeal of professional sports organisations in new contexts should therefore consider the social and cultural forces which these organisations are subject to. In the field of football studies researchers have documented the multiple ways in which fandom manifests [47–49]. In Norway, for example, football fandom is increasingly transnational [50], and both British and Norwegian fans have polygamous relationships with different clubs [51]. Integrating interventions into localised fandoms is likely to ensure that the appeal of the professional sports-based setting is maximised and that interventions engage with these fanbases with appropriate sensitivity. ## Strengths and Limitations A major strength of this study is that it analysed data from four countries, in fifteen club settings, and derived the findings from a large body of data collected through multiple methods, allowing for triangulation across data sources. The study also deliberately employed behaviour change strategies informed by contemporary motivation theories. These strengths and the findings from the study underpin our confidence in the utility of the EuroFIT theory of change as a representation of the program’s mechanisms of action for men in the four countries studied. A limitation is that although we went to considerable efforts to standardise qualitative data collection and analysis across countries it was conducted by multiple teams, with varying levels of experience, in four different languages. The research team which produced the convergence matrix had limited ability to go back to original sources to check interpretations for themselves. This was mitigated through extensive cross-country dialogue, but remains a shortcoming that is faced in all cross-cultural work in which all analysts are not poly-lingual (see e.g., [52]). ## Conclusions This paper has explored the processes through which EuroFIT achieved its outcomes and the extent to which the underpinning theory of change represented the program’s mechanisms of action. Through a mixed-methods approach, we have concluded that the theory of change is an effective representation of the EuroFIT program, with only three small areas of ambiguity. In relation to this latter point, we have suggested: that the dual focus on changing physical activity and sedentary time may have led participants to select the behaviour they found most easy to grasp and alter (i.e. physical activity); that bespoke technologies can be complex to implement and bespoke social support platforms may have low take up; and that the appeal of the club as a site for health-related change may be culturally-mediated. Future deliveries should consider whether the distinction between decreasing sedentary behaviour and increasing physical activity is worth pursuing, should use easily available technologies to support self-monitoring and social interaction and how different forms of football fandom can be exploited to attract participants. ## Supplementary Information Additional file 1. ## References 1. Lee IM, Shiroma EJ, Lobelo F, Puska P, Blair SN, Katzmarzyk PT. Effect of physical inactivity on major non-communicable diseases worldwide: an analysis of burden of disease and life expectancy. The Lancet (2012.0) 380 219-229 2. Guthold R, Stevens GA, Riley LM, Bull FC. Worldwide trends in insufficient physical activity from 2001 to 2016: a pooled analysis of 358 population-based surveys with 1.9 million participants. Lancet Glob Health. (2018.0) 6 e1077-e1086. PMID: 30193830 3. Robertson C, Archibald D, Avenell A, Douglas F, Hoddinott P, Van Teijlingen E, Boyers D, Stewart F, Boachie C, Fioratou E. Systematic reviews of and integrated report on the quantitative, qualitative and economic evidence base for the management of obesity in men. Health technology assessment (Winchester, England) (2014.0) 18 v. PMID: 24857516 4. Hunt K, Wyke S, Gray CM, Anderson AS, Brady A, Bunn C, Donnan PT, Fenwick E, Grieve E, Leishman J. A gender-sensitised weight loss and healthy living programme for overweight and obese men delivered by Scottish Premier League football clubs (FFIT): a pragmatic randomised controlled trial. Lancet. (2014.0) 383 1211-21. PMID: 24457205 5. Petrella RJ, Gill DP, Zou G. DE AC, Riggin B, Bartol C, Danylchuk K, Hunt K, Wyke S, Gray CM: Hockey Fans in Training: A Pilot Pragmatic Randomized Controlled Trial. Med Sci Sports Exerc (2017.0) 49 2506-2516. PMID: 28719494 6. Maddison R, Hargreaves EA, Wyke S, Gray CM, Hunt K, Heke JI, Kara S, Mhurchu CN, Jull A, Jiang Y. Rugby Fans in Training New Zealand (RUFIT-NZ): a pilot randomized controlled trial of a healthy lifestyle program for overweight men delivered through professional rugby clubs in New Zealand. BMC Public Health (2019.0) 19 166. PMID: 30736781 7. Pietsch B, Weisser B, Hanewinkel R, Gray C, Hunt K, Wyke S, Morgenstern M. Short term effects of a weight loss and healthy lifestyle programme for overweight and obese men delivered by German football clubs. Eur J Sport Sci. (2020.0) 20 703-712. PMID: 31456489 8. Wyke S, Bunn C, Andersen E, Silva MN, Van Nassau F, McSkimming P, Kolovos S, Gill JM, Gray CM, Hunt K. The effect of a programme to improve men’s sedentary time and physical activity: The European Fans in Training (EuroFIT) randomised controlled trial. PLoS Med (2019.0) 16 e1002736. PMID: 30721231 9. Gray CM, Hunt K, Mutrie N, Anderson AS, Leishman J, Dalgarno L, Wyke S. Football Fans in Training: the development and optimization of an intervention delivered through professional sports clubs to help men lose weight, become more active and adopt healthier eating habits. BMC Public Health (2013.0) 13 232. PMID: 23496915 10. Gray CM, Wyke S, Zhang R, Anderson AS, Barry S, Boyer N, Brennan G, Briggs A, Bunn C, Donnachie C. Long-term weight loss trajectories following participation in a randomised controlled trial of a weight management programme for men delivered through professional football clubs: a longitudinal cohort study and economic evaluation. Int J Behav Nutr Phys Act (2018.0) 15 60. PMID: 29954449 11. Hunt K, Wyke S, Bunn C, Donnachie C, Reid N, Gray CM. Scale-up and scale-out of a gender-sensitized weight management and healthy living program delivered to overweight men via professional sports clubs: the wider implementation of Football Fans in Training (FFIT). Int J Environ Res Public Health (2020.0) 17 584. PMID: 31963224 12. van Nassau F, van der Ploeg HP, Abrahamsen F, Andersen E, Anderson AS, Bosmans JE, Bunn C, Chalmers M, Clissmann C, Gill JMR. Study protocol of European Fans in Training (EuroFIT): a four-country randomised controlled trial of a lifestyle program for men delivered in elite football clubs. BMC Public Health (2016.0) 16 1-15. PMID: 26728978 13. Moore GF, Audrey S, Barker M, Bond L, Bonell C, Hardeman W, Moore L, O’Cathain A, Tinati T, Wight D. Process evaluation of complex interventions: Medical Research Council guidance. BMJ. (2015.0) 350 h1258-<>h1258. PMID: 25791983 14. Grant A, Treweek S, Dreischulte T, Foy R, Guthrie B. Process evaluations for cluster-randomised trials of complex interventions: a proposed framework for design and reporting. Trials (2013.0) 14 15. PMID: 23311722 15. Wight D, Obasi A, Stephenson J, Imrie J, Bonell C. Effective sexual health interventions: issues in experimental evaluation. Sex Transm Infect. (2004.0) 80 77-78. PMID: 14755048 16. van de Glind I, Bunn C, Gray CM, Hunt K, Andersen E, Jelsma J, Morgan H, Pereira H, Roberts G, Rooksby J. The intervention process in the European Fans in Training (EuroFIT) trial: a mixed method protocol for evaluation. Trials (2017.0) 18 356. PMID: 28750673 17. Ryan RM, Deci EL. Self-determination theory and the facilitation of intrinsic motivation, social development, and well-being. Am Psychol (2000.0) 55 68. PMID: 11392867 18. Roberts GC, Treasure D. Advances in motivation in sport and exercise: Human Kinetics (2012.0) 19. Abraham C, Michie S. A taxonomy of behavior change techniques used in interventions. Health Psychol (2008.0) 27 379. PMID: 18624603 20. Michie S, Abraham C, Whittington C, McAteer J, Gupta S. Effective techniques in healthy eating and physical activity interventions: a meta-regression. Health Psychol (2009.0) 28 690. PMID: 19916637 21. Connell RW. Masculinities: Polity (2005.0) 22. Hunt K, Gray CM, Maclean A, Smillie S, Bunn C, Wyke S. Do weight management programmes delivered at professional football clubs attract and engage high risk men? A mixed-methods study. BMC Public Health (2014.0) 14 50. PMID: 24443845 23. Bunn C, Wyke S, Gray CM, Maclean A, Hunt K. ‘Coz football is what we all have’: masculinities, practice, performance and effervescence in a gender-sensitised weight-loss and healthy living programme for men. Sociol Health Illn. (2016.0) 38 812-28. PMID: 26864994 24. Martin A, Adams JM, Bunn C, Gill JMR, Gray CM, Hunt K, Maxwell DJ, van der Ploeg HP, Wyke S, Mutrie N. Feasibility of a real-time self-monitoring device for sitting less and moving more: a randomised controlled trial. BMJ Open Sport Exerc Med. (2017.0) 3 e000285-<>e000285. PMID: 29081985 25. Gill JM, Hawari NS, Maxwell DJ, Louden D, Mourselas N, Bunn C, Gray CM, Van Der Ploeg HP, Hunt K, Martin A. Validation of a novel device to measure and provide feedback on sedentary behavior. Med Sci Sports Exerc (2018.0) 50 525. PMID: 29040225 26. 26.van Nassau F, Huis A, van de Glind I, Andersen E, Bunn C, Gray CM, et al. Factors influencing the implementation of the EuroFIT lifestyle change program in professional football clubs in Europe: a qualitative study in four European countries. Behav Med Pract Policy Res. 2023. 10.1093/tbm/ibac100. 27. Røynesdal ØB, van Nassau F, Chng NR, Pereira H, Andersen E, Bunn C, Jelsma JGM, Nijhuis-van der Sanden MWG, Roberts GC, Sørensen M. Exploring the provision and motives behind the adoption of health-promotion programmes in professional football clubs across four European countries. Plos One. (2021.0) 16 e0259458-<>e0259458. PMID: 34797842 28. Geertz C. The Interpretation of Cultures Slected Essays (1977.0) 29. 29.Madden R. Being ethnographic: A guide to the theory and practice of ethnography. London: Sage; 2017. 30. Pope C, Ziebland S, Mays N. Qualitative research in health care. Analysing qualitative data BMJ (2000.0) 320 114-116. PMID: 10625273 31. Farmer T, Robinson K, Elliott SJ, Eyles J. Developing and Implementing a Triangulation Protocol for Qualitative Health Research. Qual Health Res (2006.0) 16 377-394. PMID: 16449687 32. Brown A, Crabbe T, Mellor G. ENGLISH PROFESSIONAL FOOTBALL AND ITS COMMUNITIES. Int Rev Mod Sociol (2006.0) 32 159-179 33. van der Ploeg HP, Hillsdon M. Is sedentary behaviour just physical inactivity by another name?. The international journal of behavioral nutrition and physical activity (2017.0) 14 142. PMID: 29058587 34. Martin A, Fitzsimons C, Jepson R, Saunders DH, van der Ploeg HP, Teixeira PJ, Gray CM, Mutrie N. Interventions with potential to reduce sedentary time in adults: systematic review and meta-analysis. Br J Sports Med (2015.0) 49 1056-1063. PMID: 25907181 35. Bottorff JL, Seaton CL, Johnson ST, Caperchione CM, Oliffe JL, More K, Jaffer-Hirji H, Tillotson SM. An Updated Review of Interventions that Include Promotion of Physical Activity for Adult Men. Sports Med (2015.0) 45 775-800. PMID: 25430599 36. Bunn C, Donnachie C, Wyke S, Hunt K, Brennan G, Lennox J, Maclean A, Gray CM. Can professional football clubs deliver a weight management programme for women: a feasibility study. BMC Public Health (2018.0) 18 1330. PMID: 30509224 37. Kwasnicka D, Ntoumanis N, Hunt K, Gray CM, Newton RU, Gucciardi DF, Thøgersen-Ntoumani C, Olson JL, McVeigh J, Kerr DA. A gender-sensitised weight-loss and healthy living program for men with overweight and obesity in Australian Football League settings (Aussie-FIT): A pilot randomised controlled trial. PLoS Med (2020.0) 17 e1003136. PMID: 32760144 38. Kwasnicka D, Donnachie C, Thøgersen-Ntoumani C, Hunt K, Gray CM, Ntoumanis N, McBride H, McDonald MD, Newton RU, Gucciardi DF. The Aussie-FIT process evaluation: feasibility and acceptability of a weight loss intervention for men, delivered in Australian Football League settings. Psychol Health (2022.0) 37 470-489. PMID: 33719789 39. Duncan M, Vandelanotte C, Kolt GS, Rosenkranz RR, Caperchione CM, George ES, Ding H, Hooker C, Karunanithi M, Maeder AJ. Effectiveness of a web-and mobile phone-based intervention to promote physical activity and healthy eating in middle-aged males: randomized controlled trial of the ManUp study. J Med Internet Res (2014.0) 16 e136. PMID: 24927299 40. Blunt W, Gill DP, Sibbald SL, Riggin B, Pulford RW, Scott R, Danylchuk K, Gray CM, Wyke S, Bunn C. Optimization of the Hockey Fans in Training (Hockey FIT) weight loss and healthy lifestyle program for male hockey fans. BMC Public Health (2017.0) 17 916. PMID: 29183291 41. Krustrup P, Bangsbo J. Recreational football is effective in the treatment of non-communicable diseases. Br J Sports Med. (2015.0) 49 1426-1427. PMID: 25990758 42. Yancey A, Winfield D, Larsen J, Anderson M, Jackson P, Overton J, Wilson S, Rossum A, Kumanyika S. “Live, Learn and Play”: Building strategic alliances between professional sports and public health. Prev Med (2009.0) 49 322-325. PMID: 19595702 43. Khan KM, Thompson AM, Blair SN, Sallis JF, Powell KE, Bull FC, Bauman AE. Sport and exercise as contributors to the health of nations. The Lancet (2012.0) 380 59-64 44. Finch CF, Donaldson A. A sports setting matrix for understanding the implementation context for community sport. Br J Sports Med (2010.0) 44 973-978. PMID: 19201766 45. Gray CM, Wyke S, Zhang R, Anderson AS, Barry S, Boyer N, Brennan G, Briggs A, Bunn C, Donnachie C. Long-term weight loss trajectories following participation in a randomised controlled trial of a weight management programme for men delivered through professional football clubs: a longitudinal cohort study and economic evaluation. Int J Behav Nutr Phys Act. (2018.0) 15 60. PMID: 29954449 46. Quested E, Kwasnicka D, Thøgersen-Ntoumani C, Gucciardi DF, Kerr DA, Hunt K, Robinson S, Morgan PJ, Newton RU, Gray C. Protocol for a gender-sensitised weight loss and healthy living programme for overweight and obese men delivered in Australian football league settings (Aussie-FIT): A feasibility and pilot randomised controlled trial. BMJ Open (2018.0) 8 e022663. PMID: 30337315 47. Nash R. Globalised football fandom: Scandinavian Liverpool FC supporters. Football Studies (2000.0) 3 5-23 48. 48.Onwumechili C, Akindes G. Identity and nation in African football: Fans, community and clubs. London: Springer; 2014. 49. Giulianotti R, Robertson R. Recovering the social: globalization, football and transnationalism. Global Netw (2007.0) 7 166-186 50. Hognestad H. Transglobal Scandinavian? Globalization and the contestation of identities in football. Soccer & Society (2009.0) 10 358-373 51. Hognestad HK. Split loyalties: football is a community business. Soccer & Society (2012.0) 13 377-391 52. Mangen S. Qualitative research methods in cross-national settings. Int J Soc Res Methodol (1999.0) 2 109-124
--- title: 'Examination of physical activity development in early childhood: protocol for a longitudinal cohort study of mother-toddler dyads' authors: - Sarah B. Welch - Kyle Honegger - Megan O’Brien - Selin Capan - Soyang Kwon journal: BMC Pediatrics year: 2023 pmcid: PMC10026417 doi: 10.1186/s12887-023-03910-9 license: CC BY 4.0 --- # Examination of physical activity development in early childhood: protocol for a longitudinal cohort study of mother-toddler dyads ## Abstract ### Background Physical activity (PA) development in toddlers (age 1 and 2 years) is not well understood, partly because of a lack of analytic tools for accelerometer-based data processing that can accurately evaluate PA among toddlers. This has led to a knowledge gap regarding how parenting practices around PA, mothers’ PA level, mothers’ parenting stress, and child developmental and behavioral problems influence PA development in early childhood. ### Methods The Child and Mother Physical Activity Study is a longitudinal study to observe PA development in toddlerhood and examine the influence of personal and parental characteristics on PA development. The study is designed to refine and validate an accelerometer-based machine learning algorithm for toddler activity recognition (Aim 1), apply the algorithm to compare the trajectories of toddler PA levels in males and females age 1–3 years (Aim 2), and explore the association between gross motor development and PA development in toddlerhood, as well as how parenting practices around PA, mothers’ PA, mothers’ parenting stress, and child developmental and behavioral problems are associated with toddlerhood PA development (Exploratory Aims 3a-c). ### Discussion This study will be one of the first to use longitudinal data to validate a machine learning activity recognition algorithm and apply the algorithm to quantify free-living ambulatory movement in toddlers. The study findings will help fill a significant methodological gap in toddler PA measurement and expand the body of knowledge on the factors influencing early childhood PA development. ## Background Physical activity (PA) has numerous health benefits in children, including cardiovascular health, bone health, and motor and cognitive development [1]. Given that physical inactivity habits start to develop before age 5 years and are sustained over time [2, 3], it is alarming that a substantial proportion of preschoolers (age 3 and 4 years) are not sufficiently physically active [4–6]. Moreover, some studies [4–6], but not all [7], have shown that at the preschool age, females are less physically active than males. To better understand when and how PA habits develop in childhood, research investigating PA in children needs to focus on earlier age groups. Yet PA research in toddlers (age 1 and 2 years) is currently limited [1, 8], primarily because of a lack of analytic tools, such as accelerometer-based data processing, that can accurately evaluate PA in this age group [9]. Early childhood is a critical period in which children acquire movement proficiency and rapidly develop a range of motor skills [10]. By age 1 and 2 years, most children are expected to have reached important milestones for PA, such as walking and throwing a ball. Gross motor skills are viewed as the building blocks and foundation of PA in early childhood [11]. Gross motor competence has been shown to be cross-sectionally associated with PA in early childhood [12–14], including among children 2 years of age [15]. However, the association between gross motor competence and PA, as well as their temporal relationship through early childhood, has rarely been prospectively examined. Parents [16–20], particularly mothers [21], play a critical role in shaping PA behavior during toddlerhood [22]. Parenting practices around PA are associated with a greater amount of PA in school-age children and older [23, 24]. In early childhood, parental modeling has been highlighted as a significant factor in determining child PA [25]. Maternal mental health status, such as depressive symptoms or psychological distress, is also associated with obesity in young children with and without developmental delay [26, 27]. However, the influence of parenting practices around PA, mothers’ PA, mothers’ parenting stress, and child developmental and behavioral problems on PA development in early childhood remains unclear. Filling these knowledge gaps requires the development and application of new methodological approaches to accelerometer data processing for toddlers due to the lack of analytic tools to measure PA for this age group. We conducted a pilot study applying machine learning to the analysis of raw acceleration signals of accelerometer data among toddlers [28]. This study demonstrated that the machine learning analytic approach holds enormous potential for toddler activity recognition, with $79\%$ accuracy [28]. We also conducted a feasibility study [8], which demonstrated acceptable compliance among toddlers to wearing a hip accelerometer in free-living settings. Building on these two pilot studies, we designed the longitudinal Child and Mother Physical Activity Study (CAMPAS) to achieve the following specific aims:Aim 1: To refine and validate an accelerometer-based algorithm for toddler activity recognition; Hypothesis: a newly developed algorithm will have higher accuracy than the Albert algorithm ($79\%$) from our pilot study. Aim 2: To compare the trajectory of PA levels from age 1 to 3 years between males and females; Hypothesis: males will show a greater increase in PA between the ages of 1 and 3 years compared to females. To explore the extent to which mothers’ and children’s characteristics influence the PA trajectories evaluated in Aim 2, we also have three exploratory aims:Aim 3a: To examine the association between gross motor development and PA development in toddlerhood. Aim 3b: To examine whether parenting practices around PA and mothers’ PA are associated with PA development in toddlerhood. Aim 3c: To examine whether mothers’ parenting stress and child developmental and behavioral problems are associated with PA development in toddlerhood. ## Study design The CAMPAS is a longitudinal study to observe PA development in toddlerhood and examine the influences of personal and parental characteristics on PA development. We plan to recruit 124 children (10–15 months of age) and their mothers, and conduct five waves of assessments at 6-month intervals (at the approximate child chronological ages of 12, 18, 24, 30 and 36 months). For Aim 1, a subsample of the child cohort ($$n = 62$$) will complete a validation study in waves 1 to 4. This subsample will be selected to ensure that half of the sample is female (31 children) and at least a third (21 children) comprises children from a preterm follow-up clinic or a developmental and behavioral pediatric clinic. In the Aim 1 validation study, each child participant will wear an accelerometer on one hip to collect data for 20 min in each of four settings in waves 1–4 during which period their movements will also be video recorded. The matched accelerometer and annotated video data will be used to develop and validate our algorithm for toddler PA recognition. For Aims 2 and 3, each child participant will wear an accelerometer on one hip for 24 h a day for 7 days in each of the five waves. For Aim 3a, mothers will proxy-report child gross motor development in all five waves. For Aim 3b, mothers will report their own PA and parenting practices around PA in waves 1, 3, and 5. For Aim 3c, mothers will report their parenting stress and child developmental and behavioral problems in waves 2 and 4. ## Study setting The CAMPAS will be conducted in the Chicago area in Illinois, USA. Chicago is home to diverse racial and ethnic groups: $33\%$ non-Hispanic White, $29\%$ non-Hispanic Black, $29\%$ Hispanic, and $7\%$ Asian [29]. Study sites for data collection will include two commercial child play spaces (one located in the city of Chicago and the other in a Chicago suburb), participants’ homes, and public parks. ## Participant eligibility criteria The eligibility criteria for child participants are age 10 to 15 months at the baseline assessment, without cerebral palsy or other medical conditions precluding physical movement. Only one child per mother will be eligible. The eligibility criteria for mother participants are age 18 years or older, self-identifies as the participating child’s mother, can communicate in English or Spanish, and lives with the participating child at least $50\%$ of the time. ## Participant recruitment Mother–child dyads will be recruited during a 2-year period, primarily from two source populations: patients from various pediatric clinics and users of partnered community-based organizations/programs. Recruitment pediatric clinics will include a preterm follow-up clinic, a developmental and behavioral pediatric clinic, and multiple pediatric primary care sites. The preterm follow-up clinic serves patients who were discharged from the Neonatal Intensive Care Unit (NICU) or Cardiac Intensive Care Unit (CICU) to monitor their development until age 5 years. The developmental and behavioral pediatric clinic serves pediatric patients for diagnosis and treatment of developmental behavioral disorders. These clinics were selected to recruit a developmentally diverse sample (e.g., preterm births). Partnered community organizations/programs will include a Chicago commercial child playroom, a suburban park district, and the Maternal Infant and Early Childhood Home Visiting (MIECHV) Program. MIECHV is a national effort to support pregnant people and parents with early childhood support to increase child health and well-being [30]. We conducted our pilot study [28] and a community-based PA intervention program with these organizations, and we will leverage these existing partnerships for the CAMPAS. Recruitment flyers will be placed in clinics (e.g., waiting and exam rooms) and community partner facilities and posted on social media. Flyers will be given to home visitors to distribute to participants of the MIECHV program. The flyers will be created in both English and Spanish and include a QR code that will take interested adults directly to an online screening survey. A phone number will also be provided so individuals can call the research team directly to learn more and complete the screening survey by phone. After eligibility is established, a member of the research team will begin the consent process with the potential participant by providing consent documents for their review, answering any initial questions, and scheduling the first visit for completion of the consent process, enrollment, and the baseline assessment. ## Participant retention Because the CAMPAS is a 2-year longitudinal study, participant retention is critical. Our retention goal is ≥ $90\%$ for a subsequent assessment, anticipating a final (wave 5) retention rate of $66\%$ (0.9 × 0.9 × 0.9 × 0.9) over the 2-year follow-up period. The study will employ multiple strategies to reach this retention goal. To build rapport, each participant will have an assigned primary research staff member. This staff member will seek to develop relationships with their assigned families by managing all contact and interaction with them including scheduling and conducting their study visits; engaging in regular communication through phone calls, short text messages, and emails; and sending the participants birthday cards. Data collection visits will occur every 6 months and communication reminders halfway between these visits will ensure that research staff are in contact with participants every 3 months. We will obtain the phone number and email of another contact person who may be reached (e.g., aunt, sister, cousin) in case the participant contact information changes. Finally, if a participant misses an assessment at the proposed age, we will continue to invite the participant to complete five assessments at any age between 10 and 40 months, if there is at least a 4-month interval between assessments. If a participant moves out of the area, we will offer remote assessment options, such as Zoom meetings, online surveys, and mailing an accelerometer package and other study materials. We will provide incremental compensation over waves ($40 at wave 1, $50 at wave 2, $60 at wave 3, $70 at wave 4, and $80 at wave 5; validation study participants (Aim 1) will receive an additional $10 for each of waves 1–4). Monthly retention goals will be set and monitored. ## Study assessments (Table 1) Table 1 \| Instrument \| Wave 1(12 Mo) \| Wave 2(18 Mo) \| Wave 3(24 Mo) \| Wave 4(30 Mo) \| Wave 5(36 Mo) \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| Participant demographics \| X \| \| X \| \| \| \| Anthropometry (recumbent length/height) \| X \| X \| X \| X \| X \| \| Accelerometer and video recording (validation sub-sample; Aim 1) \| X \| X \| X \| X \| \| \| Accelerometer (7-day wear; Aims 2–3) \| X \| X \| X \| X \| X \| \| Ages & Stages Questionnaire (Aim 3a) \| X \| X \| X \| X \| X \| \| Preschooler Physical Activity Parenting Practices (Aim 3b) \| X \| \| X \| \| X \| \| Physical Activity Questionnaire for mothers (Aim 3b) \| X \| \| X \| \| X \| \| Parental Stress Scale (Aim 3c) \| \| X \| \| X \| \| \| Parenting Daily Hassles Scale (Aim 3c) \| \| X \| \| X \| \| \| Strengths and Difficulties Questionnaire (Aim 3c) \| \| X \| \| X \| \| ## Validation study for Aim 1 To further develop and validate our accelerometer-based activity recognition algorithm, we will longitudinally collect validation study data at waves 1–4 in four different settings where toddlers often spend their time: home, indoor play space, childcare class, and outdoor playground. Data collection from child participants must be completed at each of the four main locations in waves 1–4, but the data collection may be completed in any order. We will collect accelerometer data during automobile rides from a sub-sample (20 participants) to collect accelerometer signals in this location. We will not collect validation study data at wave 5 (age 36 months), because physical/motor development in late toddlerhood is slower than in early toddlerhood and because a validated algorithm for children 3 to 5 years of age (the Trost algorithm) already exists [31]. The subsample of child participants ($$n = 62$$) included in the validation study will wear an ActiGraph wGT3X-BT accelerometer (a range of ± 6 g; sampling at 100 Hz; Pensacola, FL) on the hip. ActiGraph accelerometers are the most widely used activity monitor in pediatric PA research. The child participants will then be allowed to engage in natural free-living activities in the given setting (e.g., indoor play space) for 20 min, such as interacting with their caregiver, siblings, and/or environment. Simultaneously, their activities will be video recorded using a GoPro Hero10 Black camera. ## Child physical activity measurement for Aims 2 and 3 To assess their PA, child participants will wear an ActiGraph wGT3X-BT accelerometer for 7 days, for 24 h a day, in each of waves 1–5. The accelerometer will be worn at the level of the anterior superior iliac spine, underneath or on top of clothes, and on the side of the body (right or left) chosen by participants or caregivers for comfort. We elected to place one accelerometer per participant because wearing multiple accelerometers increases participant burden and reduces compliance and because prior preschooler studies showed that combined hip and wrist data did not significantly improve the machine learning algorithm performance, compared to hip data only [31, 32]. Caregivers will be asked to complete a monitor on/off time and sleep time log sheet. After the 7-day wear is complete, parents/caregivers will be asked to schedule a time for accelerometer pickup or instructed to mail the study package using the pre-paid envelope. With the chosen ActiGraph settings, the expected battery life exceeds the requested 7-day wear period. To promote wear-time adherence, a research staff member will contact the family midway through the 7-day wear period to check that the child participant is wearing the accelerometer and to troubleshoot any issues they might be having. As research staff receive the returned accelerometers, they will download, visually inspect, and pre-analyze the accelerometer data using the ActiLife software [33] to determine whether there are sufficient non-zero (wear) data (defined as ≥ 500 min per day for ≥ 3 days). If not, participants will be asked to re-wear an accelerometer to meet the 3-day criteria. ## Gross motor competency for Aim 3a Gross motor skills will be assessed by mothers using the Ages & Stages Questionnaire gross motor subscale (6 items) at each of the 5 waves [34]. The Ages & Stages Questionnaires are tailored to child age; the appropriate age-corresponding questionnaire will be used at each study visit. ## PA parenting practices and mothers’ PA for Aim 3b PA parenting practices will be measured using the Preschooler Physical Activity Parenting Practices (PPAPP) questionnaire at child age 12, 24, and 36 months (waves 1, 3, and 5) [35]. We selected this preschooler tool because there is no validated tool for toddlers. The questionnaire contains 30 items organized into 5 subscales: 15 items for PA encouragement, 3 items for inactivity promotion, 5 items for psychological control, 3 items for screen time promotion, and 4 items for restriction for safety concerns. Mother respondents will be instructed to choose the best answer on a 5-point Likert response scale (1 = never, 2 = rarely, 3 = sometimes, 4 = often, and 5 = always). The responses will be coded as 1 to 5 points and a subscale score will be calculated by averaging response points. To test the hypothesis that maternal PA behavior influences child PA behavior via maternal PA modeling, we will also collect data on mothers’ PA as a parenting practice. At child age 12, 24, and 36 months (waves 1, 3, and 5), mothers will complete a self-report PA questionnaire that asks about the weekly frequency and duration of walking and other moderate- and vigorous-intensity PA, adopted from the National Health and Nutrition Examination Survey (NHANES) PA questionnaire [36] and the Family Life, Activity, Sun, Health and Eating (FLASHE) Study PA questionnaire [37]. Mothers will also report their average step counts for the prior week and month from their personal fitness tracking device or smartphone application (e.g., Fitbit, Health app for iPhone, Google Fit for Android), if available. ## Mothers’ parenting stress for Aim 3c Parenting stress will be measured using the Parental Stress Scale (PSS) [38] and the Parenting Daily Hassles Scale (PDHS) questionnaires, completed by participant mothers at child ages 18 and 30 months (waves 2 and 4) [39]. The PSS is an 18-item self-report scale that measures the level of stress experienced by a parent in the past month, accounting for both positive (e.g., emotional benefits, personal development) and negative (e.g., demands on resources, restriction) aspects of parenting. Each item is rated on a 5-point Likert scale (1 = strongly disagree to 5 = strongly agree). The PDHS is a 20-item self-report scale for assessing the frequency and intensity of 20 experiences that can be a “hassle” to parents of young children in the past month (e.g., “continuously cleaning up messes of toys or food”). The frequency provides an objective marker of how often those events occur, while intensity indicates a subjective appraisal of how much those events “hassle” or affect a parent. The frequency will be rated on a 4-point scale (1 = rarely, 2 = sometimes, 3 = a lot, and 4 = constantly). Intensity will be rated on a 5-point scale (from 1 = low to 5 = high). The PDHS response will be calculated by summing the scores from the frequency scale and the intensity scale. ## Child developmental and behavioral problems for Aim 3c To assess child externalizing behavioral problems, we will use the Strengths and Difficulties Questionnaire [40]. At child ages 18 and 30 months (waves 2 and 4), mothers will complete the Conduct Problem (5 items) and Hyperactivity (5 items) subscales of the Strengths and Difficulties Questionnaire version modified for children 1 and 2 years of age [41]. We elected to measure child externalizing behavior problems from age 18 months because externalizing behaviors begin to increase at age 12–17 months [41]. We believe that these 10 items are sensitive enough to capture externalizing behavior problems, given that in a study using only 5 of the 10 items, we identified $33\%$ of preschool-age children with externalizing behavior problems in a US representative sample [42]. Each item will be rated on a 3-point scale (0 = not true, 1 = somewhat true, and 2 = very true). The externalizing behavioral problem score will be calculated by summing the scores of the 10 items. ## Other measurements We will collect participant information on race/ethnicity, main caregiver/childcare attendance, household structure (e.g., number of siblings in household), mother’s employment, mother’s education, neighborhood characteristics, brief child health history, and other caregiver information at wave 1. Items likely to change will be updated at wave 3. Research staff trained in pediatric anthropometry assessment will measure child recumbent length and weight in each of the five waves. ## Data collection procedures All communication with participants will occur, and all written materials will be available, in English or Spanish, depending on participant preference. At each study visit, a research team member will collect all scheduled demographic data, anthropometric measurements, and survey items, and distribute the accelerometer, two waist belts, an instruction sheet, a wear time log sheet, and a pre-paid envelope for accelerometer return. If the child is included in the subsample of 62 participants in the validation study (Aim 1), they will play for 20 min while wearing an accelerometer and while the research team member video tapes them. At the completion of all study visit activities, participants will be given their 7-day wear accelerometer, shown how to wear it on their waist and be given instructions to take home along with a wear-time log sheet. The log sheet will be used to report monitor on/off time and sleep time for the 7 days of accelerometer wear following the study visit. As research staff receive the returned accelerometers, they will download, visually inspect, and pre-analyze the accelerometer data using the ActiLife software to determine whether there are sufficient wear data, defined as ≥ 500 min per day and ≥ 3 days. If there are not sufficient data, research team members will ask the participant to re-wear the accelerometer for another week. All project data will be collected and stored in REDCap [43, 44]. ## Accelerometer and video data processing for Aim 1 Three-axis accelerometer raw data (g-force) will be extracted from the device using the ActiLife software. Accelerometer data will be segmented into non-overlapping sliding windows, based on multiple window sizes suggested in prior studies (e.g., 5, 10, and 15 s) [28, 31]. We will extract the engineered features from the Albert and Trost algorithms (e.g., percentiles, dominant frequencies, standard deviation in lag and lead windows, etc.) [ 28, 31, 32] and from angle analysis for posture estimation [45]. The angle analysis was added to reduce confusion between “sedentary” and “standing” and between “walking/running” and “carried,” as shown in our pilot study [28]. We will engineer additional features by examining the angle values (defined as incident accelerometer orientation in relation to the gravity vector) for posture estimation [45], using the pilot data. Ground truth activity types will be labeled based on video data using Behavioral Observation Research Interactive Software (BORIS) [46]. The coding guidelines developed for the pilot study [28] will be updated as needed. We will also reference a list of the 23 activity types for preschoolers used for the development of the Trost algorithm [31, 47, 48]. Two research staff will independently label the first four participant videos. The coded data (activity class label and activity start/end times in seconds) will be compared between the two coders to identify any disagreements. To resolve disagreements, the research team will review the video images associated with the disagreements, discuss, come to consensus, and update the coding guidelines, as needed. Based on the updated guidelines, the coders will independently label the next four participant videos and interobserver reliability will be calculated. This reconciliation process will be repeated until the inter-observer reliability reaches ≥ $95\%$. The remaining video files will be divided among the two coders for coding. After coding is complete, $10\%$ of the coded files will be randomly selected to calculate the final inter-observer reliability. Detailed activity types will be re-classified into the five activity classes of interest: “walking/running,” “non-walking/running PA,” “light PA,” “sedentary,” and “being carried by an adult.” ## Accelerometer data processing for Aims 2 and 3 The 7-day raw accelerometer data will be extracted from the device using the ActiLife software. Non-wear and sleep periods will be identified. A non-wear period will be defined as ≥ 20 consecutive minutes of zero counts [49–51]. Participants with at least 500 min of wear per day and 3 wear-days will be included for analysis. A sleep period, defined based on mother-reported sleep onset (bedtime) and sleep offset (waking time), will be identified based on a sleep algorithm developed by Tracy et al. [ 52]. The wear time log sheet will also be used to cross-check non-wear and sleep periods. ## Aim 1 analysis We will further develop our machine learning algorithm from our pilot study to classify PA in toddlers. Ground truth activities (from the video coders) will be synchronized with the accelerometer data based on detection of a specific movement pattern simultaneously acquired by camera and accelerometer. Data from the validation study will be split into a training set for model development and a testing set for model validation. A random forest classifier has been selected for primary development because it is a widely implemented ensemble-based classifier [31], showing the highest accuracy for activity recognition [28, 32, 48, 53]. Additional machine learning algorithms may be considered to maximize classification performance for the dataset. ## Model training A random forest classifier will be trained to classify the five activities (walking/running, non-walking/running PA, standing, sedentary, and being carried), using the feature set from the Albert [28] and the Trost algorithms [31]. Feature selection and dimensionality reduction techniques (e.g., principal component analysis, feature embedding) will be considered to minimize risk of overfitting. A grid search approach will be used to explore a range of window sizes (5, 10, and 15 s) [54] and hyperparameter values (e.g., the number of trees used in the random forest). From these combinations, the optimal set of parameters will be chosen based on cross-validated performance on a validation sample. A leave-one-subject-out (LOSO) cross-validation will be performed to assess overall classification accuracy. Precision, recall, F1 score, and confusion matrices of the classification will be examined to learn about the patterns of misclassifications and refine the model. ## Model evaluation Using the testing dataset, we will generate confusion matrices of the five-activity classification. Model performance (precision, recall, and F1 score) for activity classification will be estimated per participant and averaged to determine overall performance. We will evaluate the overall model performance using weighted kappa [47]. Next, we will evaluate the validity of the new algorithm for estimation of walking/running time (minutes) and overall PA time (the sum of walking/running time and non-walking/running PA time; minutes) against the ground truth data, using confusion matrices [55] as well as Bland Altman plots and root mean squared error. To explore whether the model performance differs by sex or age, we will repeat these analyses by sex assigned at birth and by age (1 year vs. 2 years of age). ## Aim 2 and 3 analyses To be included in the PA trajectory analysis (Aim 2), participants should complete at least three of the five waves of PA assessments. We will estimate daily walking/running time (minutes/day) and daily overall PA time (minutes/day) using the algorithm developed and validated in Aim 1. Multiple days of PA data per assessment wave will be averaged. As we hypothesize that individuals will present diverse start levels (intercept) and change rates (slope) of the outcomes, a growth curve model will be used to describe deviation or variation of individuals’ outcome values from the population growth curve. A growth curve model will be fit to depict the trajectory of daily walking/running time over age in months from age 12 to 36 months. A quadratic model will be initially fit to allow for a non-linear trajectory. The growth curve model analysis will be repeated for the daily overall PA time outcome. To examine whether PA trajectory differs by sex, the sex variable will be added as a predictor. The interaction of sex and age (sexage) will also be considered. The model will also account for the season of assessment, distinguishing between winter months (December to March) vs. other seasons. ## Aim 3a analysis To examine the association of child PA with gross motor skills, we will expand the growth curve model fitted in Aim 2 and conduct mixed model analysis by adding gross motor skill scores as a predictor. ## Aim 3b and 3c analyses We will conduct bivariate analyses between child PA the potential correlates (parenting practices around PA, mothers’ PA, mothers’ parenting stress, and child behavioral problems). Any factors that are found to be statistically significantly associated with child PA in the bivariate analyses ($p \leq 0.05$) will be selected to be included in the subsequent multivariable mixed model analysis. The mixed linear regression model will predict child PA by the selected factors. The model will include within-subject random effects for repeated measures. In addition, the following covariates will be considered: mother’s racial and ethnic background, mother’s education level, mother’s employment status, the number of young children in the household, child’s childcare attendance, child sex, and child age. ## Sample size calculation for Aim 1 We determined a sample size for Aim 1 based on a prior study [47] that examined the validity of a machine learning activity recognition algorithm for preschool-age children. That study had a sample size of 31 children and collected data during a 20-minute free-play session. The study found that overall accuracy for a hip random forest classifier (15-s window) was $81\%$ ($95\%$ confidence interval [CI] = 79–$83\%$; width of CI = $4\%$), which was significantly higher than the accuracy of a wrist random forest classifier ($75\%$; $95\%$ CI = 73–$77\%$). Given that the Albert algorithm from our pilot study showed $79\%$ accuracy [28], we estimated that a sample size of 31 children would be sufficient to detect at least $4\%$ accuracy improvement from $79\%$ to ≥ $84\%$. As we will equally split the validation study data into a training set and a testing set, 62 participants (31 participants for training and 31 participants for testing) will be required for Aim 1. ## Sample size calculation for Aims 2 and 3 Aim 2 and 3 statistical analyses will include participants who have at least three waves of accelerometer data. As wave 3 is predicted to have the smallest sample size among waves 1, 2, and 3 due to drop out, we calculated a required sample size to detect sex difference at wave 3. Based on our pilot study that showed 8 min less moderate and vigorous PA among females than males with a standard deviation (SD) of 14 min, 100 participants (50 males and 50 females) would be required to reject the null hypothesis of equal means when the population mean difference is 8 min with a SD for both groups of 14 min (effect size of 0.57) at $80\%$ power and a significance level of 0.05 using a two-sided two-sample equal-variance t-test. Also, published recommendations regarding a sample size for growth curve modeling suggest that a sample size approaching at least 100 is preferred and at least 3 repeated measures per individual are typically required [56]. Therefore, our goal is to retain at least 100 participants at wave 3. Assuming a $90\%$ retention at each subsequent wave or an $81\%$ (0.9 × 0.9) retention rate at wave 3, we will need to recruit 124 participants (100 ÷ 0.81) at baseline. ## Discussion To our knowledge, the CAMPAS will be one of the first studies to use longitudinal data to validate a machine learning activity recognition algorithm and apply the algorithm to quantify free-living ambulatory movement in toddlerhood. This study will help fill a significant methodological gap in toddler PA measurement and expand the body of knowledge in early childhood PA. The algorithm that will be further developed and validated in the CAMPAS will improve upon current algorithms used for activity classification of young children. The Albert algorithm was developed in activity trials and, as such, its performance is expected to decline substantially when applied in free-living settings [57, 58]. Current algorithms, both cut-point and machine learning algorithms, are mostly based on laboratory data, which often exhibits a substantial reduction in accuracy (20–$50\%$ in adults [59–61] and 11–$15\%$ for preschoolers [48]) when implemented in free-living conditions. Model training and validation on free-living data is crucial because the algorithms will ultimately be utilized for analyzing free-living data. Recently, Ahmadi and Trost demonstrated that among preschool-age children, machine learning-based PA classification model showed substantially higher accuracy (weighted kappa = 0.72) than four cut-point methods (weighted kappa = 0.31 to 0.44) [47]. Therefore, a machine learning approach has the potential to perform more accurate toddler activity recognition than the traditional cut-point approach. ## Potential issues and limitations While we have made every attempt to identify and prepare for practical or operational issues, we know that behavioral research including human subjects is inherently susceptible to deviations from the planned protocol. We may find that the new algorithm developed in Aim 1 does not perform better than the existing Albert algorithm or the Trost algorithm. In this case, we would use the existing algorithm to estimate walking/running time and overall PA time for Aims 2 and 3. Our approach of assigning a single activity to a window with multiple activities could cause classification error. To assess its impact on classifier performance, we will conduct sensitivity analyses to examine whether classifier performance differs with and without the mixed-activity windows. Activity labeling is a time-consuming task and could pose a potential practical challenge. We will explore innovative ways to reduce the burden, for example, by implementing partially automated video-based labeling and unsupervised learning techniques [62]. Although we plan to collect data in four common free-living settings where most toddlers often spend their time, these settings still do not fully represent all free-living environments. In Aims 2 and 3, some participants may drop out or have incomplete accelerometer data in some waves even after the imputation of incomplete accelerometer data. Growth curve modeling allows for missing data and provides valid results under the assumption of missing data at random where the missing-ness is “ignorable.” However, the possibility of non-ignorable missing-ness cannot be ruled out. Therefore, we will conduct sensitivity analyses using non-ignorable pattern-mixture and selection models to assess the robustness of our conclusions across these different models for missing data [63]. For Aim 2, even if we do not find sex differences in PA trajectory, gross motor skill and PA parenting practice data can still be utilized to examine the associations with PA trajectory in exploratory Aim 3. In Aim 3c, mothers’ self-report of psychological stress is prone to measurement error. To complement the parenting stress measure, we will collect the objective markers of stressful events using the PDHS. Mothers’ self-reported PA is also a limitation. To complement mothers’ self-reported PA data, we will collect step count data recorded in the fitness tracking application of their smartphone. Finally, we understand that sex, gender, identity, and primary caregiving role are neither static nor uniform, and that families come in many different configurations and not all of them include mothers. The focus on mothers and the exclusion of parents who do not identify as the child's mother is a consequence of the existing literature base, which specifically reports the influence of the mother on child PA. Further, the literature suggests that children and mothers bidirectionally influence each other’s PA behaviors. Mothers’ PA reduction is largely explained by lack of time and fatigue [64–66] due to new and increased stressors related to childcare [67, 68]. Parenting stress is particularly high among mothers of young children with externalizing behavior problems, which are known to peak at age 2–3 years [69] and present in one in three young children in the US [42]. Our study protocol recognizes that parents [16–20], particularly mothers [21], play a critical role in shaping child PA habits. ## Study implications Our study will develop and validate a monitor-independent machine learning algorithm for toddler activity recognition using longitudinal free-living accelerometer data. Utilizing this algorithm, we will describe the PA trajectory from age 12 to 36 months, with a focus on sex differences in PA development during this period and the influence of mothers on PA behaviors in toddlers. The output of this study will contribute to knowledge of PA development among males and females in early childhood (age 1 and 2 years) and the role that mothers play in influencing toddler PA practices. The new analytic tool developed and validated in Aim 1 will be made publicly available to facilitate dissemination and application by other researchers. Understanding the trajectory of toddler PA and the influence of mothers on PA behaviors during this time is critical for revealing new opportunities for program and policy interventions that will promote increased toddler PA. Ultimately, this study will provide necessary epidemiologic data that can contribute to establishing Physical Activity Guidelines for children under age 3 years. ## References 1. 1.2018 Physical Activity Guidelines Advisory Committee. 2018 Physical Activity Guidelines Advisory Committee Scientific Report. Washington, DC: U.S. Department of Health and Human Services; 2018. 2. Kwon S, Janz K, Letuchy E, Trudy B, Steven L. Developmental trajectories of physical activity, sports, and television viewing during childhood to young adulthood. JAMA Pediatr* (2015.0) 169 666-672. DOI: 10.1001/jamapediatrics.2015.0327 3. Telama R. Tracking of physical activity from childhood to adulthood: a review. Obes Facts (2009.0) 2 187-195. DOI: 10.1159/000222244 4. 4.Hinkley T, Salmon J, Okely AD, Crawford D, Hesketh K. Preschoolers’ physical activity, screen time, and compliance with recommendations. Med Sci Sports Exerc. 2012;44(3):458–65. 5. Beets MW, Bornstein D, Dowda M, Pate RR. Compliance with national guidelines for physical activity in U.S. preschoolers: measurement and interpretation. Pediatrics (2011.0) 127 658-664. DOI: 10.1542/peds.2010-2021 6. 6.Pate RR, O’Neill JR, Brown WH, Pfeiffer KA, Dowda M, Addy CL. Prevalence of compliance with a new physical activity guideline for preschool-age children. Child Obes. 2015;11(4):415–20. 7. 7.Kwon S, Tandon PS, O’Neill ME, et al. Cross-sectional association of light sensor-measured time outdoors with physical activity and gross motor competency among U.S. preschool-aged children: the 2012 NHANES National Youth Fitness Survey. BMC Public Health. 2022;22:833. 10.1186/s12889-022-13239-0. 8. Kwon S, Honegger K, Mason M. Daily physical activity among toddlers: hip and wrist accelerometer assessments. Int J Environ Res Public Health (2019.0) 16 4244. DOI: 10.3390/ijerph16214244 9. 9.Bruijns BA, Truelove S, Johnson AM, Gilliland J, Tucker P. Infants’ and toddlers’ physical activity and sedentary time as measured by accelerometry: a systematic review and meta-analysis. Int J Behav Nutr Phys Act. 2020;17(1):14. 10. Gallahue D, Ozmun J, Goodway J. Understanding motor development: infants, children, adolescents, adults (2012.0) 11. Veldman SLC, Jones RA, Santos R, Sousa-Sá E, Pereira JR, Zhang Z, Okely AD. Associations between gross motor skills and physical activity in Australian toddlers. J Sci Med Sport (2018.0) 21 817-821. DOI: 10.1016/j.jsams.2017.12.007 12. Fisher A, Reilly JJ, Kelly LA, Montgomery C, Williamson A, Paton JY, Grant S. Fundamental movement skills and habitual physical activity in young children. Med Sci Sports Exercise (2005.0) 37 684-688. DOI: 10.1249/01.MSS.0000159138.48107.7D 13. 13.Foweather L, Knowles Z, Ridgers ND, O’Dwyer MV, Foulkes JD, Stratton G. Fundamental movement skills in relation to weekday and weekend physical activity in preschool children. J Sci Med Sport. 2015;18(6):691–6. 14. 14.Kwon S, O'Neill M. Socioeconomic and familial factors associated with gross motor skills among US children aged 3–5 Years: The 2012 NHANES National Youth Fitness Survey. Int J Environ Res Public Health. 2020;17(12). 15. Wijtzes AI, Kooijman MN, Kiefte-de Jong JC, de Vries SI, Henrichs J, Jansen W, Jaddoe VW, Hofman A, Moll HA, Raat H. Correlates of physical activity in 2-year-old toddlers: the generation R study. J Pediatr (2013.0) 163 791-799. DOI: 10.1016/j.jpeds.2013.02.029 16. Golan M. Parents as agents of change in childhood obesity–from research to practice. Int J Pediatr Obes (2006.0) 1 66-76. DOI: 10.1080/17477160600644272 17. 17.Hnatiuk J, Salmon J, Campbell KJ, Ridgers ND, Hesketh KD. Early childhood predictors of toddlers’ physical activity: longitudinal findings from the Melbourne InFANT Program. Int J Behav Nutr Phys Act. 2013;10:123. 18. 18.Hnatiuk JA, Ridgers ND, Salmon J, Hesketh KD. Maternal correlates of young children’s physical activity across periods of the day. J Sci Med Sport. 2017;20(2):178–83. 19. 19.Patrick H, Hennessy E, McSpadden K, Oh A. Parenting styles and practices in children’s obesogenic behaviors: scientific gaps and future research directions. Child Obes. 2013;9(Suppl):S73-86. 20. Xu H, Wen LM, Rissel C. Associations of parental influences with physical activity and screen time among young children: a systematic review. J Obes (2015.0) 2015 546925. DOI: 10.1155/2015/546925 21. Solomon-Moore E, Toumpakari Z, Sebire SJ, Thompson JL, Lawlor DA, Jago R. Roles of mothers and fathers in supporting child physical activity: a cross-sectional mixed-methods study. BMJ Open (2018.0) 8 e019732. DOI: 10.1136/bmjopen-2017-019732 22. Welk G, Wood L, Moss G. Parental influence on physical activity in children: am exploration of potential mechanisms. Ped Exerc Sci (2009.0) 15 19-33. DOI: 10.1123/pes.15.1.19 23. Telford RM, Telford RD, Olive LS, Cochrane T, Davey R. Why are girls less physically active than boys? Findings from the LOOK longitudinal study. PLoS ONE (2016.0) 11 e0150041. DOI: 10.1371/journal.pone.0150041 24. Trost SG, Sallis JF, Pate RR, Freedson PS, Taylor WC, Dowda M. Evaluating a model of parental influence on youth physical activity. Am J Prev Med (2003.0) 25 277-282. DOI: 10.1016/S0749-3797(03)00217-4 25. Crumbley CA, Ledoux TA, Johnston CA. Physical activity during early childhood: the importance of parental modeling. Am J Lifestyle Med (2020.0) 14 32-35. DOI: 10.1177/1559827619880513 26. Lakes KD, Neville RD, Abdullah M, Donnelly J. Psychological determinants of physical activity and development in early childhood among children with developmental delays: the role of parent beliefs regarding the benefits of physical activity. Front Sports Active Living (2020.0) 2 104. DOI: 10.3389/fspor.2020.00104 27. Tate EB, Wood W, Liao Y, Dunton GF. Do stressed mothers have heavier children? A meta-analysis on the relationship between maternal stress and child body mass index. Obes Rev (2015.0) 16 351-361. DOI: 10.1111/obr.12262 28. 28.Albert MV, Sugianto A, Nickele K, Zavos P, Sindu P, Ali M, Kwon S. Hidden Markov model-based activity recognition for toddlers. Physiol Meas. 2020;41(2):025003. 29. 29.QuickFacts: Chicago City, Illinois [https://www.census.gov/quickfacts/chicagocityillinois]. 30. 30.Maternal, Infant, and Early Childhood Home Visiting (MIECHV) Program [https://mchb.hrsa.gov/programs-impact/programs/home-visiting/maternal-infant-early-childhood-home-visiting-miechv-program]. 31. 31.Ahmadi MN, Pavey TG, Trost SG. Machine learning models for classifying physical activity in free-living preschool children. Sensors (Basel). 2020;20(16). 32. Trost SG, Cliff DP, Ahmadi MN, Tuc NV, Hagenbuchner M. Sensor-enabled activity class recognition in preschoolers: hip versus wrist data. Med Sci Sports Exerc (2018.0) 50 634-641. DOI: 10.1249/MSS.0000000000001460 33. 33.ActiLife Software [https://actigraphcorp.com/support/software/actilife/]. 34. Squires J, Bricker D. Ages & Stages Questionnaires®, Third Edition (ASQ®-3): A Parent-Completed Child Monitoring System (2009.0) 35. 35.O’Connor TM, Cerin E, Hughes SO, Robles J, Thompson DI, Mendoza JA, Baranowski T, Lee RE. Psychometrics of the preschooler physical activity parenting practices instrument among a Latino sample. Int J Behav Nutr Phys Act. 2014;11:3. 36. 36.Centers for Disease Control and Prevention. 2021-2022 NHANES Physical Activity and Physical Fitness - PAQ. https://wwwn.cdc.gov/nchs/data/nhanes/2021-2022/questionnaires/PAQ-L-508.pdf. 37. 37.National Cancer Institute. FLASHE - Annotated Parent Physical Activity Survey. https://cancercontrol.cancer.gov/sites/default/files/2020-06/parent_pa_puf_instrument.pdf. 38. Berry JO, Jones WH. The parental stress scale: initial psychometric evidence. J Soc Pers Relationships (1995.0) 12 463-472. DOI: 10.1177/0265407595123009 39. Crnic KA, Greenberg MT. Minor parenting stresses with young children. Child Dev (1990.0) 61 1628-1637. DOI: 10.2307/1130770 40. Goodman R. The strengths and difficulties questionnaire: a research note. J Child Psychol Psychiatry (1997.0) 38 581-586. DOI: 10.1111/j.1469-7610.1997.tb01545.x 41. Patel J, Smith R, O’Farrelly C, Iles J, Rosan C, Ryan R, Ramchandani P. Assessing behavior in children aged 12–24 months using the strengths and difficulties questionnaire. Infancy (2021.0) 26 724-734. DOI: 10.1111/infa.12425 42. 42.Kwon S, O'Neill ME, Foster CC. The Associations of Child's Clinical Conditions and Behavioral Problems with Parenting Stress among Families of Preschool-Aged Children: 2018-2019 National Survey of Child Health. Children (Basel). 2022;9(2):241. 10.3390/children9020241. 43. 43.Harris PA, Taylor R, Minor BL, Elliott V, Fernandez M, O’Neal L, McLeod L, Delacqua G, Delacqua F, Kirby J, et al. The REDCap consortium: Building an international community of software platform partners. J Biomed Informat. 2019;95: 103208. 44. Harris PA, Taylor R, Thielke R, Payne J, Gonzalez N, Conde JG. Research electronic data capture (REDCap)—a metadata-driven methodology and workflow process for providing translational research informatics support. J Biomed Informat (2009.0) 42 377-381. DOI: 10.1016/j.jbi.2008.08.010 45. Vähä-Ypyä H, Husu P, Suni J, Vasankari T, Sievänen H. Reliable recognition of lying, sitting, and standing with a hip-worn accelerometer. Scand J Med Sci Sports (2018.0) 28 1092-1102. DOI: 10.1111/sms.13017 46. Friard O, Gamba M. BORIS: a free, versatile open-source event-logging software for video/audio coding and live observations. Methods Ecol Evol (2016.0) 7 1325-1330. DOI: 10.1111/2041-210X.12584 47. Ahmadi MN, Trost SG. Device-based measurement of physical activity in pre-schoolers: comparison of machine learning and cut point methods. PLoS ONE (2022.0) 17 e0266970. DOI: 10.1371/journal.pone.0266970 48. Ahmadi MN, Brookes D, Chowdhury A, Pavey T, Trost SG. Free-living evaluation of laboratory-based activity classifiers in preschoolers. Med Sci Sports Exerc (2020.0) 52 1227-1234. DOI: 10.1249/MSS.0000000000002221 49. Hnatiuk J, Ridgers ND, Salmon J, Campbell K, McCallum Z, Hesketh K. Physical activity levels and patterns of 19-month-old children. Med Sci Sports Exerc (2012.0) 44 1715-1720. DOI: 10.1249/MSS.0b013e31825825c4 50. Kuzik N, Clark D, Ogden N, Harber V, Carson V. Physical activity and sedentary behaviour of toddlers and preschoolers in child care centres in Alberta Canada. Can J Public Health (2015.0) 106 e178-183. DOI: 10.17269/cjph.106.4794 51. Lee EY, Hesketh KD, Hunter S, Kuzik N, Rhodes RE, Rinaldi CM, Spence JC, Carson V. Meeting new Canadian 24-hour movement guidelines for the early years and associations with adiposity among toddlers living in Edmonton, Canada. BMC Public Health (2017.0) 17 840. DOI: 10.1186/s12889-017-4855-x 52. Tracy JD, Donnelly T, Sommer EC, Heerman WJ, Barkin SL, Buchowski MS. Identifying bedrest using waist-worn triaxial accelerometers in preschool children. PLoS ONE (2021.0) 16 e0246055. DOI: 10.1371/journal.pone.0246055 53. Ahmadi MN, Pfeiffer KA, Trost SG. Physical activity classification in youth using raw accelerometer data from the hip. Meas Phys Educ Exerc Sci (2020.0) 24 129-136. DOI: 10.1080/1091367X.2020.1716768 54. Banos O, Galvez JM, Damas M, Pomares H, Rojas I. Window size impact in human activity recognition. Sensors (Basel) (2014.0) 14 6474-6499. DOI: 10.3390/s140406474 55. Ellis K, Kerr J, Godbole S, Staudenmayer J, Lanckriet G. Hip and wrist accelerometer algorithms for free-living behavior classification. Med Sci Sports Exerc (2016.0) 48 933-940. DOI: 10.1249/MSS.0000000000000840 56. Curran PJ, Obeidat K, Losardo D. Twelve frequently asked questions about growth curve modeling. J Cogn Dev (2010.0) 11 121-136. DOI: 10.1080/15248371003699969 57. Farrahi V, Niemelä M, Kangas M, Korpelainen R, Jämsä T. Calibration and validation of accelerometer-based activity monitors: A systematic review of machine-learning approaches. Gait Posture (2019.0) 68 285-299. DOI: 10.1016/j.gaitpost.2018.12.003 58. 58.Narayanan A, Desai F, Stewart T, Duncan S, Mackay L. Application of Raw Accelerometer Data and Machine-Learning Techniques to Characterize Human Movement Behavior: A Systematic Scoping Review. J Phys Act Health. 2020;17(3):360–83. 10.1123/jpah.2019-0088. 59. Gyllensten IC, Bonomi AG. Identifying types of physical activity with a single accelerometer: evaluating laboratory-trained algorithms in daily life. IEEE Trans Biomed Eng (2011.0) 58 2656-2663. DOI: 10.1109/TBME.2011.2160723 60. Sasaki JE, Hickey AM, Staudenmayer JW, John D, Kent JA, Freedson PS. Performance of activity classification algorithms in free-living older adults. Med Sci Sports Exerc (2016.0) 48 941-950. DOI: 10.1249/MSS.0000000000000844 61. Bastian T, Maire A, Dugas J, Ataya A, Villars C, Gris F, Perrin E, Caritu Y, Doron M, Blanc S. Automatic identification of physical activity types and sedentary behaviors from triaxial accelerometer: laboratory-based calibrations are not enough. J Appl Physiol (1985) (2015.0) 118 716-722. DOI: 10.1152/japplphysiol.01189.2013 62. van Kuppevelt D, Heywood J, Hamer M, Sabia S, Fitzsimons E, van Hees V. Segmenting accelerometer data from daily life with unsupervised machine learning. PLoS ONE (2019.0) 14 e0208692. DOI: 10.1371/journal.pone.0208692 63. 63.Hedeker D, Gibbons RD. Longitudinal data analysis. Hoboken: Wiley-Interscience; 2006. p. 279. 64. 64.Albright CL, Saiki K, Steffen AD, Woekel E. What barriers thwart postpartum women’s physical activity goals during a 12-month intervention? A process evaluation of the Nā Mikimiki Project. Women Health. 2015;55(1):1–21. 65. Brown PR, Brown WJ, Miller YD, Hansen V. Perceived constraints and social support for active leisure among mothers with young children. Leisure Sci (2001.0) 23 131-144. DOI: 10.1080/014904001316896837 66. Evenson KR, Aytur SA, Borodulin K. Physical activity beliefs, barriers, and enablers among postpartum women. J Womens Health (Larchmt) (2009.0) 18 1925-1934. DOI: 10.1089/jwh.2008.1309 67. Bartley SJ, Blanton PW, Gilliard JL. Husbands and wives in dual-earner marriages: decision-making, gender role attitudes, division of household labor, and equity. Marriage Family Rev (2005.0) 37 69-94. DOI: 10.1300/J002v37n04_05 68. Hamilton K, White KM. Parental physical activity: exploring the role of social support. Am J Health Behav (2010.0) 34 573-584. DOI: 10.5993/AJHB.34.5.7 69. 69.Tremblay RE. Developmental origins of disruptive behaviour problems: the “original sin” hypothesis, epigenetics and their consequences for prevention. J Child Psychol Psychiat. 2010;51(4):341–67.
--- title: 'Coronary CT angiography for preoperative evaluation of non-cardiac surgery in patients with thoracic tumors: preliminary exploratory analysis in a retrospective cohort' authors: - Meng Liao - Mingyue Tang - Xu Cao - Gao Liang - Mingguo Xie - Peng Zhou journal: Journal of Cardiothoracic Surgery year: 2023 pmcid: PMC10026420 doi: 10.1186/s13019-022-02096-y license: CC BY 4.0 --- # Coronary CT angiography for preoperative evaluation of non-cardiac surgery in patients with thoracic tumors: preliminary exploratory analysis in a retrospective cohort ## Abstract ### Purpose Noninvasive coronary CT angiography (CCTA) was used to retrospectively analyze the characteristics of coronary artery disease (CAD) in patients with thoracic tumors and the impact of the results on clinical surgery decision-making, thus increasing the understanding of perioperative cardiac risk evaluation. ### Method A total of 779 patients (age 68.6 ± 6.6 years) with thoracic tumor (lung, esophageal, and mediastinal tumor) scheduled for non-cardiac surgery were retrospectively enrolled. Patients were divided into two groups: accepted or canceled surgery. Clinical data and CCTA results were compared between the two groups, and multivariate logistic regression analysis was performed to determine predictors of the events of cancellations of scheduled surgeries. ### Results 634 patients ($81.4\%$) had non-significant CAD and 145 patients ($18.6\%$) had significant CAD. Single‑, 2‑, and 3‑ vessel disease was found in 173 ($22.2\%$), 93 ($11.9\%$) and 50 ($6.4\%$) patients, respectively. 500 ($64.2\%$), 96 ($12.3\%$), 96 ($12.3\%$), 56 ($7.2\%$) and 31 ($4.0\%$) patients were rated as CACS 0, 1–99, 100–399, 400–999 and > 1000, respectively. Cancellations of scheduled procedures continue to increase based on the severity of the stenosis and the number of major coronary artery stenosis. The degree of stenosis and the number of vascular stenosis were independent predictors of cancelling scheduled surgery. ### Conclusions For patients with thoracic tumors scheduled for non-cardiac surgery, the results suggested by CCTA significantly influenced surgery planning and facilitated to reduce perioperative cardiovascular events. ## Introduction Surgery is one of the most common treatments for patients with thoracic tumors, such as lung, esophageal and mediastinal tumors. These patients may have occult CAD with inconspicuous and atypical clinical symptoms, thus has not been diagnosed. Major adverse cardiac events (MACE) that occur after non-cardiac surgery are usually associated with prior CAD. Risk assessment of perioperative cardiovascular events is important for clinical surgical planning. CCTA is a noninvasive examination with high sensitivity and specificity in the detection or exclusion of CAD [1]. Although current guidelines do not include CCTA as a routine preoperative examination for patients undergoing non-cardiac surgery [2], CCTA is increasingly being used for preoperative screening as a reliable method of diagnosing CAD in clinical practice. Previous research has shown that the severity and extent of CAD in CCTA in non-cardiac surgery patients are associated with perioperative MACE [3]. However, in clinical practice, little attention has been paid to whether coronary stenosis and calcification on CCTA has an impact on surgical planning. Therefore, we retrospectively analyzed the characteristics of CAD in patients with thoracic tumors scheduled for non-cardiac surgery and the impact of CCTA results on scheduled surgery to increase our understanding of perioperative management. ## Study population The study was approved by a local institutional review committee. Due to the retrospective design of this study, all subjects waived informed consent. From January 2015 to June 2019, we enrolled a total of 795 patients with non-cardiovascular thoracic tumor surgery who underwent preoperative CCTA for screening of CAD. Exclusion criteria in this study were: left ventricular ejection fraction < $40\%$, renal insufficiency (glomerular filtration rate < 30 ml/min/1.7 m2), severe heart failure, severe arrhythmia, iodine contrast agent allergy, and substandard image quality for imaging analysis. Finally, we recruited 779 patients (Fig. 1) with thoracic tumor, among them, there were 289 cases of lung tumor, 470 cases of esophagus tumor and 20 cases of mediastinal tumor. The preoperative complications of cardiovascular diseases included 248 patients with hypertension, 234 patients with hyperlipidemia, 50 patients with confirmed CAD, and 92 patients with the positive electrocardiogram (ECG).Fig. 1Flow diagram of the study patients ## Scan protocols CCTA of all patients was performed on a Philips Brilliance 256-layer scanner using a retrospective electrocardiogram gated mode. Scanning scope ranged from tracheal carina to 2 cm below the apex of the heart. First, plain CT scan was used for quantitative measurement of coronary artery calcification scores. Imaging parameters: prospective gated, triggered at $75\%$ R-R interval, tube voltage 120 kV, tube current 550 mAs, slice thickness 2.5 mm, reconstruction interval 2.5 mm and rotation time 0.27 s. A 20-gauge needle with double tube high-pressure Syringe (Boluspro, Philips Healthcare, Cleveland, Ohio, USA) Contrast agent (Ultravist 370, Bayer Healthcare, Berlin, Germany) was injected through the cubital vein 60–80 mL at a flow rate of 5.0 mL/s. After injection, 30 mL of normal saline was injected at the same rate. The ascending aorta was set as the region of interest, and the trigger threshold was set at 100–120HU. After reaching the threshold, the patient was asked to hold his breath, and the scan was automatically triggered after 6 s. Scanning parameters: tube voltage 100 kV, tube current 400–500 mAs, detector collimator 128 × 0.625 mm, rotation time 0.27 s, pitch 0.18, standard reconstruction(iDose4, level 5), reconstructed slice thickness 0.9 mm, reconstructed interval 0.45 mm. ## Data analysis Two experienced radiologists independently reviewed each CT scan on a dedicated workstation (Extended Brilliance Workspace Version 4.0; Philips Healthcare). If no consensus can be reached, a third expert is consulted to make the final diagnosis. Image post-processing methods used to evaluate coronary stenosis and calcification include maximum density projection, multiplane reconstruction, curved surface reconstruction, and volume reconstruction. 1-, 2-, or 3-vessel disease was defined according to the number of epicardial arterial stenosis. In patients with multi-vessel disease, the most severe coronary artery stenosis was considered the study subject. The degree of coronary artery stenosis was classified as normal appearing, mild (< $50\%$, Fig. 2), moderate ($50\%$-$75\%$, Fig. 3), and severe (≥ $75\%$, Fig. 4) stenosis. Among them, normal appearing and mild stenosis were non-obstructive stenosis, and moderate and severe stenosis were obstructive stenosis. Coronary artery calcification score (CACS) was obtained by smartscore software and divided into 5 groups, namely 0, 1–99, 100–399, 400–999 and > 1000.Fig. 2An example of mild stenosis in 65‑year-old asymptomatic man with esophageal cancer. Localized calcified plaque (arrow) with mild stenosis in the proximal segment of right coronary artery (RCA). Abbreviations: RCA, right coronary arteryFig. 3An example of moderate stenosis in 74‑year-old asymptomatic man, non‑calcified plaque (arrow) with moderate stenosis in the proximal segment of left anterior descending coronary artery (LAD). Abbreviations: LAD, left anterior descending coronary arteryFig. 4An example of severe stenosis in 54‑year-old man with chest pain and positive ECG analysis. Mixed plaque (arrow) with severe stenosis in the middle segment of left anterior descending coronary artery (LAD). Localized calcified plaque in the left circumflex coronary artery (LCX) with no luminal stenosis. The right coronary artery (RCA) is normal appearing. Abbreviations: LAD, left anterior descending coronary artery; LCX, left circumflex coronary artery; RCA, right coronary artery Electronic medical records were reviewed retrospectively to analyze patients' clinical data, including coronary risk factors such as hypertension, diabetes, hyperlipidemia, smoking, stroke, and related clinical decisions (Table 1). In preoperative routine ECG examination, ST segment analysis is considered positive if ST segment horizontal or down-sloping depression ≥ 1 mm occurs in 2 or more consecutive leads. The impact of CCTA results on clinical decisions was determined after a multidisciplinary consultation, that is, whether surgery was delayed or canceled due to severe CAD.Table 1Patient characteristicsAccept surgeryn = 484Cancel surgeryn = 295p valueAge68.7 ± 6.468.4 ± 6.80.542Male/Female$\frac{368}{116230}$/650.536BMI22.8 ± 3.122.3 ± 3.20.032HR (beats/min)73.2 ± 31.174.3 ± 14.70.585ECG Positive/Negative$\frac{46}{43446}$/2470.011LEVF66.8 ± 7.065.9 ± 8.10.120Smoking/Non-smoking$\frac{229}{255158}$/1370.091Hypertension/ Non-hypertension$\frac{157}{32891}$/2030.680Hyperlipidemia/Non-hyperlipidemia$\frac{148}{33686}$/2090.674Stroke/Non-stroke$\frac{12}{47213}$/2820.139Diabetes/Non-Diabetes$\frac{38}{44629}$/2650.357Abbreviations: BMI, Body Mass Index; HR, heart rate; ECG, electrocardiogram; LECF, Left ventricular ejection fraction ## Statistical analysis SPSS Version 23.0 and Graph Pad Prism Version 6.0 were used for statistical analysis. Continuous data are expressed as mean ± standard deviation, while nominal variables are expressed as frequency and percentage. The t test was used for measurement data, and the chi-square test or Fisher's exact test was used for counting data. Univariate and multivariate logistic regression analyses were performed to assess which parameters were independently associated with surgical decision making in patients with thoracic tumor. A p value < 0.1 in the univariate analyses were introduced to further multivariate analysis. A double-tailed $p \leq 0.05$ was considered statistically significant. ## Patient Characteristics Cardiac CT scanning was successfully performed in all 779 patients, whose age ranged from 41 to 89 years (mean 68.6 ± 6.6 years). Baseline characteristics of subjects in our study are shown in Table 1. No ECG was performed in 6 patients. In 15 patients BMI data was lacking and in 1 patient diabetes was unknown. Among the 779 patients, 145 patients with significant coronary artery stenosis, 12 underwent invasive coronary angiography (ICA) and 1 underwent coronary intervention after coronary CTA. These patients underwent coronary artery treatment followed by non-cardiac surgery. ## The effect of coronary artery CTA on the planning of non-cardiac surgery Among CAD patients, 55, 28, and 48 patients with mild, moderate, and severe cases gave up surgery, respectively. Table 2 shows the coronary categories as determined by CT. In total, 634 ($81.4\%$) patients had non-significant CAD and 145 ($18.6\%$) patients had significant CAD. Of the patients with non-significant CAD, 463 ($59.4\%$) patients were normal and 171 ($22.0\%$) patients showed mild stenosis. Of the patients with significant CAD, 71 ($9.1\%$) patients had moderate stenosis and 74 ($9.5\%$) patients had severe stenosis. In addition, for stenosis of CAD, 1-, 2-, and 3- vessel disease was found in 173 ($22.2\%$), 93 ($11.9\%$), and 50 ($6.4\%$) patients, respectively (Table 2); and 143 ($18.4\%$) patients showed multi-vessel disease (≥ 2 branches). In addition, scheduled surgery was cancelled in 19 ($11.0\%$), 28 ($30.1\%$) and 19 ($38.0\%$) patients with 1-, 2-, and 3- vessel disease, respectively. In the grading of CACS, 500 ($64.2\%$), 96 ($12.3\%$), 96 ($12.3\%$), 56 ($7.2\%$) and 31 ($4.0\%$) patients were rated as 0, 1–99, 100–399, 400–999 and > 1000, respectively (Table 3). Univariate and multivariate logistic regression analysis showed that, the number and degree of vascular stenosis were independently correlated with the cancelation of surgery (Table 4).Table 2Degree of coronary artery stenosis and the events of abandoned surgery for the reason of cardiac riskFrequency (n, %)Event (n, %)p valueNon-significant stenosis634 < 0.001 Normal appearing463 ($59.4\%$)0 [0] Mild stenosis171 ($22.0\%$)0 [0]Significant stenosis145 Moderate stenosis71 ($9.1\%$)24 ($33.8\%$) Severe stenosis74 ($9.5\%$)42 ($56.8\%$)Number of major epicardial coronary artery stenosis Normal appearing463 ($59.4\%$)0[0]1‑vessel disease173 ($22.2\%$)19 ($11.0\%$)0.001* 2‑vessel disease93 ($11.9\%$)28 ($30.1\%$) 3‑vessel disease50 ($6.4\%$)19 ($38.0\%$) Multi-vessel disease143 ($18.4\%$)47 ($32.9\%$) < 0.001*Abbreviations: CAD coronary artery diseaseEvent compared among 1, 2 and 3-vessel diseaseEvent compared between 1-vessel diseaseTable 3Coronary artery calcification score and the events of abandoned surgery for the reason of cardiac riskFrequency (n, %)Event (n, %)p value0500 ($64.2\%$)0 [0] < 0.0011–9996 ($12.3\%$)0 [0]100–39996 ($12.3\%$)27 ($28.1\%$)400–99956 ($7.2\%$)24 ($42.9\%$) > 100031 ($4.0\%$)15 ($48.4\%$)Table 4Univariate and multifactorial logistic regression analysis of the events of abandoned surgeryVariablesUnivariate analysisOR ($95\%$ CI)p valueMultivariate analysisOR ($95\%$ CI)p valueBMI1.053 (1.004–1.103)0.0321.047 (0.996–1.100)0.070ECG0.569 (0.367–0.881)0.0120.777 (0.487–1.242)0.292HR0.998 (0.993–1.004)0.594LEVF1.016 (0.996–1.037)0.124Number of vascular stenosis0.830 (0.713–0.967)0.0171.365 (1.001–1.863)0.049Degree of stenosis0.749 (0.647–0.866) < 0.0010.671 (0.504–0.894)0.006CACS0.999 (0.999–1.000)0.0011.000 (0.999–1.000)0.082Smoking0.773 (0.578–1.035)0.0840.833 (0.612–1.133)0.244Hypertension1.076 (0.788–1.470)0.644Hyperlipidemia1.070 (0.780–1.470)0.674Diabetes0.779 (0.469–1.292)0.333Stroke0.551 (0.248–1.224)0.143Abbreviations: BMI, Body Mass Index; ECG, electrocardiogram; HR, heart rate; LECF, Left ventricular ejection fraction; CACS, Coronary artery calcification score According to the electronic hospitalization records, during the postoperative hospitalization, 1 patient (severe stenosis; 2-vessel disease) had non-fatal myocardial infarction, 2 patients died of cardiac shock, and the rest had no MACE records. ## Discussion The main findings of this study were that the number and degree of vascular stenosis suggested by preoperative CCTA in patients with thoracic tumor was independently associated with the decision to cancel surgery; surgery cancellations increased as the number or extent of the stenosis rise. Cardiovascular disease is also the leading cause of death for tumor patients, there are common risk factors between them two. Vascular endothelial damage or arterial thrombosis caused by anti-tumor treatment may increase the risk of cardiovascular disease [4]. Tumor patients at higher risk of CAD, while clinical manifestations are atypical, for example, chest pain and dyspnea less seen in tumor patients [5]. Therefore, tumor patients may have a potential risk of CAD, which needs to be paid great attention. CAD can affect or limit tumor treatment. Surgery is a common treatment method for patients with lung tumor, esophageal tumor, or mediastinal tumor, which has certain requirements for the circulatory function of tumor patients. Moreover, type of surgery is related to the cardiac risk. As a high-risk operation, the MACE risk of thoracic surgery is ≥ $5\%$ [6]. Surgical stress leads to inflammation and hypercoagulability, triggering plaque instability or rupture, and subsequent thrombosis, which accounts for $50\%$ of perioperative acute coronary events [7, 8]. Therefore, CAD will limit the feasibility of surgery. Perioperative MACE was defined as non-fatal stroke, non-fatal myocardial infarction, congestive heart failure, and cardiogenic death that occurred within 30 days after surgery. Worldwide, more than 300 million patients undergo non-cardiac surgery each year [9]. Cardiovascular complications are one of the major causes of MACE in patients undergoing non-cardiac surgery. As the tumor patients with cardiovascular disease undergoing non-cardiac surgery continue to increase, the incidence of perioperative MACE is also increased, which seriously affects the safety of surgery and the management of postoperative complications. Therefore, preoperative risk assessment of the cardiovascular event is of great significance for tumor patients. CCTA is a non-invasive examination for the assessment of CAD. It can clearly display the type and composition of plaque and accurately evaluate the extent and degree of coronary artery stenosis [10, 11]. In contrast to invasive coronary angiography (ICA), CCTA can show plaques remodeling outward without lumen narrowing [12]. With relatively high sensitivity and specificity, low cost and low radiation, CCTA has become the preferred method of noninvasive examination for diagnosis of CAD [10]. The addition of an appropriate CCTA to enhanced CT in patients with thoracic tumors does not significantly increase radiation exposure or contrast material administration, and providing a practical improvement in cardiovascular risk stratification in these patients [13]. Patients with severe stenosis can be improved by revascularization, while patients with mild or moderate stenosis can be treated with medication [14]. Moreover, patients who were first assessed as inoperable by CCTA may regain the opportunity of surgery after the relevant treatment. Considering the increased heart disease progression or surgical risk, scheduled surgery of tumor patients with CAD may be delayed or cancelled. Therefore, CCTA examination before developing a treatment plan can indicate whether surgery can be performed as scheduled or should be postponed after CAD intervention or abandoned. Even though the current guidelines, CCTA has not been incorporated into the preoperative routine examination, but CCTA as a noninvasive method can be encouraged to performed on tumor patients preoperatively, if the results have a potential influence on the management of patients [2, 15, 16]. The appropriate indication for coronary CTA as part of preoperative evaluation is not specified in current European Society of Cardiology or American College of Cardiology/American Heart Association guidelines, mainly due to insufficient data on coronary CTA in preoperative risk stratification [1], which should be investigated in future research efforts. Exercise ECG test, stress echocardiography and stress myocardial perfusion imaging are recommended for the screening of CAD [17], but stress test is not suitable for patients with poor general conditions or with contraindications. For non-cardiovascular surgery patients, ICA is not routinely recommended for risk stratification, but ICA and revascularization are recommended before high-risk surgery or accompanied with severe stress ischemia. In this study, $18.6\%$ of patients were assessed as significant CAD by CCTA, but most patients did not undergo ICA. A study [18] suggested that coronary CTA and ICA are equally effective in assessing long-term risk in patients with non-ST-elevation acute coronary syndrome. Although severe calcification may affect the judgment of the degree of luminal stenosis [19], in the clinical practice of non-invasive screening of CAD before surgery for tumor patients, it is more concerned about how to screen out patients who are not suitable for surgery, rather than over-diagnosis. In our study, with the increase of degree of stenosis, patients gave up surgery has increased because of CCTA results. Some patients with mild coronary artery stenosis experienced plaque rupture leading to fatal cardiovascular events [20], thus we included patients with all grades of stenosis, not just significant stenosis. Patients with multi-vessel disease were more likely to forgo surgery for cardiovascular reasons than patients with single-vessel disease. A previous study suggested that the more extensive coronary artery calcification was associated with a higher incidence of coronary artery events, which was inconclusive [21]. But recent research suggests that calcification can predict risk of cardiovascular events and death [22–24]. A study [22] of 25,253 asymptomatic patients with long-term follow-up concluded that CACS was an independent predictor of all-cause mortality, the mortality risks of 11–100, 101–299, 300–399, 400–699, 700–999, and > 1000 scores with CACS were approximately 2.2, 4.5, 6.4, 9.2, 10.4, and 12.5 times of those with CACS 0, respectively. Coronary artery calcium scans are recommended as a class IIa in the 2019 ACC/AHA guidelines for people at intermediate risk [25]. People with CACS of zero had a lower incidence of CACS progression or risk of coronary artery disease during the 5-year warranty period [26, 27]. In our study, the number of non-stenotic and non-calcified patients was not equal, possibly because some patients only had non-calcified plaques. The probability of abandoning surgery by CCTA results was significantly different among groups with different calcification scores. However, in multivariate logistic regression analysis, CACS cannot be considered as an independent factor influencing surgical decision making. A meta-analysis [3] showed that the risk of perioperative MACE was strongly correlated with the extent and severity of coronary artery stenosis indicated by CCTA, with a greater risk of obstructive stenosis and multi-vessel disease; there was also a certain correlation between CACS and the incidence of MACE during the perioperative period (CACS ≥ 100, ≥ 400, ≥ 1000 were compared with CACS < 100, < 400, < 1000 respectively). In this study, of 484 patients who underwent surgery, only 1 had perioperative MACE (non-fatal myocardial infarction) during hospitalization, with an incidence of $0.21\%$, significantly lower than reported [28, 29]. It can be said that CCTA evaluation can effectively reduce the incidence of cardiovascular events. In addition to providing coronary artery stenosis and plaque information, CCTA can also obtain a series of hemodynamic indicators by combining advanced computational fluid dynamics methods, which were not analyzed in this paper. As a noninvasive and effective visualization tool, CCTA can provide preliminary coronary risk information and reduce cardiovascular complications by excluding some patients who are not suitable for surgery. Meanwhile, it may also exclude some patients who require surgery. The benefits to patients need to be further studied. ## Study limitations The limitations of this study are as follows. First, due to retrospective study, we could only include perioperative MACE during hospitalization. We cannot fully assess the outcome of the patients, and CCTA cannot currently be recommended as a routine test for preoperative cardiac risk stratification in patients undergoing noncardiac surgery. The second, this is a single-center retrospective study, which may have a certain center-specific bias, and a larger cohort multi-center study should be conducted in the future to investigate the association between CCTA and cardiac risk. The third, CACS for preoperative cardiac risk assessment needs further study. ## Conclusion For patients with thoracic tumors scheduled for non-cardiac surgery, preoperative CCTA characterizes coronary artery stenosis and calcification to facilitate detecting CAD and risk stratification, thereby influencing clinical surgery decisions. ## References 1. 1.Hwang JW, Kim EK, Yang JH, Chang SA, Song YB, Hahn JY, Choi SH, Gwon HC, Lee SH, Kim SM, et al. Assessment of perioperative cardiac risk of patients undergoing noncardiac surgery using coronary computed tomographic angiography. Circul Cardiovas Imaging. 2015;8(3). 2. Arnett DK, Blumenthal RS, Albert MA, Buroker AB, Goldberger ZD, Hahn EJ, Himmelfarb CD, Khera A, Lloyd-Jones D, McEvoy JW. 2019 ACC/AHA guideline on the primary prevention of cardiovascular disease: executive summary: a report of the American College of Cardiology/American Heart Association Task Force on Clinical Practice Guidelines*. Circulation* (2019.0) 140 e563-e595. PMID: 30879339 3. Koshy AN, Ha FJ, Gow PJ, Han HC, Amirul-Islam FM, Lim HS, Teh AW, Farouque O. Computed tomographic coronary angiography in risk stratification prior to non-cardiac surgery: a systematic review and meta-analysis. Heart (British Cardiac Society) (2019.0) 105 1335-1342. PMID: 31018953 4. Nohria A, Groarke JD. Management of acute coronary syndromes in patients with cancer: room for improvement. Eur Heart J (2019.0) 40 1801-1803. DOI: 10.1093/eurheartj/ehz013 5. Guha A, Dey AK, Jneid H, Addison D. Acute coronary syndromes in cancer patients. Eur Heart J (2019.0) 40 1487-1490. DOI: 10.1093/eurheartj/ehz267 6. Smilowitz NR, Berger JS. Perioperative cardiovascular risk assessment and management for noncardiac surgery: a review. JAMA (2020.0) 324 279-290. DOI: 10.1001/jama.2020.7840 7. Ghadri JR, Fiechter M, Veraguth K, Gebhard C, Pazhenkottil AP, Fuchs TA, Templin C, Gaemperli O, Kaufmann PA. Coronary calcium score as an adjunct to nuclear myocardial perfusion imaging for risk stratification before noncardiac surgery. J Nucl Med: Off Publ Soci Nuclear Med (2012.0) 53 1081-1086. DOI: 10.2967/jnumed.111.100206 8. Smilowitz NR, Berger JS. Perioperative management to reduce cardiovascular events. Circulation (2016.0) 133 1125-1130. DOI: 10.1161/CIRCULATIONAHA.115.017787 9. Smilowitz NR, Gupta N, Ramakrishna H, Guo Y, Berger JS, Bangalore S. Perioperative major adverse cardiovascular and cerebrovascular events associated with noncardiac surgery. JAMA Cardiol (2017.0) 2 181-187. DOI: 10.1001/jamacardio.2016.4792 10. Joshi H, Shah R, Prajapati J, Bhangdiya V, Shah J, Kandre Y, Shah K. Diagnostic accuracy of computed tomography angiography as compared to conventional angiography in patients undergoing noncoronary cardiac surgery. Heart Views: Off J Gulf Heart Assoc (2016.0) 17 88-91. DOI: 10.4103/1995-705X.192555 11. Nicol ED, Norgaard BL, Blanke P, Ahmadi A, Weir-McCall J, Horvat PM, Han K, Bax JJ, Leipsic J. The future of cardiovascular computed tomography: advanced analytics and clinical insights. JACC Cardiovasc Imaging (2019.0) 12 1058-1072. DOI: 10.1016/j.jcmg.2018.11.037 12. Pflederer T, Marwan M, Schepis T, Ropers D, Seltmann M, Muschiol G, Daniel WG, Achenbach S. Characterization of culprit lesions in acute coronary syndromes using coronary dual-source CT angiography. Atherosclerosis (2010.0) 211 437-444. DOI: 10.1016/j.atherosclerosis.2010.02.001 13. Bossone E, Cademartiri F, AlSergani H, Chianese S, Mehta R, Capone V, Ruotolo C, Tarrar IH, Frangiosa A, Vriz O. Preoperative assessment and management of cardiovascular risk in patients undergoing non-cardiac surgery: implementing a systematic stepwise approach during the covid-19 pandemic era. J Cardiovasc Develop Disease (2021.0) 8 126. DOI: 10.3390/jcdd8100126 14. van Rosendael AR, Bax JJ. Improved risk stratification with computed tomographic coronary angiography in patients with suspected coronary artery disease. Eur Heart J Cardiovasc Imaging (2017.0) 18 849-850. DOI: 10.1093/ehjci/jex029 15. Abbara S, Blanke P, Maroules CD, Cheezum M, Choi AD, Han BK, Marwan M, Naoum C, Norgaard BL, Rubinshtein R. SCCT guidelines for the performance and acquisition of coronary computed tomographic angiography: a report of the society of Cardiovascular Computed Tomography Guidelines Committee: Endorsed by the North American Society for Cardiovascular Imaging (NASCI). J Cardiovasc Comput Tomogr (2016.0) 10 435-449. DOI: 10.1016/j.jcct.2016.10.002 16. Fleisher LA, Fleischmann KE, Auerbach AD, Barnason SA, Beckman JA, Bozkurt B, Davila-Roman VG, Gerhard-Herman MD, Holly TA, Kane GC. 2014 ACC/AHA guideline on perioperative cardiovascular evaluation and management of patients undergoing noncardiac surgery: executive summary: a report of the American College of Cardiology/American Heart Association Task Force on Practice Guidelines. Circulation (2014.0) 130 2215-2245. DOI: 10.1161/CIR.0000000000000105 17. Chou R. Cardiac screening with electrocardiography, stress echocardiography, or myocardial perfusion imaging: advice for high-value care from the American College of Physicians. Ann Intern Med (2015.0) 162 438-447. DOI: 10.7326/M14-1225 18. Kofoed KF, Engstrøm T, Sigvardsen PE, Linde JJ, Torp-Pedersen C, de Knegt M, Hansen PR, Fritz-Hansen T, Bech J, Heitmann M. Prognostic value of coronary CT angiography in patients with non-ST-segment elevation acute coronary syndromes. J Am Coll Cardiol (2021.0) 77 1044-1052. DOI: 10.1016/j.jacc.2020.12.037 19. Pontone G, Bertella E, Mushtaq S, Loguercio M, Cortinovis S, Baggiano A, Conte E, Annoni A, Formenti A, Beltrama V. Coronary artery disease: diagnostic accuracy of CT coronary angiography—A comparison of high and standard spatial resolution scanning. Radiology (2014.0) 271 688-694. DOI: 10.1148/radiol.13130909 20. Cohen MC, Aretz TH. Histological analysis of coronary artery lesions in fatal postoperative myocardial infarction. Cardiovasc Pathol (1999.0) 8 133-139. DOI: 10.1016/S1054-8807(98)00032-5 21. Wexler L, Brundage B, Crouse J, Detrano R, Fuster V, Maddahi J, Rumberger J, Stanford W, White R, Taubert K. Coronary artery calcification: pathophysiology, epidemiology, imaging methods, and clinical implications. A statement for health professionals from the American Heart Association. Writing Group. Circulation (1996.0) 94 1175-1192. DOI: 10.1161/01.CIR.94.5.1175 22. Budoff MJ, Shaw LJ, Liu ST, Weinstein SR, Mosler TP, Tseng PH, Flores FR, Callister TQ, Raggi P, Berman DS. Long-term prognosis associated with coronary calcification: observations from a registry of 25,253 patients. J Am Coll Cardiol (2007.0) 49 1860-1870. DOI: 10.1016/j.jacc.2006.10.079 23. Valenti V, Hartaigh B, Cho I, Schulman-Marcus J, Gransar H, Heo R, Truong QA, Shaw LJ, Knapper J, Kelkar AA. Absence of coronary artery calcium identifies asymptomatic diabetic individuals at low near-term but not long-term risk of mortality: a 15-year follow-up study of 9715 patients. Circ Cardiovasc Imaging (2016.0) 9 e003528. DOI: 10.1161/CIRCIMAGING.115.003528 24. Chaikriangkrai K, Jhun HY. Palamaner Subash Shantha G, Bin Abdulhak A, Sigurdsson G, Nabi F, Mahmarian JJ, Chang SM. Coronary artery calcium score as a predictor for incident stroke: Systematic review and meta-analysis. Int J Cardiol (2017.0) 236 473-477. DOI: 10.1016/j.ijcard.2017.01.132 25. Arnett DK, Blumenthal RS, Albert MA, Buroker AB, Goldberger ZD, Hahn EJ, Himmelfarb CD, Khera A, Lloyd-Jones D, McEvoy JW. 2019 ACC/AHA guideline on the primary prevention of cardiovascular disease: a report of the American College of Cardiology/American Heart Association Task Force on Clinical Practice Guidelines. Circulation (2019.0) 140 e596-e646. PMID: 30879355 26. Lehmann N, Erbel R, Mahabadi AA, Rauwolf M, Möhlenkamp S, Moebus S, Kälsch H, Budde T, Schmermund A, Stang A. Value of progression of coronary artery calcification for risk prediction of coronary and cardiovascular events: result of the HNR Study (Heinz Nixdorf Recall). Circulation (2018.0) 137 665-679. DOI: 10.1161/CIRCULATIONAHA.116.027034 27. Shen YW, Wu YJ, Hung YC, Hsiao CC, Chan SH, Mar GY, Wu MT, Wu FZ. Natural course of coronary artery calcium progression in Asian population with an initial score of zero. BMC Cardiovasc Disord (2020.0) 20 212. DOI: 10.1186/s12872-020-01498-x 28. Hansen PW, Gislason GH, Jørgensen ME, Køber L, Jensen PF, Torp-Pedersen C, Andersson C. Influence of age on perioperative major adverse cardiovascular events and mortality risks in elective non-cardiac surgery. Eur J Intern Med (2016.0) 35 55-59. DOI: 10.1016/j.ejim.2016.05.028 29. Sunny JC, Kumar D, Kotekar N, Desai N. Incidence and predictors of perioperative myocardial infarction in patients undergoing non-cardiac surgery in a tertiary care hospital. Indian Heart J (2018.0) 70 335-340. DOI: 10.1016/j.ihj.2017.08.010
--- title: Impact of dysautonomic symptom burden on the quality of life in Neuromyelitis optica spectrum disorder patients authors: - Lili Yang - Wenjing Li - Yan Xie - Shuai Ma - Xiaobo Zhou - Xinyue Huang - Song Tan journal: BMC Neurology year: 2023 pmcid: PMC10026430 doi: 10.1186/s12883-023-03162-1 license: CC BY 4.0 --- # Impact of dysautonomic symptom burden on the quality of life in Neuromyelitis optica spectrum disorder patients ## Abstract ### Background This study aimed to investigate the clinical risk factors of dysautonomic symptom burden in neuromyelitis optica spectrum disorder (NMOSD) and its impact on patients’ quality of life. ### Methods A total of 63 NMOSD patients and healthy controls were enrolled. All participants completed the Composite Autonomic Symptom Score 31 (COMPASS-31) to screen for symptoms of autonomic dysfunction. A comprehensive clinical evaluation was performed on NMOSD patients, such as disease characteristics and composite evaluations of life status, including quality of life, anxiety/depression, sleep, and fatigue. Correlated factors of dysautonomic symptoms and quality of life were analyzed. ### Results The score of COMPASS-31 in the NMOSD group was 17.2 ± 10.3, significantly higher than that in healthy controls ($$P \leq 0.002$$). In NMOSD patients, the higher COMPASS-31 score was correlated with more attacks ($r = 0.49$, $P \leq 0.001$), longer disease duration ($r = 0.52$, $P \leq 0.001$), severer disability ($r = 0.50$, $P \leq 0.001$), more thoracic cord lesions ($r = 0.29$, $$P \leq 0.02$$), more total spinal cord lesions ($r = 0.35$, $$P \leq 0.005$$), severer anxiety ($r = 0.55$, $P \leq 0.001$), severer depression ($r = 0.48$, $P \leq 0.001$), severer sleep disturbances ($r = 0.59$, $P \leq 0.001$), and severer fatigue ($r = 0.56$, $P \leq 0.001$). The disability, total spinal cord lesions, and fatigue were revealed to be independently associated factors. Further analysis revealed that the COMPASS-31 score was independently correlated with scores of all the domains of patients’ quality of life scale ($P \leq 0.05$). ### Conclusions Dysautonomic symptom burden is correlated with decreased quality of life and certain clinical characteristics such as disability, the burden of spinal cord lesions, and fatigue in NMOSD patients. Investigation and proper management of autonomic dysfunction may help to improve the quality of life in patients with NMOSD. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12883-023-03162-1. ## Background Neuromyelitis optica spectrum disorder (NMOSD), an idiopathic inflammatory central nervous system (CNS) disorder, is characterized by the presence of an antibody for the water channel aquaporin-4 (AQP4) and clinical characteristics, which are distinguished from multiple sclerosis (MS) [1]. The incidence of NMOSD is estimated at 0.57–4.52 per 100,000, with a high female-to-male ratio (3–9:1) [2, 3]. The incidence in the Chinese population is as high as $\frac{3.31}{100}$,000 [4]. The most common clinical phenotypes are transverse myelitis (TM) and optic neuritis (ON) [1, 5]. Dysautonomic symptom burden is common in patients with neurological diseases, including MS [6, 7]. It has been found to seriously impact the patient’s quality of life [8–10] and is closely correlated with depression and fatigue [10, 11]. In recent years, research on autonomic dysfunction in NMOSD patients has been obtained. Two studies compared autonomic dysfunction between patients with NMOSD and those with MS [12, 13], finding a significant proportion of NMOSD patients suffered autonomic dysfunction and were more often symptomatic than MS patients in certain domains of dysautonomia. Another study suggested $74.1\%$ of patients with NMOSD suffered dysautonomic symptoms and investigated the correlated clinical and MRI factors of dysautonomia [14]. However, current research on autonomic dysfunction in NMOSD has not been able to explore its correlation with patients' life status (depression, anxiety, sleep, fatigue), and the extent of its impact on patients’ quality of life. Herein, we investigated dysautonomic symptom burden by using a well-recognized autonomic symptom questionnaire, the Composite Autonomic Symptom Score (COMPASS-31), in patients with NMOSD, as well as explored its impact on patient’s quality of life and its clinical correlates among demographic features, disease characteristics, and life status evaluations. ## Participants We prospectively enrolled patients with NMOSD, who visited our clinic from June 2021 to July 2022. NMOSD was diagnosed according to the 2015 International Panel for Neuromyelitis Optica diagnosis (IPND) criteria [1]. The exclusion criteria were as follows: [1] history of drug or alcohol abuse or other major clinical or psychiatric conditions, especially those that could result in autonomic dysfunction, such as parkinsonism, Sjogren's syndrome, amyloidosis, leprosy, polyneuropathy, diabetes, etc.; [ 2] receiving acute immunoregulatory treatment; [3] with cerebral lesions or severe visual impairments; [4] taking medications such as antihypertensives, anticholinergics, and antiarrhythmics, which may influence autonomic function; and [5] inability to complete all questionnaires with the assistance of neurologists. Demographic data and disease features were collected including gender, age, body mass index (BMI), the presence of antibodies associated with CNS demyelinating disease in serum (including AQP4-Ab, anti-myelin oligodendrocyte glycoprotein antibody (MOG-Ab), and anti-glial fibrillary acidic protein antibody (GFAP-Ab)), the number of clinical attacks, disease duration, the segments of spinal cord lesions (cervical, thoracic, lumbosacral and total), degree of disability, clinical phenotype, and current preventive therapy. Antibodies in serum were tested with a transfected cell-based assay [15]. The degree of disability was evaluated by two neurologists according to the Expanded Disability Status Scale (EDSS) score [16]. Severe visual impairment was referred to as a visual function subscore of 6 according to the EDSS score. The segments of spinal cord lesions were acquired from 3T spine magnetic resonance images. Healthy controls (HCs) were recruited among individuals who attended the hospital for annual health check-ups, with no major clinical or psychiatric conditions. Demographic data and BMI regarding the HCs were collected. The Ethics Committee of Sichuan Provincial People’s Hospital approved the study. Recruited participants provided written informed consent before enrolling in the study. ## Composite Autonomic Symptom Score 31 All participants were requested to complete the COMPASS 31 questionnaire independently according to the actual situation. COMPASS-31 comprises 6 domains with 31 items (orthostatic intolerance 4 items, vasomotor 3 items, secretomotor 4 items, gastrointestinal 12 items, bladder 3 items, and pupillomotor 5 items) and provides the minimal weighted total score equals 0 and the maximum weighted total score equals 100 [17]. The higher the score, the more severe the dysautonomic symptoms. ## Clinical composite evaluation of living status in NMOSD patients All enrolled NMOSD patients were requested to fulfill a composite evaluation of living status, including anxiety, depression, sleep quality, and fatigue. The Hospital Anxiety and Depression Scale (HADS) used in this study was developed to identify cases of anxiety disorders and depression among patients in nonpsychiatric hospital clinics [18]. It is divided into an anxiety subscale (HADS-A) and a depression subscale (HADS-D), both containing seven intermingled items. The score of the HADS-A and HADS-D both ranges from 0 to 21, and higher scores indicate more severe anxiety/depression. The sleep quality of patients was assessed through the Pittsburgh Sleep Quality Index (PSQI) [19]. The global PSQI score ranges from 0 to 21, with higher scores indicating worse sleep quality. Fatigue was evaluated with the Fatigue Severity Scale (FSS), which was a self-report instrument to evaluate patients' perceptions of the functional limitations caused by fatigue within the last week [20]. Possible global scores range from 7 to 63, where higher scores indicate more severe fatigue. ## Quality of life evaluation of NMOSD patients All NMOSD patients completed the evaluation of their quality of life via the 36-item short-form health survey (SF-36). SF-36 evaluates 8 dimensions: physical functioning (10 items), physical role fulfillment (4 items), bodily pain (2 items), general health (5 items), vitality (4 items), social functioning (2 items), emotional role fulfillment (3 items), and mental health (5 items) [21]. The scale’s total possible score is 145, with a higher score reflecting better quality of life. All the scales used in the present study have been utilized in previous NMOSD studies [13, 14, 22–24]. All the Chinese versions of these scales have been validated previously [25–29]. A psychologist administered the HADS questionnaires; the other questionnaires were completed by the participants themselves in the presence of a neurologist, who assisted the participants in reading and understanding the items. ## Statistical analysis All statistical analyses were carried out using the statistical software GraphPad Prism (version 8, San Diego, CA). To compare the demographic characteristics between the NMOSD and HC groups, Fisher’s exact test was used in the gender ratio analysis, and the Mann–Whitney U test was used in the comparisons of age, BMI, COMPASS-31 score, and its subscores. To analyze the related factors of dysautonomic symptoms in NMOSD, Mann–Whitney and Kruskal–Wallis tests were used to determine whether COMPASS-31 scores/subscores differed among groups defined by gender, AQP4 seropositivity or seronegativity, clinical phenotype, and current preventive therapy. Spearman’s ranked correlation analysis was used to explore the relationships between the COMPASS-31 score/subscores and the independent variables, including age, BMI, number of attacks, disease duration, EDSS, segments of spinal cord lesions (cervical, thoracic, lumbosacral, total separately), HADS-A, HADS-D, PSQI, and FSS score. Multiple linear regression was used to further assess the independent factor of the COMPASS-31 score/subscore (checking the normality of residuals). Age, gender, BMI, clinical phenotype, number of attacks, disease duration, current therapy, EDSS, segments of total spinal cord lesions, HADS, PSQI, and FSS score were included as possible independent variables for the multiple linear regression model. $P \leq 0.05$ was considered statistically significant. To analyze the influence of dysautonomic symptom burden on patients’ quality of life, another multiple linear regression model was established to assess the independent contributor of different domains of SF-36 score in NMOSD patients (checking the normality of residuals). Age, gender, BMI, clinical phenotype, number of attacks, disease duration, current therapy, EDSS, COMPASS-31, HADS, PSQI, and FSS score were included as possible independent variables for this multiple linear regression model. $P \leq 0.05$ was considered statistically significant. ## Demographic data of NMOSD patients and HCs A total of 63 NMOSD patients and 63 HCs were enrolled, with no significant difference in age or gender (seen in Table 1).Table 1Demographic, clinical characteristics and composite evaluation in NMOSD and HCNMOSD ($$n = 63$$)HC ($$n = 63$$)P valueAge (years), mean [SD] (range)41 [13.8] (18–69)37.8 [12.5] (20–60)0.50Sex (F), N (%)50 ($79.4\%$)47 ($74.6\%$)0.67BMI, mean [SD] (range)22.7 [4.0] (12.6–35.4)22.7 [4.1] (17.1–39.0)0.59Antibody in serum AQP4 seropositivity, N(%)55 ($87.3\%$) MOG seropositivity, N(%)4 ($6.3\%$) Negative4 ($6.3\%$)Number of attacks, mean [SD] (range)3.7 [3.9] (1–21)Disease duration (year), mean [SD] (range)4.2 [4.3] (0.25–20)EDSS, mean [SD] (range)3.1 [1.9] (0–7.5)Clinical phenotype, N(%) ON15 ($23.8\%$) TM22 ($34.9\%$) ON + TM26 ($41.3\%$)Segments of MR lesions, mean [SD] (range) Cervical cord lesions2.2 [2.2] (0–7) Thoracic cord lesions2.2 [3.4] (0–12) Total number of spinal cord lesion4.3 [4.3] (0–17)Therapy, N(%) MMF37 ($58.7\%$) AZA5 ($7.9\%$) RTX17 ($30.0\%$) periodic IVIG1 ($1.6\%$) only low-dose prednisolone3 ($4.8\%$)COMPASS-31, mean [SD] (range)17.2 [10.3] (1–43)11.6 [7.3] (0–32)0.002 Orthostatic intolerance1.8 [2.3] (0–10)1.1 [1.9] (0–6)0.003 Vasomotor0.8 [1.7] (0–6)0.2 [0.8] (0–5)0.007 Secretomotor2.1 [1.7] (0–6)1.3 [1.4] (0–6)0.006 Gastrointestinal6.2 [4.0] (0–15)5.2 [3.8] (0–16)0.18 Bladder1.4 [1.7] (0–6)0.4 [0.9] (0–4) < 0.001 Pupillomotor4.7 [3.5] (0–14)3.4 [2.9] (0–10)0.04F female, M male, BMI Body Mass Index, AQP4-IgG IgG autoantibodies to aquaporin 4, EDSS Expanded Disability Status Scale, ON optica neuritis, TM transverse myelitis, MMF mycophenolate mofetil, AZA azathioprine, RTX rituximab, IVIG intravenous immunoglobulin, COMPASS-31 Composite Autonomic Symptom Score 31 The disease characteristics of NMOSD patients were also summarized in Table 1. The distributions of the clinical phenotype were as follows: 15 patients had only ON ($23.8\%$), 22 patients had only TM ($34.9\%$), and 30 patients had both ON and TM ($41.3\%$). Four patients in the ON + TM subgroup also experienced area postrema syndrome. The segments of spinal cord lesions were collected, while none of the patients has a lumbosacral spinal cord involvement. Regarding the distribution of current preventive therapy, 42 patients ($66.7\%$) used immunosuppressants such as mycophenolate mofetil (MMF, 37 patients) or azathioprine (AZA, 5 patients); 17 patients ($30.0\%$) received rituximab (RTX); and the remaining 4 patients used other therapies, such as periodic intravenous immunoglobulin (1 patient) or only low-dose prednisolone (3 patients). ## The comparison of COMPASS score/subscore between NMOSD patients and HCs The COMPASS-31 score in NMOSD patients was 17.2 ± 10.3, significantly higher than that in HCs ($$P \leq 0.002$$). For the six subdomains of COMPASS-31, the NMOSD patients had significantly higher scores than HCs in orthostatic intolerance ($$P \leq 0.003$$), vasomotor ($$P \leq 0.007$$), secretomotor ($$P \leq 0.006$$), bladder ($P \leq 0.001$), and pupillomotor ($$P \leq 0.04$$) scores. The statistical results could be seen in Table 1. We also analyzed the frequency of patients with general dysautonomic symptoms (COMPASS-31 score > 0), as well as the frequency of those with dysfunction in each dimension (the corresponding subscore > 0). The results showed that all 63 NMOSD patients suffered dysautonomic symptoms, the number and frequency of NMOSD patients suffering dysfunction in each dimension were as follows: 40 patients ($63.5\%$) with orthostatic intolerance, 42 ($66.7\%$) with vasomotor symptoms, 57 ($90.5\%$) with secretomotor symptoms, 60 ($95.2\%$) with gastrointestinal symptoms, 49 ($77.8\%$) with bladder symptoms, and 58 ($92.1\%$) with pupillomotor symptoms. ## Clinical factors associated with the COMPASS-31 total score/subscore in NMOSD patients We further analyzed clinical factors associated with dysautonomic symptoms in NMOSD patients, to reveal the relevant factors and intervention targets. Firstly, we analyzed the difference in COMPASS-31 total scores/subscores among NMOSD subgroups distributed by gender, AQP4 seropositivity or seronegativity, clinical phenotype, and current preventive therapy, considering each of these independent variables separately. We found that the COMPASS-31 total scores were significantly different among patients with different clinical phenotypes ($$P \leq 0.03$$) (Fig. 1, Supplementary Table S1). Regarding subscores, the scores of vasomotor symptoms were significantly different in patients with different preventive therapy ($$P \leq 0.04$$). Meanwhile, female patients showed higher scores of gastrointestinal symptoms than male patients ($$P \leq 0.02$$). There were no significant differences in the COMPASS-31 total score and subscore among the other NMOSD subgroups. Fig. 1The comparisons of COMPASS-31 score/subscore among NMOSD subgroups, only displaying the significantly different results ($P \leq 0.05$). The detailed results of the comparisons of COMPASS-31 score/subscore among NMOSD subgroups distributed by gender, AQP4 seropositivity or seronegativity, clinical phenotype, and current preventive therapy could be seen in Supplementary Table S1. COMPASS-31, Composite Autonomic Symptom Score 31. ON, optica neuritis; TM, transverse myelitis; MMF, mycophenolate mofetil; AZA, azathioprine; RTX, rituximab Furthermore, we performed correlation analyses between COMPASS-31 scores/subscores and other quantitative variables separately. The results demonstrated that higher COMPASS-31 scores were correlated with more attacks ($r = 0.49$, $P \leq 0.001$), longer disease duration ($r = 0.52$, $P \leq 0.001$), higher EDSS score ($r = 0.50$, $P \leq 0.001$), more thoracic cord lesions ($r = 0.29$, $$P \leq 0.02$$), more total spinal cord lesions ($r = 0.35$, $$P \leq 0.005$$), higher HADS-A ($r = 0.55$, $P \leq 0.001$), higher HADS-D ($r = 0.48$, $P \leq 0.001$), higher PSQI ($r = 0.59$, $P \leq 0.001$), and higher FSS ($r = 0.56$, $P \leq 0.001$) in NMOSD patients (displayed in Fig. 2).Fig. 2Correlation and multiple linear regression analysis between COMPASS-31 score and several independent variables. The r values and P values are labeled. A multiple linear regression model was used to distinguish any factors that were independently related to the COMPASS-31 score, and Pmulti indicates the statistical result. A value of $P \leq 0.05$ was considered significant. * Asterisks mark the variables that are significantly correlated with COMPASS-31 scores in the correlation analysis; COMPASS-31 scores were correlated with the number of attacks, disease duration, EDSS, segments of spinal cord lesions (thoracic, total), HADS-A, HADS-D, PSQI, and FSS. # This symbol indicates that the variable was independently correlated with the COMPASS-31 score in the multiple linear regression model. EDSS, segments of total spinal cord lesions, and FSS were the independent correlated factor. COMPASS-31, Composite Autonomic Symptom Score 31; EDSS, Expanded Disability Status Scale; HADS-A, Hospital Anxiety and Depression Scale–Anxiety; HADS-D, Hospital Anxiety and Depression Scale–Depression; PSQI, Pittsburgh Sleep Quality Index; FSS, Fatigue Severity Scale Regarding the subscore of COMPASS-31, only the score of vasomotor symptoms was moderately correlated with HADS-A ($r = 0.35$, $$P \leq 0.004$$), while other correlations were mildly correlated with limited significance. The detailed results could be seen in Supplementary table S2. In the multivariable linear regression analysis model, we found the independently associated variables of the COMPASS-31 score were the EDSS score ($$P \leq 0.002$$), the total number of spinal cord lesions ($$P \leq 0.009$$), and the FSS score ($$P \leq 0.031$$) (seen in Fig. 2). There were no independently associated variables of COMPASS-31 subscores. The detailed statistical results could be seen in Supplementary table S3. ## The influencing factors of the quality of life in NMOSD patients Through a multivariable linear regression model, which included comprehensive underlying factors affecting the patient’s quality of life, we found that the COMPASS-31 score was the independently correlated factor of all the domains of SF-36 ($P \leq 0.05$). BMI was also found to be the independently correlated factor of the score of general health in SF-36 ($$P \leq 0.030$$). The detailed statistical data could be seen in Table 2.Table 2Multivariable linear regression of the SF-36 in NMOSD patientsVariablesBS.E$95\%$CIPPhysical Functioning COMPASS-31-1.880.44-2.76 to -0.99 < 0.001Role Physical COMPASS-31-1.950.65-3.27 to -0.640.004Bodily Pain COMPASS-31-1.120.29-1.71 to -0.54 < 0.001General Health COMPASS-31-1.180.27-1.72 to -0.63 < 0.001 BMI-1.280.57-2.43 to -0.130.030Vitality COMPASS-31-1.150.35-1.85 to -0.460.002Social Functioning COMPASS-31-1.430.38-2.19 to -0.67 < 0.001Role Emotional COMPASS-31-2.600.66-3.94 to -1.27 < 0.001Mental Health COMPASS-31-0.580.19-0.96 to -0.210.003Reported Health Transition COMPASS-311.140.530.08 to 2.200.035SF-36 the 36-item short-form health survey, BMI Body Mass Index, COMPASS-31 Composite Autonomic Symptom Score 31 ## Discussion Our study found that NMOSD patients reported more severe dysautonomic symptoms than the HCs with age and gender matched. We further revealed the associated factors of dysautonomic symptoms in NMOSD patients. Among the correlated factors, the EDSS score, the total number of spinal cord lesions, and fatigue were independent risk factors. Further analysis showed that dysautonomic symptom burden was an independent influencing factor in all domains of scale on patients’ quality of life in NMOSD. To our knowledge, it’s the first investigation of the impacts of dysautonomic symptom burden on patients’ quality of life and the associations of dysautonomic symptoms with the life status evaluations in NMOSD. The autonomic function is commonly assessed in two ways, patient-reported symptom studies (often using different questionnaires) and assessment of autonomic function/dysfunction in the laboratory [30]. Laboratory evaluation, although more objective, is time-consuming since it requires multiple tests regarding different domains of autonomic function. Moreover, it may only be positive in patients with severe autonomic dysfunction. For example, a previous study combined the self-reported questionnaires and laboratory measurements in NMOSD patients showed that all patients had self-reported dysautonomic symptoms (20 patients, COMPASS > 0), but only 11 ($55\%$) were positive for tests [13]. The patients with negative laboratory tests do endure dysautonomic symptoms subjectively, which would impact their health status. These subjective feelings should not be directly ignored when investigating the impact of dysautonomic symptom burden on the quality of life in NMOSD patients. Previous MS studies have proved that patients with laboratory-confirmed autonomic dysfunction score higher on certain domains of the COMPASS-31 scale [30, 31], suggesting that the quantified COMPASS-31 score was in good agreement with laboratory tests. Taking into account all these considerations, we used the COMPASS-31 scale to analyze the autonomic function in NMOSD patients, to comprehensively and sensitively identify the patients’ dysautonomic symptom burden. Autonomic dysfunction is common in the general population, especially in the elderly [11], since hormonal changes and oxidative stress also participate in the progress of autonomic dysfunction [32]. However, the previous NMOSD study of autonomic dysfunction did not enroll healthy participants with age and gender matched as controls [13]. Our study had filled this blank. Through comparison with age-gender-matched HCs, patients with NMOSD were found to have significantly higher scores than HCs in multiple domains of dysautonomic symptoms, except gastrointestinal symptoms. This finding suggested that NMOSD patients were more likely to suffer orthostatic intolerance, vasomotor symptoms, secretomotor symptoms, bladder symptoms, and pupillomotor symptoms than the general population, while the gastrointestinal symptoms in NMOSD patients might not be disease-related. We further analyzed clinical factors associated with dysautonomic symptoms and found that the degree of disability (EDSS score), spinal cord lesion burden, and fatigue were independent risk factors for dysautonomic symptoms in patients with NMOSD. Our findings were consistent with the previous studies in NMOSD [14] and earlier studies in MS [30, 33] regarding the association of dysautonomic symptoms with disability and spinal cord lesion burden. Our stratified analysis in NMOSD patients with myelitis also showed that the more severe sequela after myelitis (represented by the sum scores of pyramidal, sensory, bowel & bladder, and walking subdomains in EDSS) was correlated with more severe autonomic dysfunction in NMOSD patients. It can be explained by the anatomical features of the autonomic nervous system, where the preganglionic cells of the sympathetic nervous system are located between the thoracic and upper lumbar segments of the spinal cord [14]. However, autonomic dysfunction may not only be a consequence of disease progression, but it may also be a contributor. Previous studies suggested that the autonomic nervous system and the immune system interacted on several levels [34]. The parasympathetic nervous system may play an important role in alerting the CNS to the presence of inflammation through the "cholinergic anti-inflammatory pathway" [30]. It is speculated that this pathway suppresses inflammation and immune responses by integrating signals from the immune system and the nervous system [7, 35]. Another branch of the ANS, the sympathetic system, was also found to modulate the CNS inflammatory response in MS through laboratory and genetic studies [36, 37]. Whether the autonomic dysfunction in NMOSD was a promoting factor in the disease progression needed to be further researched. The close relationship between dysautonomic symptoms and fatigue in NMOSD patients was revealed for the first time in this study. Fatigue is extremely common among NMOSD patients and can significantly influence patients’ quality of life, according to recent clinical findings [38–40]. As a complex and multifactorial problem, the underlying pathophysiology of fatigue is unclear. A close relationship between dysautonomic symptoms and fatigue had also been found in MS [10], and some hypothesized that fatigue may be attributed to the orthostatic intolerance of MS patients [41, 42]. Researchers also demonstrated that MS patients with fatigue had adrenergic hyporesponsive [41], which was not found in MS patients without fatigue or normal controls. Our findings suggested that autonomic dysfunction might play an important role in the underlying mechanism of NMOSD fatigue. Further, through a multiple regression model integrating multifactorial variables, we were surprised to find that dysautonomic symptom burden was an independent influencing factor of all the domains of the quality of life assessed by SF-36. Dysautonomic symptoms have been demonstrated to cause physical and emotional discomfort, limit the activities of daily living and social participation, and ultimately affect the health-related quality of life [11]. Studies in many diseases provide overwhelming evidence for an association between the presence of autonomic symptoms and reduced quality of life [8, 10, 43]. In the study of NMOSD, earlier studies showed that bowel and bladder dysfunction was associated with quality of life or spinal cord atrophy [44, 45]. We confirmed the close relationship between the quality of life in NMOSD patients and the self-reported overall autonomic dysfunction. Our findings suggest that more attention should be paid to screening and managing the dysautonomic symptoms of NMOSD patients, which may be vital in improving patients’ quality of life. ## Limitation This study has some limitations. First, this study was a cross-sectional observational study that only evaluated the influence of dysautonomic symptoms on quality of life, while further prospective studies are needed to judge the impact of dysautonomic symptoms on disease activity or prognosis. Second, the COMPASS 31 questionnaire requires memory data on various autonomic symptoms that reflect subjective experiences and feelings. The subjective nature of the assessment tools employed sacrificed certain accuracies. Third, this study was a single-center clinical study with a limited sample size, future studies with larger samples and stratified analyses in patients with different disease severity (for example, different degrees of disability or spinal cord injury) are needed to validate our findings and to reveal promising therapeutic targets for improving dysautonomic symptoms. ## Conclusion The present study investigated self-perceived dysautonomic profiles by using COMPASS-31 in patients with NMOSD and found that the dysautonomic symptom burden in NMOSD patients was much more severe than HCs. Disability, burden of spinal cord lesions, and fatigue were independently correlated with dysautonomic symptoms in NMOSD patients. Moreover, the dysautonomic symptom burden was an independent influencing factor of the patients’ quality of life in NMOSD. These findings might help us identify patients with a high risk of autonomic dysfunction and suggest that the evaluation and management of autonomic dysfunction are of great significance for improving the quality of life in NMOSD patients. ## Supplementary Information Additional file 1: Supplementary Table S1. The comparisons of COMPASS-31 score/subscore among NMOSD subgroups distributed by gender, serum AQP4-IgG positive or negative, clinical phenotype, and current preventive therapy. Supplementary Table S2. The correlation between COMPASS-31 score/subscore and clinical variables in NMOSD patients. Supplementary Table S3. Multivariable linear regression of the COMPASS-31 score in NMOSD patients. ## References 1. Wingerchuk DM, Banwell B, Bennett JL, Cabre P, Carroll W, Chitnis T. International consensus diagnostic criteria for neuromyelitis optica spectrum disorders. Neurology (2015) 85 177-189. DOI: 10.1212/WNL.0000000000001729 2. Bukhari W, Prain KM, Waters P, Woodhall M, O'Gorman CM, Clarke L. Incidence and prevalence of NMOSD in Australia and New Zealand. J Neurol Neurosurg Psychiatry (2017) 88 632-638. DOI: 10.1136/jnnp-2016-314839 3. Lana-Peixoto MA, Talim NC, Pedrosa D, Macedo JM, Santiago-Amaral J. Prevalence of neuromyelitis optica spectrum disorder in Belo Horizonte, Southeast Brazil. Mult Scler Relat Disord (2021) 50 102807. DOI: 10.1016/j.msard.2021.102807 4. Hor JY, Lim TT, Chia YK, Ching YM, Cheah CF, Tan K. Prevalence of neuromyelitis optica spectrum disorder in the multi-ethnic Penang Island, Malaysia, and a review of worldwide prevalence. Mult Scler Relat Disord (2018) 19 20-24. DOI: 10.1016/j.msard.2017.10.015 5. Li Q, Wang B, Cheng B, Liu C, Li N, Dai G. Efficacy and safety of rehabilitation exercise in neuromyelitis optica spectrum disorder during the acute phase: a prospective cohort study. Mult Scler Relat Disord (2022) 61 103726. DOI: 10.1016/j.msard.2022.103726 6. Habek M, Crnosija L, Lovric M, Junakovic A, KrbotSkoric M, Adamec I. Sympathetic cardiovascular and sudomotor functions are frequently affected in early multiple sclerosis. Clin Auton Res (2016) 26 385-393. DOI: 10.1007/s10286-016-0370-x 7. Racosta JM, Kimpinski K. Autonomic dysfunction, immune regulation, and multiple sclerosis. Clin Auton Res (2016) 26 23-31. DOI: 10.1007/s10286-015-0325-7 8. Tomic S, Rajkovaca I, Pekic V, Salha T, Misevic S. Impact of autonomic dysfunctions on the quality of life in Parkinson's disease patients. Acta Neurol Belg (2017) 117 207-211. DOI: 10.1007/s13760-016-0739-6 9. Merola A, Romagnolo A, Rosso M, Suri R, Berndt Z, Maule S. Autonomic dysfunction in Parkinson's disease: A prospective cohort study. Mov Disord (2018) 33 391-397. DOI: 10.1002/mds.27268 10. Cortez MM, Nagi Reddy SK, Goodman B, Carter JL, Wingerchuk DM. Autonomic symptom burden is associated with MS-related fatigue and quality of life. Mult Scler Relat Disord (2015) 4 258-263. DOI: 10.1016/j.msard.2015.03.007 11. Renno-Busch S, Hildesheim H, van Uem JMT, Sünkel U, Röben B, Brockmann K. Autonomic symptoms in older adults are common and associated with health-related quality of life. Front Neurol (2021) 12 757748. DOI: 10.3389/fneur.2021.757748 12. Habek M, Andabaka M, Fanciulli A, BreclJakob G, Drulović J, Leys F. Sudomotor dysfunction in people with neuromyelitis optica spectrum disorders. Eur J Neurol (2022) 29 2772-2780. DOI: 10.1111/ene.15413 13. Crnošija L, KrbotSkorić M, Andabaka M, Junaković A, Martinović V, Ivanović J. Autonomic dysfunction in people with neuromyelitis optica spectrum disorders. Mult Scler (Houndmills, Basingstoke, England) (2020) 26 688-695. DOI: 10.1177/1352458519837703 14. Kwon S, Kim YS, Kim J, Kim BJ, Min JH. Clinical and MRI correlates of autonomic dysfunction in neuromyelitis optica spectrum disorder. Mult Scler Relat Disord (2020) 43 102215. DOI: 10.1016/j.msard.2020.102215 15. Waters PJ, McKeon A, Leite MI, Rajasekharan S, Lennon VA, Villalobos A. Serologic diagnosis of NMO: a multicenter comparison of aquaporin-4-IgG assays. Neurology (2012) 78 665-71; discussion 9. DOI: 10.1212/WNL.0b013e318248dec1 16. Kurtzke JF. Rating neurologic impairment in multiple sclerosis: an expanded disability status scale (EDSS). Neurology (1983) 33 1444-1452. DOI: 10.1212/WNL.33.11.1444 17. Sletten DM, Suarez GA, Low PA, Mandrekar J, Singer W. COMPASS 31: a refined and abbreviated Composite Autonomic Symptom Score. Mayo Clin Proc (2012) 87 1196-1201. DOI: 10.1016/j.mayocp.2012.10.013 18. Zigmond AS, Snaith RP. The hospital anxiety and depression scale. Acta Psychiatr Scand (1983) 67 361-370. DOI: 10.1111/j.1600-0447.1983.tb09716.x 19. Buysse DJ, Reynolds CF, Monk TH, Berman SR, Kupfer DJ. The Pittsburgh Sleep Quality Index: a new instrument for psychiatric practice and research. Psychiatry Res (1989) 28 193-213. DOI: 10.1016/0165-1781(89)90047-4 20. Krupp LB, LaRocca NG, Muir-Nash J, Steinberg AD. The fatigue severity scale. Application to patients with multiple sclerosis and systemic lupus erythematosus. Arch Neurol (1989) 46 1121-3. DOI: 10.1001/archneur.1989.00520460115022 21. Ware JE, Kosinski M, Gandek B. SF-36 Health Survey: Manual and Interpretation Guide (1993) 22. Kim S, Lee EJ, Kim KW, Seo D, Moon S, Kim KK. Quality of life of patients with multiple sclerosis and neuromyelitis optica spectrum disorders: cross-sectional and longitudinal analysis. Mult Scler Relat Disord (2022) 58 103500. DOI: 10.1016/j.msard.2022.103500 23. Haji Molla Rabi S, Shahmirzaei S, Sahraian MA, KazemiMozdabadi RS, RezaeiAliabadi H, Gheini MR. Sleep disorders as a possible predisposing attack factor in neuromyelitis optica spectrum disorder (NMOSD): a case-control study. Clin Neurol Neurosurg (2021) 204 106606. DOI: 10.1016/j.clineuro.2021.106606 24. Wang T, Ruan H, Fan P, Jia M, Qiu W, Li K. Social participation and quality of life among patients with neuromyelitis optica spectrum disorders: the mediating effects of depression. Mult Scler Relat Disord (2022) 57 103445. DOI: 10.1016/j.msard.2021.103445 25. Zhang Z, Ma Y, Fu L, Li L, Liu J, Peng H. Combination of Composite Autonomic Symptom Score 31 and Heart Rate Variability for Diagnosis of Cardiovascular Autonomic Neuropathy in People with Type 2 Diabetes. J Diabetes Res (2020) 2020 5316769. DOI: 10.1155/2020/5316769 26. Leung CM, Ho S, Kan CS, Hung CH, Chen CN. Evaluation of the Chinese version of the Hospital Anxiety and Depression Scale. A cross-cultural perspective. Int J Psychosom (1993) 40 29-34. PMID: 8070982 27. Tsai PS, Wang SY, Wang MY, Su CT, Yang TT, Huang CJ. Psychometric evaluation of the Chinese version of the Pittsburgh Sleep Quality Index (CPSQI) in primary insomnia and control subjects. Qual Life Res (2005) 14 1943-1952. DOI: 10.1007/s11136-005-4346-x 28. Li L, Wang HM, Shen Y. Chinese SF-36 Health Survey: translation, cultural adaptation, validation, and normalisation. J Epidemiol Community Health (2003) 57 259-263. DOI: 10.1136/jech.57.4.259 29. Wang MY, Liu IC, Chiu CH, Tsai PS. Cultural adaptation and validation of the Chinese version of the Fatigue Severity Scale in patients with major depressive disorder and nondepressive people. Qual Life Res (2016) 25 89-99. DOI: 10.1007/s11136-015-1056-x 30. Habek M. Immune and autonomic nervous system interactions in multiple sclerosis: clinical implications. Clin Auton Res (2019) 29 267-275. DOI: 10.1007/s10286-019-00605-z 31. Ruška B, Pavičić T, Pavlović I, Junaković A, Adamec I, Crnošija L. Performance of the COMPASS-31 questionnaire with regard to autonomic nervous system testing results and medication use: a prospective study in a real-life setting. Neurol Sci (2018) 39 2079-2084. DOI: 10.1007/s10072-018-3542-8 32. Saffrey MJ. Cellular changes in the enteric nervous system during ageing. Dev Biol (2013) 382 344-355. DOI: 10.1016/j.ydbio.2013.03.015 33. Kale N, Magana S, Agaoglu J, Tanik O. Assessment of autonomic nervous system dysfunction in multiple sclerosis and association with clinical disability. Neurol Int (2009) 1 e5. DOI: 10.4081/ni.2009.e4 34. Benarroch EE. Autonomic nervous system and neuroimmune interactions: new insights and clinical implications. Neurology (2019) 92 377-385. DOI: 10.1212/WNL.0000000000006942 35. Tracey KJ. Reflex control of immunity. Nat Rev Immunol (2009) 9 418-428. DOI: 10.1038/nri2566 36. Zaffaroni M, Marino F, Bombelli R, Rasini E, Monti M, Ferrari M. Therapy with interferon-beta modulates endogenous catecholamines in lymphocytes of patients with multiple sclerosis. Exp Neurol (2008) 214 315-321. DOI: 10.1016/j.expneurol.2008.08.015 37. KrbotSkorić M, Crnošija L, Gabelić T, Barun B, Adamec I, Junaković A. Autonomic symptom burden can predict disease activity in early multiple sclerosis. Mult Scler Relat Disord (2019) 28 250-255. DOI: 10.1016/j.msard.2019.01.005 38. Shi Z, Chen H, Lian Z, Liu J, Feng H, Zhou H. Factors that impact health-related quality of life in neuromyelitis optica spectrum disorder: anxiety, disability, fatigue and depression. J Neuroimmunol (2016) 293 54-58. DOI: 10.1016/j.jneuroim.2016.02.011 39. Seok JM, Choi M, Cho EB, Lee HL, Kim BJ, Lee KH. Fatigue in patients with neuromyelitis optica spectrum disorder and its impact on quality of life. PLoS One (2017) 12 e0177230. DOI: 10.1371/journal.pone.0177230 40. Pan J, Zhao P, Cai H, Su L, Wood K, Shi FD. Hypoxemia, Sleep Disturbances, and Depression Correlated with Fatigue in Neuromyelitis Optica Spectrum Disorder. CNS Neurosci Ther (2015) 21 599-606. DOI: 10.1111/cns.12411 41. Flachenecker P, Rufer A, Bihler I, Hippel C, Reiners K, Toyka KV. Fatigue in MS is related to sympathetic vasomotor dysfunction. Neurology (2003) 61 851-853. DOI: 10.1212/01.WNL.0000080365.95436.B8 42. Keselbrener L, Akselrod S, Ahiron A, Eldar M, Barak Y, Rotstein Z. Is fatigue in patients with multiple sclerosis related to autonomic dysfunction?. Clin Auton Res (2000) 10 169-175. DOI: 10.1007/BF02291352 43. Allan L, McKeith I, Ballard C, Kenny RA. The prevalence of autonomic symptoms in dementia and their association with physical activity, activities of daily living and quality of life. Dement Geriatr Cogn Disord (2006) 22 230-237. DOI: 10.1159/000094971 44. Wang Y, Wu A, Chen X, Zhang L, Lin Y, Sun S. Comparison of clinical characteristics between neuromyelitis optica spectrum disorders with and without spinal cord atrophy. BMC Neurol (2014) 14 246. DOI: 10.1186/s12883-014-0246-4 45. Mutch K, Zhao S, Hamid S, Methley A, Elsone L, Singh G. Bladder and bowel dysfunction affect quality of life A cross sectional study of 60 patients with aquaporin-4 antibody positive Neuromyelitis Optica spectrum disorder. Mult Scler Relat Disord (2015) 4 614-8. DOI: 10.1016/j.msard.2015.07.015
--- title: Robust classification using average correlations as features (ACF) authors: - Yannis Schumann - Julia E. Neumann - Philipp Neumann journal: BMC Bioinformatics year: 2023 pmcid: PMC10026437 doi: 10.1186/s12859-023-05224-0 license: CC BY 4.0 --- # Robust classification using average correlations as features (ACF) ## Abstract ### Motivation In single-cell transcriptomics and other omics technologies, large fractions of missing values commonly occur. Researchers often either consider only those features that were measured for each instance of their dataset, thereby accepting severe loss of information, or use imputation which can lead to erroneous results. Pairwise metrics allow for imputation-free classification with minimal loss of data. ### Results Using pairwise correlations as metric, state-of-the-art approaches to classification would include the K-nearest-neighbor- (KNN) and distribution-based-classification-classifier. Our novel method, termed average correlations as features (ACF), significantly outperforms those approaches by training tunable machine learning models on inter-class and intra-class correlations. Our approach is characterized in simulation studies and its classification performance is demonstrated on real-world datasets from single-cell RNA sequencing and bottom-up proteomics. Furthermore, we demonstrate that variants of our method offer superior flexibility and performance over KNN classifiers and can be used in conjunction with other machine learning methods. In summary, ACF is a flexible method that enables missing value tolerant classification with minimal loss of data. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12859-023-05224-0. ## Background With the increasing availability of high quality data across various disciplines, researchers commonly employ data mining techniques such as classification, clustering or regression to answer the research questions under consideration. Classification aims to assign new observations to one (or multiple) classes based on a set of training instances, e.g. assigning a diagnosis (sick/healthy) to a patient. While powerful classifiers have been successfully implemented for commonly studied omic types, such as DNA-methylation [1] or the transcriptome [2], a widespread problem associated with many emerging omics technologies such as proteomics and single-cell RNA sequencing (scRNA-seq) is the strong prevalence of missing values, which hampers the direct applicability of most classification algorithms. The number of missing values is often additionally amplified by the integration of multiple individual datasets which is a common strategy to add statistical power to a study [3]. In order to overcome those hurdles, researchers often either delete all features with missing values (leading to significant loss of information) or use imputation methods that do not generalize well across datasets [4, 5] and that have been demonstrated to introduce false positives and irreproducible differential expression in certain cases [6]. In this paper, we present an approach that relies on pairwise correlations to train tunable machine learning models in a modular fashion. We make use of inter- and intra-class correlations and respectively pairwise deletions [7], resulting in minimal data loss and independence of potentially error-prone imputation. ## Related work Multiple mechanisms contribute to the presence of missing data in omics datasets, such as for instance biologic differences between the samples, technical reasons (e.g. detection thresholds) or limitations of the bioinformatics pipeline (e.g. misidentification of peptides in mass spectrometric data). Based on such mechanisms, Rubin [8] introduced the established discrimination between different types of missing values into MCAR (missing completely at random), MAR (missing at random) and MNAR (missing not at random). As elaborated by Emmanuel et al. [ 7], strategies to handle these missing values can be broadly divided into deletion and imputation. The latter uses the measured data to predict and replace the missing values. Extensive studies have been conducted to evaluate the strengths and weaknesses of various imputation methods on different omics types [6, 9, 10]. Lazar et al. [ 4] conclude that the involved missing value mechanism impacts the performance of imputation on label-free quantitative proteomics data and advocated the development of hybrid strategies that consider the coexistence of different types of missing values. Lately, tools have been developed to select suitable imputation methods in a data driven fashion to tailor the type of imputation to the dataset under consideration [5]. Most recently, Linderman et al. published the ALRA-algorithm, which discriminates between biological and technical zeros for scRNA-seq data and only imputes the latter [11]. Deletion-based approaches can be divided into listwise and pairwise deletion (cf. [ 7]). Listwise deletion refers to deleting all features that contain missing values whereas pairwise deletion reduces each pairwise computation on samples to features that were observed in both. The exact pairwise operations performed depend on the task under consideration. In this paper, we restrict our considerations to pairwise correlations, as they represent a particularly flexible class of metrics (e.g. rank correlation as opposed to correlation for continuous variables) which are by definition in the range \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$[-1,1]$$\end{document}[-1,1]. They provide a way to summarize relationships in an easily interpretable number and are commonly used by the biomedical community. As two representative examples, we consider Pearson correlation and Spearman’s rank correlation. There are only few different classifiers capable of working with correlations in the commonly used vectorial representation among which we focus on the K-Nearest-Neighbor (KNN) classifier and the distribution-based classification (DBC) method. The KNN algorithm [12, 13] is a classical, yet state-of-the-art approach which is capable of classifying observations by using those pairwise correlations (cf. [ 14, 15]). The KNN classifier assigns a class to a test instance by performing a majority vote among the K nearest neighbors. It thereby omits all additional information, such as (potentially meaningful) correlations to instances from other classes. Authors have introduced partitioning strategies, such as KD-tree and Ball-tree, that can accelerate the nearest-neighbor search [16, 17]. Since not all of them are applicable to correlations (eg. Ball-tree requires mathematical distance metrics), we restrict the considerations in this paper to the brute search for nearest neighbors. Distribution-based classification (DBC) is a method introduced by Wei and Li [18] which compares similarity-score distributions within and between classes by means of the Kullback–Leibler (KL) distance. Among other metrics, they applied their method to pairwise correlations and found their approach to perform comparable or better than several other popular machine learning methods. Although KNN and DBC both show excellent performance on some classification tasks, they offer very limited capabilities of being adapted to the data at hand. Our approach provides a modular concept with exchangeable baseline classifiers, each of which may be tuned specifically to the problem under consideration. For instance, a limited overlap of the expressed genes between two specific classes could render the correlation between samples from those classes essentially meaningless. Both KNN and DBC would still consider those correlations as equally important to other class combinations, whereas a RandomForest as baseline classifier would intrinsically assign lower feature importance to those meaningless values during the training process. ## Algorithm In this section we describe the proposed method and compare our concept to the KNN and DBC classifiers. Furthermore, we introduce two modifications of our approach that allow to reduce the execution time and to neglect specific types of bias (e.g. batch effects). The proposed method focuses on three key aspects:Tolerate close to arbitrarily many missing values without relying on imputation. Make use of (potentially discriminative) cross-correlations between classes (eg. instances from class A and B exhibit high mutual correlation, but while instances from A also have a high average correlation with instances from class C, those from class B don’t).Provide tuning options via modular components and parameters, allowing to even exchange the incorporated machine learning models. In order to address all three aspects, we propose to fit tunable machine learning models to the empirical estimates for the average pairwise correlation between samples from each combination of classes (cf. Fig. 1A). We term this approach ACF (average correlations as features).Fig. 1Concepts and results regarding the comparison of correlation-based classifiers on simulated datasets. A Concept sketch of ACF. B Concept sketch of the data-generating process for datasets with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{A}$$\end{document}NA, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{B}$$\end{document}NB, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_{C}$$\end{document}NC samples from each of the three classes A, B, C respectively. C Dependency of the properties of correlation matrices from the data-generating process on the percentage of missing values and the standard deviation per feature. Lines indicate the mean value per combination of classes A, B and C over 20 repetitions. D Dependency of the macro-averaged F1-score of ACF (red), DBC (green), KNN (blue) and KNN with random oversampling (KNN+ROs, orange) on the average relative noise. ACF may incorporate additional predictive covariates (violet). The predictiveness of the covariate alone is indicated in black. Solid lines and shaded areas indicate mean and standard deviations over 10 repetitions. For ACF, we tested multiple baseline classifiers (support-vector-classifier/RandomForest/ridge) and report the scores for the SVC which performed best in most cases. E Dependency of the macro-averaged F1-score (averaged over 10 repetitions) of the considered correlation-based classifiers on class imbalance. F Top: Dependency of the macro-averaged F1-score (averaged over 10 repetitions) of the considered correlation-based classifiers on the size of the training set. Bottom: Mean and standard deviation of the selected number of nearest neighbors for KNN and KNN+ROs. Results are averaged over 10 repetitions. G Mean and standard deviation over 10 repetitions of the runtime per predicted test instance for varying number of training instances. Linestyles indicate the respective algorithms (KNN, DBC, F-ACF and F-DBC) and colors represent the respective number of reference instances per class (10, 20, 30 and naïve (all)). For F-ACF, we chose a support-vector-classifier (kernel = “rbf”, $C = 100$) as baseline classifier. H Visualization of the trade-off between the number of reference instances for F-ACF and the F1-macro score achieved at a fixed noise level (averaged over 10 repetitions). Both ACF (orange) and F-ACF (blue) use a support-vector-machine as baseline classifier This method is particularly appropriate, if we assume the matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\textbf{C}_{Train}$$\end{document}CTrain of all pairwise correlations between training observations to exhibit block structure upon ordering (cf. in “Discussion” section). This means, that the pairwise correlation \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\textbf{C}_{Train}[s_{1}, s_{2}]$$\end{document}CTrain[s1,s2] between two samples \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$s_{1}$$\end{document}s1, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$s_{2}$$\end{document}s2 is solely determined by their respective classes \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_{1}$$\end{document}C1, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_{2}$$\end{document}C2, i.e.1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \textbf{C}_{Train}[s_{1}, s_{2}] = \mu \left(C_{1}, C_{2}\right) + \epsilon, \end{aligned}$$\end{document}CTrain[s1,s2]=μC1,C2+ϵ,where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu (C_{1}, C_{2})$$\end{document}μ(C1,C2) denotes the expected correlation of samples from classes \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_{1}$$\end{document}C1 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C_{2}$$\end{document}C2 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\epsilon$$\end{document}ϵ is a random variable that we assume to be normally distributed. In the most simplistic form, the proposed approach proceeds as follows (cf. Fig. 1A): Compute the average correlations \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \mu (s_{i}, s_{k}), \quad y(s_{k}) = C, \quad k \ne i \end{aligned}$$\end{document}μ(si,sk),y(sk)=C,k≠i of each training sample \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$s_{i}$$\end{document}si to all training observations \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$s_{k}$$\end{document}sk per class C (self-correlations must not contribute, since they do not carry any information and would lead to biased estimates of the mean correlations).Select a suitable classification model (e.g. RandomForest) and train it on the empirical estimates obtained in step 1. Additionally, other relevant covariates may be included, such as age or symptoms of a patient. The underlying classification model is also termed baseline classifier in the following. Common hyperparameter tuning approaches can be used to select and further adapt it to the problem. Compute the average correlations of each test instance to all training instances per class. Use the trained baseline classifier on the estimates obtained in step 3 (and all considered covariates) to predict the classes of the test instances. In contrast to the KNN classifier, ACF intrinsically considers all cross-correlations between classes, without limiting itself to certain elements of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\textbf{C}_{Train}$$\end{document}CTrain. DBC also incorporates cross-correlations but relies on a fixed claiming-scheme and weighted Kullback–Leibler (KL) decision rules. For ACF, the baseline classifier may instead be chosen depending on the data and can be further adapted, e.g. increasing the depth of decision trees or applying regularization. Both ACF and DBC require the computation of all pairwise correlations between training instances as well as between test instances and training instances. This raises important concerns regarding their computational performance, especially when compared to the KNN classifier that does neither require the computation of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\textbf{C}_{Train}$$\end{document}CTrain nor a training step prior to prediction. Due to the computation of all pairwise correlations, the asymptotic time complexities of ACF are2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \mathcal {O}_{Train}^{ACF}&= \mathcal {O}\left(n^{2}_{Train}\right) + \mathcal {O}_{Train}^{Baseline Classifier} \end{aligned}$$\end{document}OTrainACF=OnTrain2+OTrainBaselineClassifier3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \mathcal {O}_{Test}^{ACF}&= \mathcal {O}\left(n_{Train}n_{Test}\right) + \mathcal {O}_{Test}^{Baseline Classifier} \end{aligned}$$\end{document}OTestACF=OnTrainnTest+OTestBaselineClassifierfor training and prediction respectively. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n_{Train}$$\end{document}nTrain and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n_{Test}$$\end{document}nTest denote the number of instances in the training set and test set respectively whereas \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal {O}_{Train/Test}^{Baseline Classifier}$$\end{document}OTrain/TestBaselineClassifier represents the time complexities of the baseline classifier for training and prediction. Optimizing the time complexities of a particular baseline classifier is not within the scope of this paper, as the methodology is intended to work with arbitrary machine learning models. It is however possible to enhance the computational performance of ACF by estimating the average classwise correlations of a training instance using only a randomly selected subset of reference instances per class1 (cf. Additional file 2: Fig. S1, left panel). We term this faster variant of our algorithm F-ACF. This approach is expected to yield coarser estimates of the average correlations, thereby coming with a trade-off in classification performance. Given a fixed number of reference samples and a baseline classifier with suitable time complexity, the prediction time of F-ACF is independent of the number of training instances. This differs from the prediction time complexity of the brute KNN classifier, which is at least linear in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n_{Train}$$\end{document}nTrain due to the necessary \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n_{Train}$$\end{document}nTrain distance computations. This can conceptually not be reduced as it can for ACF.2 Therefore, our method brings additional flexibility in terms of time complexity and is particularly advantageous over the KNN classifier for large training sets. Apart from computational performance, another relevant concern regards the existence of biased values in the correlation matrix. Such biased elements would also bias the obtained average correlations, which is why we propose to omit those biased correlations when performing the averaging (cf. Additional file 2: Fig. S1, right panel). As we assume block-structured correlation matrices (cf. Eq. 1), where the unbiased elements exhibit high redundancy, this does not affect classification performance as long as the number of biased elements remains relatively low. We refer to this modified algorithm as B-ACF.3 A particular weakness of this approach is that the exact location of the biased elements has to be known beforehand. As will be demonstrated in the “Results” section, this holds true for a simple model of batch effects in multiplexed proteomic measurements. ## Implementation All studies have been conducted in Python. ACF, DBC and the KNN classifier are implemented as estimators compatible with current standards for machine learning modules. To ensure the correctness of our implementation, we validated our DBC-implementation against results from the original publication. Although other packages were used as well, the developed software and conducted analyses rely to large extent on the python-packages scikit-learn [19], optuna [20], numpy [21], scipy [22], pandas [23, 24], seaborn [25] and matplotlib [26]. All source code is publicly available at GitHub [27]. ## Consideration of classification performance First, we analyzed the impact of noise present on the correlation matrix on classification performance. For this, we used the procedure described in the “Methods” section (cf. Fig. 1B, C) to generate simulated datasets of three classes with 70, 30 and 50 instances respectively. The first two classes (denoted as A and B in the following) were closely correlated but strongly differed in their correlation to the third class (denoted as C). Additionally, we generated an artificial covariate, which follows a Gaussian distribution with standard deviation 0.015 around the class centers at 0.15, 0.2 and 0.25 for A, B and C respectively. We measure the reliability of class predictions (also referred to as classification performance in this paper) as the macro-averaged \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-score, which is the harmonic mean of the average precision and average recall per class [28]. We report the dependency of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-score on the average relative noise \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma _{rel} = \frac{\sigma }{\mu _{AA}-\mu _{AB}}$$\end{document}σrel=σμAA-μAB, averaged over 10 independent simulations. Here, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma$$\end{document}σ denotes the standard deviation of the noise on the correlation matrix and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _{AA}$$\end{document}μAA, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _{AB}$$\end{document}μAB represent the average pairwise correlations between instances from A, A and A, B respectively. Although Spearman’s correlation works as well, we employ Pearson correlation for the simulation studies presented below, since we don’t expect strong outliers in the data and Pearson correlation will therefore yield unbiased estimates that capture more information than rank-based correlations. As shown in Fig. 1D, the score of the KNN classifier decreased drastically with increasing relative noise. Using random oversampling improved the performance of the KNN classifier by allowing the hyperparameter optimization to select a higher number of nearest neighbors, which resulted in better averaging for the class prediction. For ACF, we tested three different baseline classifiers (support-vector-classifier/RandomForest/ridge) which yielded comparable \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-scores. On average, the support-vector-classifier performed best. Both ACF and DBC maintain high \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-macro scores even for much higher relative noise than the nearest-neighbor based approaches, which is most likely due to fact that they intrinsically consider cross-correlations. ( While the average correlations \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _{AA}$$\end{document}μAA, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _{BB}$$\end{document}μBB, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _{AB}$$\end{document}μAB, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu _{BA}$$\end{document}μBA might be indistinguishable at a relative noise of \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, the discriminative cross-correlation with class C is only obscured at much higher noise, thereby allowing ACF and DBC to still yield reliable class predictions.) The \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-score achieved by ACF exceeds the score of DBC, which we attribute to better adaption of the underlying support-vector-classifier via hyperparameter optimization for ACF instead of a fixed claiming-scheme as for DBC. Furthermore, ACF proved capable of further enhancing classification performance by considering additional covariates, as was demonstrated with the generated artificial covariate. For all considered methods, the standard deviation of the macro-averaged \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-score increased with the relative noise \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma _{rel}$$\end{document}σrel. This is to be expected, since the computed correlations resemble coarser estimates of their true values, potentially moving samples closer to the decision boundaries of the corresponding classifier. At maximum \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma _{rel}$$\end{document}σrel, the DBC classifier, KNN (with and without oversampling) and ACF without artificial covariate exhibited similar standard deviations of their \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-scores (0.030, 0.040, 0.033, 0.035 respectively). Once the artificial covariate, which exhibited a constant standard deviation independent of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma _{rel}$$\end{document}σrel, was included, ACF achieved macro-averaged \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-scores with considerably lower standard deviation than the other methods (0.019). At maximum relative noise, the average \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-scores of each pair of methods were at least 1.50 standard deviations apart, indicating very robust results. Further generated simulation data allows to discuss the effect of class imbalance on the performance of the respective classification approaches under consideration. We generated datasets of 150 instances with an average relative noise of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma _{rel} = 2.9$$\end{document}σrel=2.9. Class C had 50 observations, whereas the remaining samples were split between class A and class B with a varying ratio. As depicted in Fig. 1E, all considered classifiers achieved their highest performance on balanced datasets. The \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-macro score of the KNN classifier decreased drastically with increasing class imbalance. Using the KNN classifier with random oversampling to artificially balance the dataset yields equivalent performance in the balanced scenario, but mitigated the problem to a certain extent in the unbalanced scenarios, by allowing a larger number of nearest neighbors to be selected during hyperparameter optimization. This led to enhanced averaging of the class prediction, which countered the effect of noise to a certain degree. In the most strongly imbalanced scenarios, the average \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-scores of KNN with and without random oversampling differed by at least 1.33 standard deviations. Again, we observed only subtle differences between the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-scores of ACF with different baseline classifiers, but the support-vector-classifier (SVC) performed best. With this SVC as baseline classifier, ACF surpassed the performance of both KNN (with and without random oversampling) and DBC for all class ratios. The latter is of particular interest, since DBC is conceptually very robust to class imbalance, because all classes contribute equally regardless of their relative abundance. The robustness of ACF can be explained by the fact that many classical machine learning models, including the support-vector-classifier, offer balanced class weights as a possible hyperparameter. This allows the hyperparameter optimization procedure to ensure equal contribution of all classes during the training process of the baseline classifier, regardless of their respective number of instances. While the difference between ACF and DBC is relatively small in the perfectly balanced scenario (at least 1.24 standard deviations), their difference is very pronounced in the strongly imbalanced scenarios (at least 3.02 standard deviations), indicating particularly robust results of the ACF approach. Lastly, we expect the nearest-neighbors based approaches to be highly dependent on the total number of instances. To demonstrate this, we generated datasets of various sizes, each exhibiting relative class abundances of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{7}{15}, \frac{3}{15}$$\end{document}715,315 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{5}{15}$$\end{document}515, as well as an average relative noise of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma _{rel} = 2.6$$\end{document}σrel=2.6. Figure 1F reports the macro-averaged \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-score in dependency of the total number of instances in the dataset. Among the different baseline classifiers, ACF performed best in combination with a support-vector-classifier. It is apparent that this combination of SVC and ACF outperformed the KNN classifier (with and without random oversampling) as well as DBC for small datasets. For medium-sized and large datasets, our approach exhibited a close-to-ideal \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-macro score of approximately 1, whereas DBC and the KNN classifier with oversampling slowly converged towards this value. Without oversampling, the KNN classifier showed severely reduced scores, which can be attributed to the low number of considered nearest neighbors that was on average selected in the hyperparameter optimizations. With increasing size of the dataset, the macro-averaged \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-scores of all considered methods exhibited a decreasing standard deviation. This is to be expected, since the higher number of training instances moves new samples away from the decision boundaries of the corresponding classifiers. In the considered scenarios, nearest-neighbor based approaches exhibited scores which were typically well separated from other methods by multiples of their respective standard deviations, while the difference between ACF and DBC was typically not as strongly pronounced. ## Consideration of computational performance In this section we compare the time complexities of the fast F-ACF algorithm, DBC and the (brute) KNN classifier. Furthermore, we compare the macro-averaged \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-score of F-ACF and ACF. To compare the asymptotic time complexities of the respective algorithms, we generated various datasets with 60 test instances and between 100 and 240 training instances. For DBC, we considered both the naïve DBC algorithm as described by Wei et al. [ 18], as well as a similar modification as in F-ACF, where the intra- and interclass distributions are approximated using only a fixed number of reference instances per class. We term this modified variant F-DBC. We selected a support-vector-classifier ($C = 100$, rbf kernel) as baseline classifier for F-ACF and tested F-ACF and F-DBC with 10, 20 and 30 reference instances per class. Figure 1G reports the prediction time per test instance for each considered algorithm, averaged over 10 independent measurements. We expect the computation of pairwise correlations to dominate the runtime. This was confirmed experimentally by the observed linear scaling of KNN and the naïve DBC classifier, as well as the independent scaling for F-ACF and F-DBC. For all numbers of reference instances, F-ACF was faster than the corresponding F-DBC implementation. Furthermore, even without hyperparameter optimization for F-ACF, the average \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-scores of F-ACF were in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$65.3\%$$\end{document}$65.3\%$ of the dataset configurations higher than the scores of the respective F-DBC algorithm (Additional file 2: Fig. S2, left panel). Employing hyperparameter optimization increases this ratio to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$86.7\%$$\end{document}$86.7\%$, but also increases the training time (Additional file 2: Fig. S2, right panel).Fig. 2Characteristics of the considered datasets and approaches. A Class distributions of the considered scRNA-seq datasets. B Mean and standard deviations of the macro-averaged F1-scores of ACF (blue), KNN (orange), DBC (green) and conventional machine learning models (listwise deletion, red) on the three considered scRNA-seq datasets over 10 repetitions. For ACF and listwise deletion, we report the results for the baseline classifier maximizing the reported score. C Representative example of the approximated distributions of the correlations between samples from each pair of classes on the dataset by Xin et al. For the classes PP and delta (top and bottom), close similarities between inter- and intraclass distributions are observed. D Class distributions of the considered dataset from multiplexed proteomics. E Mean and standard deviations of the macro-averaged F1-scores of the considered classifiers on the considered proteomic dataset over 10 repetition. Scores are reported for raw and for batch-corrected (IRS) data. F Representative example of the improved prediction of proteomic subtypes on the dataset by Petralia et al. when considering histopathologic diagnoses. Scores refer to the baseline classifier maximizing the classification performance. G, H Importance of average class-wise correlations for the prediction of each class for the datasets by Xin et al. ( G) and Petralia et al. ( H). Variable importance was measured by the decrease in class-wise F1-score of an unoptimized support-vector-classifier ($C = 100$, kernel = rbf, balanced class weights) when removing the considered correlation The observed independent scaling as well as the high \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-scores compared to F-DBC demonstrate that F-ACF might be particularly suited for classification tasks with large training sets. Reducing the number n of reference instances per class decreases prediction time, but will generally yield coarser estimates for the average correlations, thereby leading to lower classification performance of F-ACF. Figure 1H illustrates the trade-off between the number of reference instances per class and the achieved \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-macro score at various noise levels. Furthermore, we also report the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-macro score of the ACF algorithm at the same relative noise. For comparability, both algorithms used a support-vector-classifier as baseline classifier. At low relative noise, a small number of considered instances per class was sufficient to yield reliable estimates of the average correlations (and therefore a high \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-macro score of F-ACF). Higher relative noise however required a larger number of references to yield comparable classification performance. Unsurprisingly, the highest macro-averaged \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-score was obtained using all accessible instances per class (ACF). In an application, the number of reference instances required by F-ACF to achieve good classification performance would be determined automatically using a hyperparameter optimization library, such as optuna [20]. ## Comparison with KNN, DBC and conventional machine learning methods on biologic datasets The results from the previous section were based on simulated datasets that were engineered to exhibit discriminative cross-correlations. However, the benefit of applying our model has still to be demonstrated on real-world datasets. In this section, we apply our approach4 to datasets from scRNA-seq and proteomics and validate it against KNN, DBC and established, conventional machine learning models. For the latter, we consider the models that were used as baseline classifiers for ACF and directly apply them to the gene expression data, handling the missing values using listwise deletion. We considered datasets from three scRNA-seq experiments by Baron et al. [ 29], Xin et al. [ 30] and 10XGenomics [31]. The respective class distributions of the datasets are schematically summarized in Fig. 2A. Further information on the datasets can be found in Additional file 2. Figure 2B reports the respective macro-averaged \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-scores obtained on the three scRNA-seq datasets. Since listwise deletion removed all genes on the dataset by Baron et al, the reported scores of the conventional machine learning models on that dataset were determined by random class assignment. The proposed approach, ACF, strongly outperformed the other methods with significant differences on all three datasets, regardless of the selected baseline classifier (cf. Additional file 1 and Additional file 2: Table S6). The differences between different baseline classifiers for ACF were not always significant, although a support-vector-classifier generally appeared to be a good choice. The three correlation-based approaches, ACF, KNN and DBC, outperformed the combination of conventional machine learning methods with listwise deletion on two out of three datasets. This demonstrates that our initial motivation of avoiding data loss was highly reasonable in the context of imputation-free classification. Although DBC generally had better \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-scores than KNN, the difference between these two methods was rather small. We attribute this to the indistinguishability of inter- and intra-class distributions for many classes on the datasets (cf. Fig. 2C). This reduces the probability for true positives in the claiming-scheme of DBC and makes false positives more likely at the same time. This explanation is supported by the fact that we also observe reduced precision and recall of DBC for these classes (cf. Additional file 2: Table S8 for an example). The proteomic datasets differ from the scRNA-seq datasets in multiple ways: Methods such as isobaric labelling of peptides and other technologies can increase the number of identified peptides so that the missing value problem is not as prevalent as in scRNA-seq data. Furthermore, combining several multiplexed experiments introduces a bias (the so called batch effect) among instances from different experiments. The existence of such biases poses a major challenge for biologic analysis as well as classification. Common techniques to correct batch effects include the usage of internal reference samples (IRS) between experiments [32] and the application of batch effect correcting algorithms such as ComBat [33]. In this study, we employ internal references for batch correction. We consider two proteomic datasets by Petralia et al. [ 34] and Krug et al. [ 35]. Their respective class distributions are visualized in Fig. 2D. Further details on the datasets are provided in Additional file 2. Figure 2E reports the respective macro-averaged \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-scores achieved by the individual classification models. For the batch-effect corrected data (Fig. 2E, 1st and 3rd column), the combination of conventional machine learning models with listwise deletion performed best, closely followed by ACF which yields significantly higher classification performance than the other correlation-based approaches (cf. Additional file 1 and Additional file 2: Table S7). For the uncorrected data (Fig. 2E, 2nd and 4th column), we employ B-ACF and a similarly modified version of DBC. For this, we modeled the batch effect to bias only the correlations of samples from the same batch (cf. Additional file 2: Fig. S3), which is a simplistic approximation, but proves to be powerful by enhancing classification performance. This flexibility is however not offered for the KNN classifier that only works on the entire (unmasked) correlation matrix. On this unadjusted data, B-ACF outperformed both KNN and the modified DBC with significant differences (cf. Additional file 2: Table S1). This supports our simplistic model of the batch effect and demonstrates the flexibility of the classification approach presented here. The modularity of ACF even allows to integrate deep-learning based methods, such as a multi-layer perceptron (MLP) as baseline classifier. As proof-of-concept, we conducted experiments on one exemplary scRNA-seq dataset and proteomic dataset each, where we tested a MLP as baseline classifier and compared the results for the deep neural network with the conventional baseline classifiers discussed before (cf. Additional file 2: Fig. S4). The results indicate that using a deep neural network as baseline classifier may offer a slight improvement over the previously discussed conventional machine learning methods. While KNN and DBC intrinsically rely on the pairwise correlations exclusively, ACF offers the flexibility to incorporate further covariates into the classification process. This advantage of ACF is demonstrated by the observation of improved \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-scores, when considering histopathologic diagnoses as covariate for the data by Petralia et al (Fig. 2G). To estimate the variable importance of each average correlation for the prediction of individual classes, we selected a support-vector-classifier with typical hyperparameters ($C = 100$, balanced class weights, rbf-kernel) as baseline classifier for ACF and employed repeated stratified cross-validation. We measured the decrease of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-score for each class, when individual average correlations were not passed to the baseline classifier. We observed discriminative cross-correlations (variable importance \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 for the average correlations to samples from other classes) on all considered datasets from each omic type (cf. Fig. 2G, H). This highlights the importance of considering all correlations when using absolute correlation values (as in ACF and DBC) instead of relative values, such as for KNN. ## Discussion The aim of this study is to explore the use of pairwise correlations for classification based on molecular data. This is motivated by the widespread use of correlations in the biomedical community as well as the fact, that they allow to summarize relationships in an easily interpretable number. Previous work on correlation-based classification is scarce, but researchers have used the K-Nearest-Neighbor (KNN) classifier and DBC (distribution based classification) [14, 15, 18]. With ACF, we present a novel method for correlation-based classification, which can be flexibly adapted to a large number of settings. By using pairwise metrics, it works in an imputation-free fashion, whilst minimizing data loss.5 While the KNN classifier only considers the k highest correlations, both ACF and DBC intrinsically consider cross-correlations. DBC however relies on a fixed claiming-scheme, whereas ACF offers the flexibility of choosing and adapting tunable classification models to the data under consideration. This makes ACF particularly suitable for the application to datasets with large portions of missing values, such as from dataset integration [3]. Candidate problems include, but are not limited to, multi-omic datasets as well as datasets assembled from multiple laboratories, leading to various kinds of missing values. The computation of pairwise correlations relies on pairwise deletion, which is rarely used compared to listwise deletion and imputation. Based on our results, we see great potential in both the application and future research on approaches based on pairwise deletion (see “Conclusion and Outlook”). ACF makes use of average correlations, whereas DBC employs the Kullback–Leibler distance between distributions, thereby capturing further potential information, such as skewness. However, we observed significantly reduced scores when combining ACF with Kullback-Leibler distances. We attribute this to the unfavorable divergence of Kullback–Leibler distance, which makes strong outliers more likely. Therefore, we focused on average correlations only, although our implementation allows the user to select other metrics such as median correlations. The simulation studies we conducted show that the proposed method yielded much higher macro-average \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-scores than the KNN classifier for noisy, small or imbalanced datasets and also performed comparable or better than the DBC method. We tested different baseline classifiers for ACF which all performed comparably well on the simulated datasets, but in the majority of cases considered in this study, a support-vector-classifier yielded slightly more favorable scores than the other methods. Additional proof-of-concept experiments indicate that using deep-learning methods (e.g. MLPs) as baseline classifiers may yield slight improvements over conventional machine learning methods. Whilst this manuscript focuses on presenting the method and applying three conventional machine learning methods as baseline classifier, an optimal baseline classifier (e.g. MLP) may in practice easily be selected as part of the hyperparameter optimization procedure. The process used for generating the simulated datasets (cf. “ Methods” section) was developed to allow precise control over the block-structure and the noise of the correlation matrix and is based on the introduction of missing values to instances that are normally distributed around correlated class centers. We argue, that this procedure is highly reasonable in a biological context: Firstly, the high dimensionality as well as the high number of missing values is common in many datasets, e.g. in scRNA-seq experiments, cf. [ 37]. Secondly, assuming instances to be normally distributed around a class-specific center is reasonable, as many relevant sources of variation, e.g. measurement error, can be approximated to be normal. Lastly, correlation matrices from biologic datasets commonly exhibit block-structure (cf. [ 38] for an example), in which some of the cross-correlations may allow class discrimination. Our findings suggest that this phenomenon occurs commonly in real-world datasets, i.e. from scRNA-seq, which in turn shows that it is reasonable for the data-generating process to generate datasets with such discriminative cross-correlations. We also demonstrated that ACF offers the flexibility to be modified in such a way that the time complexity for prediction is independent of the number of training instances (F-ACF), whereas it scales approximately linearly for the KNN classifier. The same modification is possible for DBC, but yielded both less efficient as well as less accurate predictions than F-ACF. We showed on data from three scRNA-seq experiments, that our approach significantly outperformed both KNN and DBC as well as listwise deletion combined with several conventional classifiers (RandomForest, SVM, Ridge). On datasets from proteomics, ACF yielded better \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-macro scores than the other correlation-based classifiers, especially when incorporating a simplistic model for batch effects. In summary, this work explores and compares different approaches to correlation-based classification. The proposed approach, ACF, offers peculiar advantages, such as tolerance to missing values, the consideration of cross-correlations as well as potential covariates, and the capability of being adapted to the considerd dataset by means of hyperparameter optimization. Our results demonstrate superior classification performance of ACF over established correlation-based techniques in extensive simulation studies, as well as on biologic datasets from scRNA-seq and proteomics. The assumption of block-structured correlation matrices as well as the dependence on approximate average correlations constitute two important limitations to ACF: While the former is required to establish average correlations as meaningful metrics for classification, the latter implies that the correlation between samples from two classes may not vary too strongly relative to the number of samples used for averaging, in order to obtain meaningful averages. We found the considered real-world datasets to satisfy both requirements. It is important to note that this paper focused entirely on the comparison of ACF, DBC and KNN in an imputation-free setting. We explicitly excluded imputation methods from the considerations here, since we were concerned about their generally weak generalization across different omics types and datasets as well as their applicability in presence of different types of missing values [4–6]. ## Conclusion and outlook We presented our novel correlation-based classification approach ACF. The particular advantage of ACF lies in the combination of tolerance to missing values, consideration of cross-correlations and the capability of providing tuning options via modular components and parameters. In simulation studies, we found our approach to work particularly well when considering small, imbalanced or noisy datasets, which are challenging for most algorithms. We observed statistically significant improvements over KNN and DBC on experimental data from two representative omics-technologies, namely scRNA-seq and proteomics. Furthermore, we demonstrated that ACF offers high flexibility with respect to time complexity and modeling of certain biases (e.g. batch effects), thereby enabling problem-specific adaptions to various applications. Directions of further research include the evaluation of ACF on multi-omics datasets as well as the comparison of ACF with deep learning models. For the latter, a particularly interesting approach might be to adapt the recently published DeepOmicNet architecture to classification tasks, which would allow for highly efficient training due to the usage of grouped bottleneck structures and skip connections [39]. Other architectures of interest include, but are not limited to, the multi-layer perceptron and deep-belief networks [40]. ## The ACF method The proposed method (Average Correlations as Features, ACF) aims to provide tolerance to missing values, consideration of cross-correlations, as well as the capability of being flexibly adapted to the data at hand. This is achieved by fitting tunable machine learning models to empirical estimates for the average pairwise correlations between samples from each combination of classes. The procedure starts by computing the average correlations of each training sample to all other training observations per class. ( Depending on the considered problem, other empirical metrics, such as the median, may be used as well. In this study however, we focus on average correlations exclusively.) In the next step, a machine learning model (referred to as baseline-classifier) is trained to predict class labels based on the previously obtained averages and potential other covariates, such as the age of a patient (cf. Fig. 1, panel A). For the classification of a test sample, the trained baseline classifier is provided with all potential covariates as well as with the average correlations between the test sample and the previously considered training samples per class. The computational performance of ACF may be enhanced by estimating the average correlation based on less training samples. This approach is referred to as F-ACF throughout this manuscript. This method reduces the required number of computations, but may result in coarser estimates of the average correlations which can reduce classification performance. If specific correlations between samples are known to be biased (e.g. due to batch-effects), these correlations may be masked out during the computation of the empirical averages. This modified approach is referred to as B-ACF throughout the manuscript. For a motivation of the method, as well as further details, we refer the reader to the “Algorithm” section. ## Data-generating process We aimed to generate data with block-structured correlation matrices and discriminative cross-correlations with sufficient control overthe overall block-structure of the correlation matrix,the standard deviation of the (Gaussian) noise \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma$$\end{document}σ on the correlation matrixas well as the number of classes, the number of instances per class and the size of the dataset. Meeting these requirements, we developed a data-generating process (DGP) that is based on the introduction of missing values to high-dimensional data points which are normally distributed around points that were chosen to exhibit a specified correlation matrix (see Fig. 1B). This DGP, which was used to generate datasets for all simulation studies in this paper, proceeds as follows: We choose the dataset under consideration to exhibit 10,000 features and to consist of instances from three classes. This corresponds to the minimum number of classes that can exhibit a discriminative cross-correlation. The centers of those classes are expected to be correlated with a correlation matrix of4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} \textbf{C}_{Centers} = \begin{pmatrix}1.0&{}\quad 0.9&{}\quad 0.6\\ 0.9&{}\quad 1.0&{}\quad 0.8\\ 0.6&{}\quad 0.8&{}\quad 1.0\end{pmatrix} \end{aligned}$$\end{document}CCenters=1.00.90.60.91.00.80.60.81.0Throughout this paper we focus on this specific matrix, since it provides a minimal example of discriminative cross-correlations. ( The centers for the first two classes are closely correlated but strongly differ in their correlation to the third class.) Considering variations of the individual correlations in Eq. [ 4] might offer the opportunity for further characterization in future works, but is clearly not in the scope of this paper, since they can potentially affect the noise distribution on the final correlation matrix and would therefore need to be chosen very carefully (see elaboration below). By applying a Cholesky-decomposition [41] to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\textbf{C}_{Centers}$$\end{document}CCenters and multiplying the resulting matrix with a suitably shaped random matrix drawn from a multivariate standard distribution, we obtain centers with approximately the specified correlation matrix. The final observations are then drawn from multivariate Gaussian distributions around each of the previously created centers, where the standard deviation of all Gaussian distributions is controlled via the parameter \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma _{Feature}$$\end{document}σFeature. The number of samples that are drawn per distribution determines the number of instances per class and correspondingly also the size of the dataset. Finally, we introduce a fixed percentage of completely randomly missing values to each observation. The correlation matrix of the resulting dataset follows the block-structure specified by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\textbf{C}_{Centers}$$\end{document}CCenters. We observed that the parameter \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma _{Feature}$$\end{document}σFeature introduces a scaling factor to the correlation matrix (cf. Fig. 1C, left center), which is to be expected, since \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma _{Feature}$$\end{document}σFeature determines the standard deviation per feature, thereby reducing the correlation of each pair of instances. Furthermore, the percentage of missing values determines the standard deviation of the noise (denoted \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma$$\end{document}σ in the following) on the correlation matrix (cf. Fig. 1C, top right). This is understandable, since missing values cause the pairwise correlation of each pair of samples to be computed using only a subset of features, thereby introducing variance to the individual correlations.6 Since each pairwise correlation must be in the interval \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$[-1,1]$$\end{document}[-1,1], we were initially concerned about our assumption that the noise on the correlation matrix is Gaussian (which is an unbounded distribution). Using the test from D’Agostino and Pearson [42] we found however, that suitably low average correlations allowed for very high standard deviations \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma$$\end{document}σ without significant deviation from normality (cf. Fig. 1C, bottom left and bottom right). ( Such low values can either be achieved by variation of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma _{Feature}$$\end{document}σFeature or \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\textbf{C}_{Centers}$$\end{document}CCenters.) Furthermore, we found that 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}$$\sigma$$\end{document}σ of the individual blocks of the correlation matrix were all equal and increased non-linearly with the percentage of missing values. Finally, we could show empirically, that the parameter \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma _{Feature}$$\end{document}σFeature did not impose any changes on the block-wise standard deviations as long as the values in the correlation matrix were low enough so that the range of the correlation values \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$[-1,1]$$\end{document}[-1,1] did not conflict with normality of the noise (cf. Fig. 1C, top left). If \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma _{Feature}$$\end{document}σFeature was very low, the elements of the correlation matrix became close to the boundaries for correlations, hence the noise could not be symmetric anymore. Throughout this paper, we chose \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sigma _{Feature}=2.0$$\end{document}σFeature=2.0 which allows a suitable range of missing values rates without violating the assumption of normality. ## Machine learning methodology We measure the reliability of class predictions (also referred to as classification performance in this paper) as the macro-averaged \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-score, which is the harmonic mean of the average precision and average recall per class [28]. Since all classes contribute equally to the averaging, it is particularly insensitive to the class imbalance that often occurs in biologic datasets, including the ones considered in this paper. Especially the considered proteomic datasets are small with only as few as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$<20$$\end{document}<20 samples per class, rendering it infeasible to hold out a sufficiently large, representative test set for accurate evaluation of the considered methods. We therefore employ stratified, 10-fold cross-validation to measure the macro-averaged \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-score on the considered datasets. Using this approach, the dataset is first randomly split into k disjoint sets of samples (throughout the manuscript, it is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$$k = 10$$$\end{document}$k = 10$). Since the datasets under consideration exhibit considerable class-imbalance, each of the sets is constructed to approximately preserve class frequencies, which helps to reduce experimental variance [43]. In a round-robin like fashion, one set is then held out as test set for evaluation, while the remaining \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k-1$$\end{document}k-1 sets are used for the training procedure of the corresponding classifier. In particular, the test set is not used during the training or validation procedure and the evaluated classifiers are fully independent. While DBC resembles a parameter-free algorithm and may be trained directly on the k − 1 sets, both KNN and ACF (as well as the underlying baseline classifiers themselves) allow for hyperparameter optimization during the training procedure. For the hyperparameter optimization, we sample candidate parameter values using a tree-structured Parzen estimator [44, 45]. For each parameter combination, 10 validation sets of $10\%$ size are independently and randomly drawn from the union of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k-1$$\end{document}k-1 splits. For each of the validation sets, a classifier with the considered parameter combination is trained on the remaining data. The parameter combination is then assigned the mean macro-averaged F1-score of the classifiers on the respective validation sets. After 60 iterations of hyperparameter optimization, the best parameter combination is selected and used to train a new classifier on the full \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$k-1$$\end{document}k-1 splits. Finally, this new classifier is evaluated on the respective held-out test set. This procedure is repeated for each of the k sets. To account for experimental variations, e.g. during hyperparameter optimization, the entire cross-validation procedure is repeated 10 times, resulting in a mean macro-averaged F1 test-score that we report per dataset. For ACF and its variants, we optimize various hyperparameters for different baseline classifiers: For a support-vector-classifier, we optimize the kernel (linear/rbf), the regularization parameter \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C\in [5 \times 10^{-3}, 5 \times 10^2]$$\end{document}C∈[5×10-3,5×102], the kernel coefficient \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\gamma$$\end{document}γ for rbf-kernels (scale/auto) and the class weights (balanced/None). For a RandomForest, we choose the number of decision-trees to be between 80 and 300, their depth between 2 and 40, the number of features within the entire possible interval and the class weight to be either balanced or None. The Ridge-Classifier is optimized using the regularization strength \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha \in [10^{-3}, 10^{4}]$$\end{document}α∈[10-3,104] and the class weight (balanced/None). In our proof-of-concept experiments employing a multi-layer perceptron (MLP) as baseline classifier, we set the number of epochs to 400 and optimize the initial learning rate \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\lambda \in [10^{-4},10^{-2}]$$\end{document}λ∈[10-4,10-2], the learning rate scheduling (constant/adaptive), the activation function (tanh/ReLU/logistic), the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_{2}$$\end{document}L2 regularization strength \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha \in [10^{-4},10^{-1}]$$\end{document}α∈[10-4,10-1] and the number of hidden layers between 1 and 40. When reporting the respective \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$F_{1}$$\end{document}F1-scores of the conventional machine learning methods (SVC/RandomForest/Ridge, without ACF), we optimize the same set of hyperparameters as above. All other parameters, which are not optimized, are set to their default values in the scikit-learn library [19]. For the KNN classifier, we include the number \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$K \in [1, n_{Train}]$$\end{document}K∈[1,nTrain] of considered nearest neighbors in the optimization process, as well as their weights (uniform/distance-based). DBC is parameter-free and does not require any hyperparameter optimization. We use a corrected right-tailed paired t-test for pairwise comparison of the classification performance of all considered models. A Bonferroni-correction is employed to correct for multiple testing (cf. [ 46]). ## Supplementary Information Additional file 1. Table with p-values for pairwise comparisons between the tested classification approaches on the 5 biologic datasets. Additional file 2. Supplementary Information. ## References 1. Capper D. DNA methylation-based classification of central nervous system tumours. Nature (2018.0) 555 469-474. DOI: 10.1038/nature26000 2. Rathi KS. A transcriptome-based classifier to determine molecular subtypes in medulloblastoma. PLOS Comput Biol (2020.0) 16 1008263. DOI: 10.1371/journal.pcbi.1008263 3. Voß H, Schlumbohm S, Barwikowski P, Wurlitzer M, Dottermusch M, Neumann P, Schlüter H, Neumann JE, Krisp C. HarmonizR enables data harmonization across independent proteomic datasets with appropriate handling of missing values. Nat Commun (2022.0) 13 3523. DOI: 10.1038/s41467-022-31007-x 4. Lazar C. Accounting for the multiple natures of missing values in label-free quantitative proteomics data sets to compare imputation strategies. J Proteome Res (2016.0) 15 1116-1125. DOI: 10.1021/acs.jproteome.5b00981 5. Egert J. DIMA: data-driven selection of an imputation algorithm. J Proteome Res (2021.0) 20 3489-3496. DOI: 10.1021/acs.jproteome.1c00119 6. Andrews TS, Hemberg M. False signals induced by single-cell imputation [version 2; peer review: 4 approved]. F1000Research (2019.0) 7 1740. DOI: 10.12688/f1000research.16613.2 7. Emmanuel T. A survey on missing data in machine learning. J Big Data (2021.0) 8 1-37. DOI: 10.1186/s40537-021-00516-9 8. Rubin DB. Inference and missing data. Biometrika (1976.0) 63 581-592. DOI: 10.1093/biomet/63.3.581 9. Hou W. A systematic evaluation of single-cell RNA-sequencing imputation methods. Genome Biol (2020.0) 21 1-30. DOI: 10.1186/s13059-020-02132-x 10. Jin L. A comparative study of evaluating missing value imputation methods in label-free proteomics. Sci Rep (2021.0) 11 1760. DOI: 10.1038/s41598-021-81279-4 11. Linderman GC. Zero-preserving imputation of single-cell RNA-seq data. Nat Commun (2022.0) 13 192. DOI: 10.1038/s41467-021-27729-z 12. Fix E, Hodges JL. Discriminatory analysis. Nonparametric discrimination: consistency properties. Int Stat Rev Rev Int Stati (1989.0) 57 238. DOI: 10.2307/1403797 13. Cover T, Hart P. Nearest neighbor pattern classification. IEEE Trans Inf Theory (1967.0) 13 21-27. DOI: 10.1109/tit.1967.1053964 14. Alfeilat HAA. Effects of distance measure choice on k-nearest neighbor classifier performance: a review. Big Data (2019.0) 7 221-248. DOI: 10.1089/big.2018.0175 15. 15.Chomboon K, et al. An empirical study of distance metrics for k-nearest neighbor algorithm. In: The Proceedings of the 2nd international conference on industrial application engineering 2015. The Institute of Industrial Applications Engineers; 2015. 10.12792/iciae2015.051. 16. Bentley JL. Multidimensional binary search trees used for associative searching. Commun ACM (1975.0) 18 509-517. DOI: 10.1145/361002.361007 17. 17.Omohundro SM. Five Balltree construction algorithms. Technical report. International Computer Science InstituteBerkeley; 1989. 18. Wei X, Li K-C. Exploring the within- and between-class correlation distributions for tumor classification. Proc Natl Acad Sci (2010.0) 107 6737-6742. DOI: 10.1073/pnas.0910140107 19. Pedregosa F. Scikit-learn: machine learning in python. J Mach Learn Res (2011.0) 12 2825-2830 20. 20.Akiba T, et al. Optuna: A next-generation hyperparameter optimization framework. In: Proceedings of the 25rd ACM SIGKDD international conference on knowledge discovery and data mining. 2019. 21. Harris CR. Array programming with NumPy. Nature (2020.0) 585 357-362. DOI: 10.1038/s41586-020-2649-2 22. 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 23. 23.Pandas development team T. Pandas-dev/pandas: Pandas. 10.5281/zenodo.3509134. 24. 24.McKinney W. Data structures for statistical computing in Python. In: van der Walt S, Millman J (eds) Proceedings of the 9th Python in Science Conference. 2010, pp. 56–61. 10.25080/Majora-92bf1922-00a. 25. Waskom ML. Seaborn: statistical data visualization. J Open Source Softw (2021.0) 6 3021. DOI: 10.21105/joss.03021 26. Hunter JD. Matplotlib: a 2d graphics environment. Comput Sci Eng (2007.0) 9 90-95. DOI: 10.1109/MCSE.2007.55 27. 27.Schumann Y. ACF source code. GitHub Repository. https://github.com/HSU-HPC/ACF. 28. 28.Grandini M, et al. Metrics for multi-class classification: an overview. 2020. arXiv:2008.05756. 29. Baron M. A single-cell transcriptomic map of the human and mouse pancreas reveals inter- and intra-cell population structure. Cell Syst (2016.0) 3 346-3604. DOI: 10.1016/j.cels.2016.08.011 30. Xin Y. RNA sequencing of single human islet cells reveals type 2 diabetes genes. Cell Metab (2016.0) 24 608-615. DOI: 10.1016/j.cmet.2016.08.018 31. 31.10XGenomics: Single cell gene expression dataset by cell ranger 1.1.0. licensed under creative commons attribution license. 2016. https://support.10xgenomics.com/single-cell-gene-expression/datasets/1.1.0/pbmc3k?. 32. Plubell DL. Extended multiplexing of tandem mass tags (TMT) labeling reveals age and high fat diet specific proteome changes in mouse epididymal adipose tissue. Mol Cell Proteomics (2017.0) 16 873-890. DOI: 10.1074/mcp.m116.065524 33. Johnson WE. Adjusting batch effects in microarray expression data using empirical bayes methods. Biostatistics (2006.0) 8 118-127. DOI: 10.1093/biostatistics/kxj037 34. Petralia F. Integrated proteogenomic characterization across major histological types of pediatric brain cancer. Cell (2020.0) 183 1962-198531. DOI: 10.1016/j.cell.2020.10.044 35. Krug K. Proteogenomic landscape of breast cancer tumorigenesis and targeted therapy. Cell (2020.0) 183 1436-145631. DOI: 10.1016/j.cell.2020.10.036 36. Kim J-O, Curry J. The treatment of missing data in multivariate analysis. Sociol Methods Res (1977.0) 6 215-240. DOI: 10.1177/004912417700600206 37. 37.Jiang R, et al. Statistics or biology: the zero-inflation controversy about scRNA-seq data. 2022. 10.1101/2020.12.28.424633. 38. Mieldzioc A. Identification of block-structured covariance matrix on an example of metabolomic data. Separations (2021.0) 8 205. DOI: 10.3390/separations8110205 39. Gonçalves E, Poulos RC, Cai Z, Barthorpe S, Manda SS, Lucas N, Beck A, Bucio-Noble D, Dausmann M, Hall C, Hecker M, Koh J, Lightfoot H, Mahboob S, Mali I, Morris J, Richardson L, Seneviratne AJ, Shepherd R, Sykes E, Thomas F, Valentini S, Williams SG, Wu Y, Xavier D, MacKenzie KL, Hains PG, Tully B, Robinson PJ, Zhong Q, Garnett MJ, Reddel RR. Pan-cancer proteomic map of 949 human cell lines. Cancer Cell (2022.0) 40 835-8498. DOI: 10.1016/j.ccell.2022.06.010 40. Zhang Z, Zhao Y, Liao X, Shi W, Li K, Zou Q, Peng S. Deep learning in omics: a survey and guideline. Brief Funct Genom (2018.0) 18 41-57. DOI: 10.1093/bfgp/ely030 41. Benoit E. Note sur une méthode de résolution des équations normales provenant de l’application de la méthode des moindres carrés a un système d’équations linéaires en nombre inférieur a celui des inconnues–application de la méthode a la résolution d’un système defini d’équations linéaires. BullGéod (1924.0) 2 67-77. DOI: 10.1007/bf03031308 42. D’Agostino R, Pearson ES. Tests for departure from normality. Empirical results for the distributions of b2 and. Biometrika (1973.0) 60 613-622. DOI: 10.1093/biomet/60.3.613 43. Forman G, Scholz M. Apples-to-apples in cross-validation studies. ACM SIGKDD Explor Newsl (2010.0) 12 49-57. DOI: 10.1145/1882471.1882479 44. 44.Ozaki Y, et al. Multiobjective tree-structured Parzen estimator for computationally expensive optimization problems. In: Proceedings of the 2020 genetic and evolutionary computation conference. ACM; 2020. 10.1145/3377930.3389817 45. 45.Bergstra J, et al. Algorithms for hyper-parameter optimization. In: Inc CA (ed) Proceedings of the 24th international conference on neural information processing systems. 2011. 46. Kononenko I, Kukar M. Machine learning and data mining: introduction to principles and algorithms (2007.0)
--- title: 'Using multivariate nonlinear mixed-effects model to investigate factors influencing symptom improvement after high tibial osteotomy in combination with bone marrow concentrate injection for medial compartment knee osteoarthritis: a prospective, open-label study' authors: - Hsiao-Yi Cheng - Chun-Wei Liang - Chen-Lun Chu - Hao-Wei Hsu - Sheng-Mou Hou - Kao-Shang Shih journal: BMC Musculoskeletal Disorders year: 2023 pmcid: PMC10026441 doi: 10.1186/s12891-023-06314-z license: CC BY 4.0 --- # Using multivariate nonlinear mixed-effects model to investigate factors influencing symptom improvement after high tibial osteotomy in combination with bone marrow concentrate injection for medial compartment knee osteoarthritis: a prospective, open-label study ## Abstract ### Purpose To investigate the effects of various demographic, structural, radiographic, and clinical factors on the prognosis of patients with medial compartmental knee osteoarthritis with varus deformity undergoing medial opening wedge high tibial osteotomy (HTO) in combination with bone marrow concentrate (BMC) injection. ### Methods In this prospective study, 20 patients underwent medial opening wedge HTO in combination with BMC injection with 12 months of follow-up. The structural and radiographic outcomes were evaluated by femorotibial angle and posterior tibial slope angle. The clinical outcomes were evaluated by visual analogue scale (VAS), Western Ontario and McMaster Universities Arthritis Index (WOMAC), and The Knee injury and Osteoarthritis Outcome Score (KOOS). Multivariate nonlinear mixed-effects models with asymptotic regressions were used to model the trajectory of symptom improvement. ### Results Medial opening wedge HTO in combination with BMC corrected the malalignment of the knee and led to significant symptom relief. The improvement of clinical symptoms reached a plateau 6 months after the surgery. Greater symptom severity at baseline and lower Kellgren-Lawrance (KL) grades were correlated with better post-operative clinical outcomes. Body-Mass-Index (BMI), femorotibial angle, age, and sex may also play a role in influencing the extent of symptom relief. ### Conclusion Symptom severity at baseline is important for prognosis prediction. In clinical practice, we suggest that the evaluation of clinical features and functional status of the patients be more emphasised. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12891-023-06314-z. ## Introduction Knee osteoarthritis is one of the leading causes of knee pain and dysfunction [1, 2]. Because of the ageing population and the prevalence of obesity, the increase in global incidence rate of knee osteoarthritis has led to severe physical, psychological, and economic burden on patients’ families and the whole society [3–6]. According to the current European Society for Clinical and Economic Aspects of Osteoporosis, Osteoarthritis and Musculoskeletal Diseases (ESCEO) guideline, the Osteoarthritis Research Society International (OARSI) guideline, and American College of Rheumatology/Arthritis Foundation (ACR/AF) guideline in 2019, non-pharmacological treatments, including exercise, education, and weight loss (if overweight), should be initially prescribed [7–9]. Pharmacological agents such as non-steroidal anti-inflammatory drugs, intra-articular corticosteroids, and hyaluronic acids are subsequently added [7–9]. For patients with symptoms that can not be alleviated by conservative treatments, surgical intervention is further considered [10]. Among the various surgical interventions, total knee arthroplasty, unicompartmental knee arthroplasty, high tibial osteotomy (HTO), and distal femoral osteotomy are the most common types. For patients who want to preserve the knee joints and whose symptoms are caused by lower limb malalignment, HTO is a major trend [11–13]. Among the various HTO techniques, medial opening wedge HTO, lateral opening wedge HTO, medial closing wedge HTO, and lateral closing wedge HTO are the most common subtypes [13]. For medial compartment knee osteoarthritis with varus deformity, medial opening wedge HTO alleviates the symptoms by correcting the malalignment and redistributing the body weight loading [14, 15]. However, HTO alone does not aim to repair the structural damage of the knee. Therefore, orthobiologics, including platelet-rich plasma, bone marrow concentrate (BMC), and adipose-derived mesenchymal stem cells, have been applied for soft tissue regeneration and for their potential role in immunomodulation [16–18]. For medial opening wedge HTO, several studies have investigated the association of baseline demographic and radiographic features with clinical outcomes [19–23]. Although a few clinical studies and systematic reviews have investigated the efficacy of the combination of HTO and orthobiologics [24–26], few have reported and evaluated various structural, radiographic, and clinical factors that influenced the prognosis [27]. In our study, patients with medial compartment knee osteoarthritis were recruited. Medial opening wedge HTO was performed and BMC were postoperatively injected. Various demographic, structural, radiographic, and clinical factors that could potentially affect the outcomes were investigated to provide a more profound understanding of the efficacy of the combination therapy of medial opening wedge HTO and BMC. ## Patient enrollment This prospective, open-label study was performed in line with the principles of the Declaration of Helsinki. The protocol was approved by the institutional review board (IRB) of Shin Kong Wu Ho-Su Memorial Hospital (Taipei, Taiwan) (IRB number: 20181006R). Prior to the recruitment of the study, the informed consent was obtained from all participants. We prospectively collected the data of 20 patients who underwent HTO in combination with BMC injection between June 2019 and May 2021. The inclusion criteria were as follows: [1] medial knee pain that could not be alleviated by non-pharmacological or pharmacological treatments; [2] radiographs showing moderate to severe medial compartment knee osteoarthritis with Kellgren-Lawrance (KL) grade II–IV; [3] patients with tibiofemoral angle less than valgus 5° or with varus deformity measured by standing anteroposterior radiographs; and [4] age between 20 and 70 years old. The exclusion criteria were as follows: [1] history of knee surgery; [2] history of stem cell transplantation in the knee; [3] other pathological diseases including rheumatoid arthritis, active knee infections, haemophilia, chronic anterior or posterior cruciate ligament instability; [4] participants with severe obesity (body mass index [BMI] ≥ 35); [5] active neuromuscular injury; [6] participants with poor health conditions that cannot tolerate surgical interventions; [7] participants with severe mental illness, developmental disability, inability to read consent forms, or unable to cooperate with the researchers; and [8] participants under pregnancy or breastfeeding. A total of 20 participants were recruited and included in the analysis. None of the participants were lost to follow-up. ## Surgical procedures In the pre-operative phase, the desired correction angle and wedge size were measured and calculated from standing radiographs. Under endotracheal tube intubation general anaesthesia, the participant was placed in the supine position on a radiolucent operating table with a tourniquet applied. A medial longitudinal skin incision was made just distal to the joint line. After the pes anserinus was detached from the tibia, the superficial medial collateral ligament was exposed. The patellar tendon was protected by anterior retraction. Medial opening wedge osteotomy was carried out with a custom-made cutting jig. After temporary fixation of the cutting jig with multiple Kirschner-pins, the sawing was performed. The open wedge was subsequently filled with bone allograft, and a proximal medial tibia locking plate (APlus™) was used to fix the osteotomy. ## Preparation of the BMC Prior to bone marrow collection, 1000–1500 IU/mL heparin was used to flush and rinse all the instruments. 1.5 ml acid-citrate-dextrose solution was then added to a 10 ml syringe as an anticoagulant. Subsequently, 8.5 ml bone marrow was slowly aspirated. A total of 30–40 ml bone marrow was extracted by iliac crest aspiration. The bone marrow aspirate was centrifuged using A-BMC (Aeon™) to autologous BMC. After the HTO operation, 4–6 ml bone marrow aspiration concentrate was injected into the knee joint. ## Outcome measurement Radiographic outcomes of anatomical femorotibial angle and posterior tibial slope angle were measured at baseline and 12 months post-intervention. The measurements of anatomical femorotibial angle and posterior tibial slope angle were demonstrated in Supplementary Fig. 1. The clinical outcomes of pain (visual analogue scale [VAS]) and global function (Western Ontario and McMaster Universities Arthritis Index [WOMAC] and The Knee injury and Osteoarthritis Outcome Score [KOOS]) were obtained at baseline, 1 month, 3 months, 6 months, and 12 months postintervention. ## Statistical analysis Shapiro–Wilk test was used to check the normality of the data. To evaluate the differences of the structural and radiological outcomes before and after HTO, paired Wilcoxon signed-rank tests and paired T tests were used for non-normally and normally distributed data, respectively. To evaluate the differences of the clinical outcomes at baseline and different follow-up time points, Friedman tests and repeated measures ANOVA (analyses of variance) were performed for non-normally and normally distributed data, respectively. For repeated measures ANOVA, Mauchly’s tests were conducted to check if the sphericity assumption was met, and Greenhouse–Geisser corrections were applied if the sphericity assumption was violated. Paired Wilcoxon signed-rank tests and paired T tests with Bonferroni multiple testing correction method were used as post hoc analyses for Friedman tests and repeated measures ANOVA, respectively. Data were reported as mean ± standard deviation (SD) or median (interquartile range [IQR]). To model the change of the outcomes during the follow-up period, both linear and nonlinear mixed-effects models were used. The models with lower Akaike information criterion with correction for small sample sizes (AICc) and Bayesian information criterion (BIC) were considered to better fit the data. To further examine the factors that potentially influenced the treatment effects, covariates including age, sex, BMI, KL grade, femorotibial angle at baseline, posterior tibial slope angle at baseline, and symptoms severity at baseline were included in the models. Stepwise regressions, AICc, and BIC were used to select the optimal multivariate mixed-effects models. To validate the robustness of the multivariate models, the collinearity between the covariates were tested for significance. The values of correlation coefficient between 0 and 0.3 (0 and − 0.3) indicate a weak positive (negative) linear relationship, those between 0.3 and 0.7 (− 0.3 and − 0.7) indicate a moderate positive (negative) linear relationship, and those between 0.7 and 1.0 (− 0.7 and − 1.0) indicate a strong positive (negative) linear relationship [28]. The statistical analyses were all conducted in R (version 4.2.1), and the multivariate nonlinear mixed-effects models were performed using the “saemix” package. ## Baseline characteristics of the patients Table 1 summarises the patient characteristics prior to the surgery. The patients comprised 12 men and 8 women with a mean age of 61.4 years old. The median Kellgren-Lawrence (KL) grade was 3. The patient had varus deformity with a mean anatomical femorotibial angle of 2.6° and a mean posterior tibial slope angle of 5.9°. The patients were generally overweight, with the mean BMI of 27.7 kg/m2. Regarding the symptoms at baseline, the median VAS was 62.5 mm, the median total WOMAC score was 40.2, and the mean total KOOS score was 210.9.Table 1Baseline characteristics of the patientsVariableValueMin–MaxAge, y61.4 ± 7.143.3–67.9Sex, male/female, n$\frac{12}{8}$-Side of involvement, right/left, n$\frac{9}{11}$-KL gradeGrade II: 3, Grade III: 11, Grade IV: 6-Femorotibial angle (°)2.6 ± 4.6 (varus)11.8 (varus)–2.7 (valgus)Posterior tibial slope (°)5.9 ± 3.50.8–13.8BMI (kg/m2)27.7 ± 6.021.1–45.8VAS (mm)62.5 (30.0)5.0–95.0WOMAC total40.2 (21.8)25.2–91.2KOOS total210.9 ± 65.559.4–346.3Data presented as mean ± SD or median (IQR).The number of patients of each KL grade was presentedAbbreviations: BMI Body mass index, IQR Interquartile range, KOOS Knee injury and Osteoarthritis Outcome Score, KL Kellgren-Lawrence, max maximum; min, minimum, SD Standard deviation, VAS Visual analogue Scale, WOMAC Western Ontario and McMaster Universities Arthritis Index ## Radiographic outcomes Both the anatomical femorotibial angle and posterior tibial slope angle showed significant improvement after medial opening wedge HTO (Table 2A). The varus deformities were corrected ($p \leq 0.0001$), with the mean femorotibial angle after surgery being 8.6° valgus. A slight increase of the mean posterior tibial slope from 5.9° to 7.6° was also observed ($$p \leq 0.04$$).Table 2Improvement of radiographic and clinical outcomes over the follow-up periodA. Structural outcomesTime (month)Femorotibial angle (°)Posterior tibial slope (°)0-2.6 ± 4.6 (-11.8–2.7)5.9 ± 3.5 (0.8–13.8)128.6 ± 2.3 (4.9–11.6)7.6 ± 3.5 (1.5–14.9)Paired T test$p \leq 0.0001$p = 0.04B. VASTime (month)VAS (mm)062.5 (30.0, 5.0–95.0)140.0 (20.0, 0.0–65.0)a325.0 (10.0, 0.0–40.0)a,b614.5 (10.0, 0.0–30.0)a,b,c1210.0 (2.5, 0.0–20.0)a,b,cFriedman test χ2[4] = 68.9,$p \leq 0.0001$C. WOMACTime (month)TotalPainStiffnessPhysical function040.2 (21.8, 25.2–91.2)8.4 (5.6, 2.0–18.4)5.4 (3.7, 0.8–7.6)30.2 (14.8, 17.2–65.6)132.2 (18.0, 10.4–47.6)a6.8 (3.8, 0.0–9.6)a3.6 (2.0, 0.0–6.0)a22.6 (10.6, 8.0–34.8)a322.8 (9.9, 4.4–40.8)a,b4.8 (2.4, 0.0–7.2)a,b2.4 (1.8, 0.8–4.0)a,b15.4 (8.0, 1.6–29.6)a,b615.4 (9.4, 3.6–23.2)a,b,c2.8 (1.8, 0.8–5.6)a,b,c1.6 (0.6, 0.4–2.4)a,b,c11.2 (6.4, 1.6–15.6)a,b,c1210.0 (5.5, 2.8–20.4)a,b,c,d2.0 (1.2, 0.0–4.4)a,b,c,d0.8 (0.5, 0.4–1.6)a,b,c6.8 (4.2, 1.2–14.4)a,b,cFriedman testχ2[4] = 76.2,$p \leq 0.0001$χ2[4] = 66.0,$p \leq 0.0001$χ2[4] = 57.6,$p \leq 0.0001$χ2[4] = 74.8,$p \leq 0.0001$D. KOOSTime (month)TotalSymptomsPainFunction, ADLFunction, sports and recreational activitiesQuality of life0210.9 ± 65.5 (59.3–346.3)57.6 ± 13.3 (36.1–77.8)52.6 ± 18.9 (5.6–86.1)57.5 ± 15.3 (17.6–85.3)31.3 ± 18.5 (0.0–70.0)6.3 (18.8, 0.0–43.8)1257.9 ± 62.6 (174.0–364.4)67.8 ± 12.3 (50.0–91.7)61.4 ± 18.2 (33.3–91.7)64.9 ± 12.0 (47.1–83.8)38.5 ± 15.7 (20.0–70.0)25.0 (10.9, 6.3–50.0)a3319.7 ± 58.4 (208.4–405.7)a,b75.1 ± 8.0 (61.1–88.9)a,b70.8 ± 13.7 (50.0–97.2)a,b72.8 ± 14.3 (47.1–94.1)a,b52.5 ± 14.2 (20.0–75.0)a,b50.0 (9.4, 12.5–68.8)a,b6368.8 ± 43.8 (269.1–433.0)a,b,c81.9 ± 7.0 (66.7–100.0)a,b,c74.7 ± 11.0 (52.8–97.2)a,b79.9 ± 12.1 (55.9–94.1)a,b,c66.0 ± 8.5 (50.0–75.0)a,b,c75.0 (20.3, 43.8–81.2)a,b,c12387.3 ± 36.6 (320.6–451.4)a,b,c82.9 ± 6.0 (66.7–94.4)a,b,c79.3 ± 11.2 (55.6–97.2)a,b,c83.5 ± 8.1 (70.6–95.6)a,b,c70.3 ± 9.5 (50.0–85.0)a,b,c75.0 (6.3, 50.0–87.5)a,b,cRepeated measures ANOVAF(2.0, 38.3) = 75.2,$p \leq 0.0001$F(2.1, 40.1) = 30.9,$p \leq 0.0001$F(2.0, 38.2) = 26.3,$p \leq 0.0001$F(1.7, 33.1) = 29.8,$p \leq 0.0001$F(2.8, 52.5) = 45.9,$p \leq 0.0001$χ2[4] = 72.0,$p \leq 0.0001$†Data presented as mean ± SD (min–max) or median (IQR, min–max)Femorotibial angle: varus = negative value, valgus = positive valueAbbreviations: ADL Activities of daily living, ANOVA Analysis of variance, IQR Interquartile range, KOOS Knee injury and Osteoarthritis Outcome Score, max maximum, min minimum, SD Standard deviation, VAS Visual analogue Scale, WOMAC Western Ontario and McMaster Universities Arthritis Index Because of the nonparametric testing, only p values were provided† Friedman test used due to non-normal distribution of the dataa Significant difference from baselineb Significant difference from 1 month post-interventionc Significant difference from 3 months post-interventiond Significant difference from 6 months post-intervention ## Improvement of clinical outcomes over the follow-up period Based on the results of the Friedman tests and repeated measures ANOVA, VAS, WOMAC, KOOS, and their respective subscales all showed significant improvement over the follow-up period (Table 2B–D and Fig. 1). Furthermore, the improvement of the symptoms increased with the progress of time. According to the post hoc tests, most of the outcomes improved significantly compared to those at the previous follow-up time points. However, VAS and KOOS revealed no significant differences between the scores at 6 months and those at 12 months after the surgery (Tables 2B and D). The results indicated a possible plateau in pain reduction and function improvement after 6 months of follow-up. Fig. 1The trajectory of clinical symptom improvement over the follow-up duration. Each dot represents an outcome measure. The trajectories of symptom improvement of the patients are connected by grey lines. The blue lines represent the mean ± standard deviation or median ± interquartile range. Abbreviations: VAS, visual analogue scale; WOMAC, Western Ontario and McMaster Universities Arthritis Index; KOOS, The Knee injury and Osteoarthritis Outcome Score ## Nonlinear mixed-effects model with asymptotic regression Because a plateau was observed in symptom improvement as the follow-up duration increased, linear mixed-effects models may be unsuitable to fit the data. Nonlinear mixed-effects models with asymptotic regression were considered. The asymptotic regression can be described as the following equation:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta (t)=a+({a}_{0}-a){e}^{-rt}$$\end{document}Δ(t)=a+(a0-a)e-rt\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t$$\end{document}t represents the duration of follow-up. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta (t)$$\end{document}Δ(t) represents the differences between patients’ post-intervention scores at follow-up time \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$t$$\end{document}t and baseline scores. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${a}_{0}$$\end{document}a0 represents the \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 at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$$t = 0$$$\end{document}$t = 0$ and should by definition be close to 0. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$a$$\end{document}a represents the asymptote, which indicates the extent of symptom reduction when the plateau was reached. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$r$$\end{document}r is the natural logarithm of the decline rate constant. To validate the superiority of the asymptotic regression, both the linear and nonlinear mixed-effects models were performed, and their AICc and BIC values were compared. We further considered other nonlinear models, including power regressions and polynomial regressions, for model fit. The nonlinear mixed-effects models with asymptotic regressions showed the best fit to the trajectory of the symptom improvement (Supplementary Table 1). ## Factors associated with greater symptom improvement Based on the asymptotic regression, we further consider the effect of other covariates, including age, sex, BMI, KL grade, anatomical femorotibial angle at baseline, posterior tibial slope angle at baseline, and symptoms severity at baseline, on modifying the asymptote, thereby influencing the extent of symptom reduction and the rate of symptom improvement. The multivariate nonlinear mixed-effects model can be described as the following equation:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta (t)=(a+\sum_{i}{\beta }_{i}{C}_{i})+\left[{a}_{0}-(a+\sum_{i}{\beta }_{i}{C}_{i})\right]{e}^{-rt}$$\end{document}Δ(t)=(a+∑iβiCi)+a0-(a+∑iβiCi)e-rt \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C$$\end{document}C represents the covariates. A linear relationship was assumed between the covariates and their effects on \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$a$$\end{document}a, with \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 implicating the correlation coefficient of the linearity. Using the method of stepwise regression to construct multivariate nonlinear mixed-effects models (Supplementary Table 2–4), symptom severity at baseline and KL grade were significantly correlated with all the three outcomes (Table 3A–C). To more clearly visualise the effects of baseline symptom severity on the treatment outcomes, the regression lines of the patients with baseline symptom severity above and below the 50th percentiles were respectively constructed (Fig. 2). Patients with more severe symptoms at baseline experienced more rapid and greater symptom relief. Lower KL grades were also positively correlated with symptom improvement (Table 3A–C). In addition to baseline symptoms and KL grades, the WOMAC and KOOS questionnaire reported better global function improvement in patients with lower BMI (Table 3B and C), and those with more prominent baseline varus deformity tended to benefit more in VAS and KOOS outcomes (Table 3A and C). Furthermore, sex and age may play a role in influencing the extent of pain reduction (Table 3A) and physical function advancement (Table 3C), respectively. Table 3Covariates in nonlinear mixed-effects modelsA. VASNonlinear mixed-model with covariates\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta (t)=(a+\sum_{i}{\beta }_{i}{C}_{i})+\left[{a}_{0}-(a+\sum_{i}{\beta }_{i}{C}_{i})\right]{e}^{-rt}$$\end{document}Δ(t)=(a+∑iβiCi)+a0-(a+∑iβiCi)e-rtParameterEstimate[$95\%$ CI]P value\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$a$$\end{document}a-9.32[-20.75, 2.11]-\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\beta }_{Sex}$$\end{document}βSex5.95[0.99, 10.91]0.009\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\beta }_{KL grade}$$\end{document}βKLgrade4.69[1.34, 8.04]0.003\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\beta }_{FTA}$$\end{document}βFTA0.45[0.02, 0.88]0.03\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\beta }_{PTS}$$\end{document}βPTS-1.25[-1.88, -0.63]< 0.0001\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\beta }_{VAS baseline}$$\end{document}βVASbaseline-0.80[-0.88, -0.72]< 0.0001\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${a}_{0}$$\end{document}a00.67[-1.43, 1.70]-\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$r$$\end{document}r0.54[0.36, 0.72]-AICc; BIC693.07; 704.25B. WOMACNonlinear mixed-model with covariates\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta (t)=(a+\sum_{i}{\beta }_{i}{C}_{i})+\left[{a}_{0}-(a+\sum_{i}{\beta }_{i}{C}_{i})\right]{e}^{-rt}$$\end{document}Δ(t)=(a+∑iβiCi)+a0-(a+∑iβiCi)e-rtParameterEstimate[$95\%$ CI]P value\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$a$$\end{document}a-14.53[-26.68, -2.38]-\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\beta }_{BMI}$$\end{document}βBMI0.28[0.01, 0.55]0.04\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\beta }_{KL grade}$$\end{document}βKLgrade3.27[0.98, 5.56]0.003\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\beta }_{WOMAC baseline}$$\end{document}βWOMACbaseline-0.87[-0.88, -0.86]< 0.0001\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${a}_{0}$$\end{document}a00.13[-0.81, 1.07]-\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$r$$\end{document}r0.47[0.27, 0.67]-AICc; BIC573.85; 583.44C. KOOSNonlinear mixed-model with covariates\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta (t)=(a+\sum_{i}{\beta }_{i}{C}_{i})+\left[{a}_{0}-(a+\sum_{i}{\beta }_{i}{C}_{i})\right]{e}^{-rt}$$\end{document}Δ(t)=(a+∑iβiCi)+a0-(a+∑iβiCi)e-rtParameterEstimate[$95\%$ CI]P value\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$a$$\end{document}a364.13[211.94, 516.32]-\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\beta }_{Age}$$\end{document}βAge1.58[0.03, 3.13]0.04\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\beta }_{BMI}$$\end{document}βBMI-2.13[-3.30, -0.06]0.02\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\beta }_{KL grade}$$\end{document}βKLgrade-10.06[-20.08, -0.04]0.04\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\beta }_{FTA}$$\end{document}βFTA-5.51[-7.85, -3.17]< 0.0001\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\beta }_{KOOS baseline}$$\end{document}βKOOSbaseline-0.91[-1.07, -0.75]< 0.0001\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${a}_{0}$$\end{document}a0-8.18[-16.80, 0.44]-\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$r$$\end{document}r0.40[0.25, 0.55]-AICc; BIC988.82; 1000.77Sex: male = 0, female = 1Femorotibial angle: varus = negative value, vulgus = positive valueAbbreviations: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\varvec{a}}$$\end{document}a asymptote, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\varvec{a}}}_{0}$$\end{document}a0 intercept, AICc Akaike information criterion with correction for small sample sizes, BIC Bayesian information criterion, BMI body mass index, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\varvec{C}}}_{{\varvec{i}}}$$\end{document}Ci the ith covariate, CV coefficient of variation, FTA femorotibial angle, KL Kellgren-Lawrence, KOOS Knee injury and Osteoarthritis Outcome Score, PTS posterior tibial slope, r natural logarithm of the rate constant, SE standard error, t follow-up time, VAS Visual analogue Scale, WOMAC Western Ontario and McMaster Universities Arthritis Index, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\varvec{\beta}}}_{{\varvec{i}}}$$\end{document}βi correlation coefficient of the ith covariate, ∆ difference between patients’ postintervention and baseline scoresFig. 2The regression lines of the patients with baseline symptom severity above and below the 50th percentiles. Each dot represents an outcome measure. The trajectories of symptom improvement of the patients are connected by grey lines. The blue and red lines represent the regression lines. Abbreviations: VAS, visual analogue scale; WOMAC, Western Ontario and McMaster Universities Arthritis Index; KOOS, The Knee injury and Osteoarthritis Outcome Score To validate the robustness of the multivariate models, the correlation coefficients among the covariates were calculated and tested for significance (Supplementary Fig. 1). Although most of the covariates showed no significant collinearity, moderate to strong collinearity was observed between sex and WOMAC at baseline and between sex and KOOS at baseline. However, sex was removed from the final model of WOMAC and KOOS using stepwise regression. The collinearity was therefore unlikely to influence the results. Moderate collinearity was also observed among some of the structural and radiographic factors. The effects of these factors should be cautiously interpreted. ## Adverse events During the entire trial period, none of the patients reported infections, rejections, or other adverse events. ## Discussion In our study, the combination of medial opening wedge HTO and BMC significantly reduced pain and improved global function in patients with medial compartment knee osteoarthritis and varus deformity. The improvement of clinical symptoms reached a plateau 6 months after the surgery. Using multivariate nonlinear mixed-effects models with asymptomatic regressions, symptom severity at baseline and KL grades were significantly correlated with the prognosis. BMI, femorotibial angle, age, and sex may also play a role in influencing the extent of symptom relief. Medial opening wedge HTO is indicated for medial compartment knee osteoarthritis with varus deformity [29, 30]. By adjusting the femorotibial angle, the weight bearing of the knee can be redistributed [14, 15]. Although no consensus has been reached in terms of the optimal alignment in medial opening wedge HTO, 8–10° vulgus of the post-operative anatomical femorotibial angle was often suggested [31, 32]. In our study, the mean anatomical femorotibial angle was adjusted to 8.6° vulgus postoperatively, which indicated an appropriate correction of the malalignment. Moreover, our study indicated that patients with more severe pre-operative varus deformity experienced better post-operative outcomes, which further confirmed the benefits of medial opening wedge HTO on medial compartmental knee osteoarthritis. In addition to femorotibial angle, a slight increase in posterior tibial slope angle by 1.7° was observed. The result is in line with a previous meta-analysis, where the posterior tibial slope angle averagely increased 2.02° ($95\%$ confidence interval, 1.38° to 2.66°) after opening wedge HTO [33]. Prior to our study, some clinical studies have revealed the potential benefits of the combination of HTO and orthobiologics injection [24, 25, 27, 34–37]. However, few have investigated factors that could potentially affect the clinical outcomes [27]. In a study by Kim et al. [ 27], the effects of several factors on Lysholm score and KOOS after HTO and BMC injection were evaluated. Patients with higher KL grades and over 70 years old were significantly correlated with unsatisfactory clinical outcomes, whereas the association between other factors and clinical outcomes were not observed. In our study, the analyses of KL grades yielded similar results to those of the study by Kim et al.[27] However, contrary to their findings, older age may be related to better KOOS improvement, but due to the small sample size and the potential overfit of the stepwise regression in this study, the result could not be definitively confirmed. Moreover, patients > 70 years old were not included in this study. The results of the two studies may therefore not be comparable. In addition to KL grade and age, higher BMI was associated with lesser WOMAC and KOOS improvement. Although prior studies of HTO in combination with orthobiologics did not report similar results of BMI [27], the negative impact of overweight and obesity on the prognosis has been revealed in some of the previous HTO studies [21, 23, 38–40]. In our study, clinical symptom severity at baseline was included as a covariate in the nonlinear mixed-effects models, which was not investigated in the study by Kim et al.[27] According to previous cross-sectional studies, KL grades were not significantly correlated with symptoms of pain and physical function at baseline in patients with knee osteoarthritis, indicating absence of collinearity between radiographic and clinical factors [41, 42]. The tests for collinearity in our study also showed similar results. Therefore, the treatment strategies can be planned based on the baseline clinical features and functional status in addition to radiographic findings [41]. Based on the nonlinear mixed-effects models, patients with more severe pain or more limited physical function at baseline experienced more rapid and greater symptom relief. To the best of our knowledge, this is the first study of HTO in combination with BMC to indicate baseline symptom severity as a prognostic factor. In clinical practice, baseline clinical symptoms may also be considered in addition to structural and radiographic features prior to the intervention. One of the highlights of our study is the use of multivariate nonlinear mixed-effects models. Using asymptotic regressions to model the trajectory of symptom improvement, the results showed some differences from those reported by the study by Kim et al. [ 27]. In the study by Kim et al.[27], the authors set a threshold and dichotomised the continuous scores into satisfactory and unsatisfactory outcomes. To model the dichotomous outcomes, multivariate logistic regressions were performed. However, in this study, we believe that the changes of symptom severity from baseline to post-intervention were of more clinical significance. Therefore, we did not dichotomise the scores and used asymptotic regressions to more precisely model the continuous change of the outcomes. Our study has some limitations. First, there was no control group; thus, the comparison between HTO alone and the combination of HTO and BMC could not be performed. Furthermore, the analyses in this study may suffer from limitations of observational investigations without control groups, including bias arising from regression to the mean. Regression to the mean is a statistical phenomenon where the patients with relatively lower symptom severity at baseline tend to experience greater symptom severity nearer the true mean in the follow-up period, and vice versa, thereby making natural variation in repeated data look like real change [43]. However, because the patients in this study all experienced improvement in the outcomes, it may be unlikely that the effect of regression to the mean contributed to the overall effect estimates. Second, the sample size of this study was relatively small, which may not yield enough statistical power. The patients also had various disease severity and different KL grades. However, we did not perform subgroup analyses based on KL grades because this may further hinder the statistical power of this study. Third, to construct the multivariate nonlinear mixed-effects models, stepwise regressions were adopted, but the method is prone to overfitting. Therefore, we only focused on the factors that showed significant correlations with multiple outcomes. For those that showed significant correlation with only one outcome, the results should be cautiously interpreted. Fourth, because of the non-randomisation design of this study, selection bias may occur. However, because there were no missing data due to loss to follow-up, incomplete data collection, or exclusion from analysis, the probability of the existence of selection bias was low. Fifth, patients with more than 70 years of age were not included in this study; therefore, our result may not be able to be generalised to patients > 70 years old. In conclusion, this is the first clinical study of medial opening wedge HTO in combination with BMC injection to demonstrate the importance of baseline symptom severity on the prognosis. In clinical practice, we suggest that the evaluation of clinical features and functional status of the patients be more emphasised. Studies with larger sample size and longer follow-up duration are also warranted to validate the current evidence. ## Supplementary Information Additional file 1: Supplementary Table 1. Comparison of linear and nonlinear mixed-effects models. Supplementary Table 2. Stepwise regression in VAS. Supplementary Table 3. Stepwise regression in WOMAC. Supplementary Table 4. Stepwise regression in KOOS. Supplementary Figure 1. Demonstration of the measurement of anatomical femorotibial angle and posterior tibial slope angle. Supplementary Figure 2. Collinearity of the covariates in multivariate nonlinear mixed-effects models. ## References 1. Jang S, Lee K, Ju JH. Recent updates of diagnosis, pathophysiology, and treatment on osteoarthritis of the knee. Int J Mol Sci (2021) 22 2619. DOI: 10.3390/ijms22052619 2. Primorac D, Molnar V, Rod E, Jeleč Ž, Čukelj F, Matišić V. Knee Osteoarthritis: a review of pathogenesis and state-of-the-art non-operative therapeutic considerations. Genes (2020) 11 854. DOI: 10.3390/genes11080854 3. Hunter DJ, Schofield D, Callander E. The individual and socioeconomic impact of osteoarthritis. Nat Rev Rheumatol (2014) 10 437-441. DOI: 10.1038/nrrheum.2014.44 4. Hunter DJ, Bierma-Zeinstra S. Osteoarthritis Lancet (2019) 393 1745-1759. PMID: 31034380 5. Kulkarni K, Karssiens T, Kumar V, Pandit H. Obesity and osteoarthritis. Maturitas (2016) 89 22-28. DOI: 10.1016/j.maturitas.2016.04.006 6. Prieto-Alhambra D, Judge A, Javaid MK, Cooper C, Diez-Perez A, Arden NK. Incidence and risk factors for clinically diagnosed knee, hip and hand osteoarthritis: influences of age, gender and osteoarthritis affecting other joints. Ann Rheum Dis (2014) 73 1659-1664. DOI: 10.1136/annrheumdis-2013-203355 7. Bruyère O, Honvo G, Veronese N, Arden NK, Branco J, Curtis EM. An updated algorithm recommendation for the management of knee osteoarthritis from the European Society for Clinical and Economic Aspects of Osteoporosis, Osteoarthritis and Musculoskeletal Diseases (ESCEO). Semin Arthritis Rheum (2019) 49 337-350. DOI: 10.1016/j.semarthrit.2019.04.008 8. Bannuru RR, Osani MC, Vaysbrot EE, Arden NK, Bennell K, Bierma-Zeinstra SMA. OARSI guidelines for the non-surgical management of knee, hip, and polyarticular osteoarthritis. Osteoarthritis Cartilage (2019) 27 1578-1589. DOI: 10.1016/j.joca.2019.06.011 9. Kolasinski SL, Neogi T, Hochberg MC, Oatis C, Guyatt G, Block J. 2019 American College of Rheumatology/Arthritis Foundation guideline for the management of osteoarthritis of the Hand, hip, and knee. Arthritis rheumatol (2020) 72 220-233. DOI: 10.1002/art.41142 10. Gademan MGJ, Hofstede SN, Vliet Vlieland TPM, Nelissen RGHH, Marang-van de Mheen PJ. Indication criteria for total hip or knee arthroplasty in osteoarthritis: a state-of-the-science overview. BMC Musculoskelet Disord (2016) 17 463. DOI: 10.1186/s12891-016-1325-z 11. Li OL, Pritchett S, Giffin JR, Spouge ARI. High Tibial osteotomy: an update for radiologists. AJR Am J Roentgenol (2022) 218 701-712. DOI: 10.2214/AJR.21.26659 12. Liu X, Chen Z, Gao Y, Zhang J, Jin Z. High Tibial osteotomy: review of techniques and biomechanics. J Healthc Eng (2019) 2019 8363128. DOI: 10.1155/2019/8363128 13. Murray R, Winkler PW, Shaikh HS, Musahl V. High tibial osteotomy for varus deformity of the knee. J Am Acad Orthop Surg Glob Res Rev. (2021) 5 e21.00141. PMID: 34242204 14. Bayam L, Erdem M, Gülabi D, Erdem AC, Uyar AÇ, Kochai A. Clinical and radiological outcomes of high tibial osteotomy with combined fixator-assisted nailing and subtubercle tibial osteotomy. Acta Orthop Traumatol Turc (2020) 54 89-96. DOI: 10.5152/j.aott.2020.01.130 15. Ollivier B, Berger P, Depuydt C, Vandenneucker H. Good long-term survival and patient-reported outcomes after high tibial osteotomy for medial compartment osteoarthritis. Knee Surg Sports Traumatol Arthrosc (2021) 29 3569-3584. DOI: 10.1007/s00167-020-06262-4 16. Chen FH, Tuan RS. Mesenchymal stem cells in arthritic diseases. Arthritis Res Ther (2008) 10 223. DOI: 10.1186/ar2514 17. Ha C-W, Park Y-B, Kim SH, Lee H-J. Intra-articular mesenchymal stem cells in osteoarthritis of the knee: a systematic review of clinical outcomes and evidence of cartilage repair. Arthroscopy (2019) 35 277-88.e2. DOI: 10.1016/j.arthro.2018.07.028 18. Shariatzadeh M, Song J, Wilson SL. The efficacy of different sources of mesenchymal stem cells for the treatment of knee osteoarthritis. Cell Tissue Res (2019) 378 399-410. DOI: 10.1007/s00441-019-03069-9 19. Birmingham TB, Giffin JR, Chesworth BM, Bryant DM, Litchfield RB, Willits K. Medial opening wedge high tibial osteotomy: a prospective cohort study of gait, radiographic, and patient-reported outcomes. Arthritis Rheum (2009) 61 648-657. DOI: 10.1002/art.24466 20. Floerkemeier S, Staubli AE, Schroeter S, Goldhahn S, Lobenhoffer P. Outcome after high tibial open-wedge osteotomy: a retrospective evaluation of 533 patients. Knee Surg Sports Traumatol Arthrosc (2013) 21 170-180. DOI: 10.1007/s00167-012-2087-2 21. Herbst M, Ahrend M-D, Grünwald L, Fischer C, Schröter S, Ihle C. Overweight patients benefit from high tibial osteotomy to the same extent as patients with normal weights but show inferior mid-term results. Knee Surg Sports Traumatol Arthrosc (2022) 30 907-917. DOI: 10.1007/s00167-021-06457-3 22. Primeau CA, Birmingham TB, Leitch KM, Willits KR, Litchfield RB, Fowler PJ. Total knee replacement after high tibial osteotomy: time-to-event analysis and predictors. CMAJ (2021) 193 E158-E166. DOI: 10.1503/cmaj.200934 23. Wang F, Ma W, Chen J, Cong W, Zhang Y, Yu T. Prognostic factors for medial open-wedge high tibial osteotomy with spacer implantation in patients with medial compartmental knee osteoarthritis. J Orthop Surg Res (2022) 17 50. DOI: 10.1186/s13018-022-02934-8 24. Betzler BK, Bin Muhammad Ridzwan Chew AH, Bin Abd Razak HR. Intra-articular injection of orthobiologics in patients undergoing high tibial osteotomy for knee osteoarthritis is safe and effective - a systematic review. J Exp Orthop (2021) 8 83. DOI: 10.1186/s40634-021-00387-2 25. Tan SHS, Kwan YT, Neo WJ, Chong JY, Kuek TYJ, See JZF. Outcomes of high Tibial osteotomy with versus without mesenchymal stem cell augmentation: a systematic review and meta-analysis. Orthop J Sports Med (2021) 9 23259671211014840. DOI: 10.1177/23259671211014840 26. Yao R-Z, Liu W-Q, Sun L-Z, Yu M-D, Wang G-L. Effectiveness of high Tibial osteotomy with or without other procedures for medial compartment osteoarthritis of knee: an update meta-analysis. J Knee Surg (2021) 34 952-961. DOI: 10.1055/s-0039-1700978 27. Kim YS, Suh DS, Tak DH, Chung PK, Kwon YB, Kim TY. Factors influencing clinical and MRI outcomes of mesenchymal stem cell implantation with concomitant high Tibial osteotomy for Varus knee osteoarthritis. Orthop J Sports Med (2021) 9 2325967120979987. DOI: 10.1177/2325967120979987 28. Ratner B. The correlation coefficient: Its values range between +1/−1, or do they?. J Target Meas Anal Mark (2009) 17 139-142. DOI: 10.1057/jt.2009.5 29. Chahla J, Dean CS, Mitchell JJ, Moatshe G, Serra Cruz R, LaPrade RF. Medial opening wedge proximal Tibial osteotomy. Arthrosc Tech (2016) 5 e919-e928. DOI: 10.1016/j.eats.2016.04.019 30. Loia MC, Vanni S, Rosso F, Bonasia DE, Bruzzone M, Dettoni F. High tibial osteotomy in varus knees: indications and limits. Joints (2016) 4 98-110. DOI: 10.11138/jts/2016.4.2.098 31. Kang BY, Lee DK, Kim HS, Wang JH. How to achieve an optimal alignment in medial opening wedge high tibial osteotomy?. Knee Surg Relat Res (2022) 34 3. DOI: 10.1186/s43019-021-00130-2 32. Sabzevari S, Ebrahimpour A, Roudi MK, Kachooei AR. High Tibial osteotomy: a systematic review and current concept. Arch Bone Jt Surg (2016) 4 204-212. PMID: 27517063 33. Nha K-W, Kim H-J, Ahn H-S, Lee D-H. Change in posterior Tibial slope after open-wedge and closed-wedge high Tibial osteotomy: a meta-analysis. Am J Sports Med (2016) 44 3006-3013. DOI: 10.1177/0363546515626172 34. Jin Q-H, Chung Y-W, Na S-M, Ahn H-W, Jung D-M, Seon J-K. Bone marrow aspirate concentration provided better results in cartilage regeneration to microfracture in knee of osteoarthritic patients. Knee Surg Sports Traumatol Arthrosc (2021) 29 1090-1097. DOI: 10.1007/s00167-020-06099-x 35. Kim YS, Koh YG. Comparative matched-pair analysis of open-wedge high Tibial osteotomy with versus without an injection of adipose-derived mesenchymal stem cells for Varus knee osteoarthritis: clinical and second-look arthroscopic results. Am J Sports Med (2018) 46 2669-2677. DOI: 10.1177/0363546518785973 36. Kim J-H, Kim K-I, Yoon WK, Song S-J, Jin W. Intra-articular injection of mesenchymal stem cells after high Tibial osteotomy in osteoarthritic knee: two-year follow-up of randomized control trial. Stem Cells Transl Med (2022) 11 572-585. DOI: 10.1093/stcltm/szac023 37. Magnanelli S, Screpis D, Di Benedetto P, Natali S, Causero A, Zorzi C. Open-wedge high Tibial osteotomy associated with Lipogems® intra-articular injection for the treatment of Varus knee osteoarthritis - retrospective study. Acta Biomed (2020) 91 e2020022. PMID: 33559633 38. Floerkemeier S, Staubli AE, Schroeter S, Goldhahn S, Lobenhoffer P. Does obesity and nicotine abuse influence the outcome and complication rate after open-wedge high tibial osteotomy? A retrospective evaluation of five hundred and thirty three patients. Int Orthop (2014) 38 55-60. DOI: 10.1007/s00264-013-2082-3 39. Gilat R, Patel S, Knapik DM, Evuarherhe A, Haunschild E, Parvaresh K. Patient factors predictive of failure following high tibial osteotomy. J Cartilage Joint Preservation (2021) 1 100012. DOI: 10.1016/j.jcjp.2021.100012 40. Tuhanioğlu Ü, Oğur HU, Seyfettinoğlu F, Çiçek H, Gültekin A. High tibial osteotomy in obese patients: Is successful surgery enough for a good outcome?. J Clin Orthop Trauma (2019) 10 S168-S173. DOI: 10.1016/j.jcot.2018.09.004 41. Cubukcu D, Sarsan A, Alkan H. Relationships between pain, function and radiographic findings in osteoarthritis of the knee: a cross-sectional study. Arthritis (2012) 2012 984060. DOI: 10.1155/2012/984060 42. Özden F, Nadiye Karaman Ö, Tuğay N, Yalın Kilinç C, Mihriban Kilinç R, Umut TB. The relationship of radiographic findings with pain, function, and quality of life in patients with knee osteoarthritis. J Clin Orthop Trauma (2020) 11 S512-S517. DOI: 10.1016/j.jcot.2020.04.006 43. Barnett AG, van der Pols JC, Dobson AJ. Regression to the mean: what it is and how to deal with it. Int J Epidemiol (2004) 34 215-220. DOI: 10.1093/ije/dyh299
--- title: 'Performance of clinical signs and symptoms, rapid and reference laboratory diagnostic tests for diagnosis of human African trypanosomiasis by passive screening in Guinea: a prospective diagnostic accuracy study' authors: - Oumou Camara - Mamadou Camara - Laura Cristina Falzon - Hamidou Ilboudo - Jacques Kaboré - Charlie Franck Alfred Compaoré - Eric Maurice Fèvre - Philippe Büscher - Bruno Bucheton - Veerle Lejon journal: Infectious Diseases of Poverty year: 2023 pmcid: PMC10026442 doi: 10.1186/s40249-023-01076-1 license: CC BY 4.0 --- # Performance of clinical signs and symptoms, rapid and reference laboratory diagnostic tests for diagnosis of human African trypanosomiasis by passive screening in Guinea: a prospective diagnostic accuracy study ## Abstract ### Background Passive diagnosis of human African trypanosomiasis (HAT) at the health facility level is a major component of HAT control in Guinea. We examined which clinical signs and symptoms are associated with HAT, and assessed the performance of selected clinical presentations, of rapid diagnostic tests (RDT), and of reference laboratory tests on dried blood spots (DBS) for diagnosing HAT in Guinea. ### Method The study took place in 14 health facilities in Guinea, where 2345 clinical suspects were tested with RDTs (HAT Sero-K-Set, rHAT Sero-Strip, and SD Bioline HAT). Seropositives underwent parasitological examination (reference test) to confirm HAT and their DBS were tested in indirect enzyme-linked immunoassay (ELISA)/Trypanosoma brucei gambiense, trypanolysis, Loopamp Trypanosoma brucei Detection kit (LAMP) and m18S quantitative PCR (qPCR). Multivariable regression analysis assessed association of clinical presentation with HAT. Sensitivity, specificity, positive and negative predictive values of key clinical presentations, of the RDTs and of the DBS tests for HAT diagnosis were determined. ### Results The HAT prevalence, as confirmed parasitologically, was $2.0\%$ ($\frac{48}{2345}$, $95\%$ CI: 1.5–$2.7\%$). Odds ratios (OR) for HAT were increased for participants with swollen lymph nodes (OR = 96.7, $95\%$ CI: 20.7–452.0), important weight loss (OR = 20.4, $95\%$ CI: 7.05–58.9), severe itching (OR = 45.9, $95\%$ CI: 7.3–288.7) or motor disorders (OR = 4.5, $95\%$ CI: 0.89–22.5). Presence of at least one of these clinical presentations was $75.6\%$ ($95\%$ CI: 73.8–$77.4\%$) specific and $97.9\%$ ($95\%$ CI: 88.9–$99.9\%$) sensitive for HAT. HAT Sero-K-Set, rHAT Sero-Strip, and SD Bioline HAT were respectively $97.5\%$ ($95\%$ CI: 96.8–$98.1\%$), $99.4\%$ ($95\%$ CI: 99.0–$99.7\%$) and $97.9\%$ ($95\%$ CI: 97.2–$98.4\%$) specific, and $100\%$ ($95\%$ CI: 92.5–$100.0\%$), $59.6\%$ ($95\%$ CI: 44.3–$73.3\%$) and $93.8\%$ ($95\%$ CI: 82.8–$98.7\%$) sensitive for HAT. The RDT’s positive and negative predictive values ranged from 45.2–$66.7\%$ and 99.2–$100\%$ respectively. All DBS tests had specificities ≥ $92.9\%$. While LAMP and m18S qPCR sensitivities were below $50\%$, trypanolysis and ELISA/T.b. gambiense had sensitivities of $85.3\%$ ($95\%$ CI: 68.9–$95.0\%$) and $67.6\%$ ($95\%$ CI: 49.5–$82.6\%$). ### Conclusions Presence of swollen lymph nodes, important weight loss, severe itching or motor disorders are simple but accurate clinical criteria for HAT referral in HAT endemic areas in Guinea. Diagnostic performances of HAT Sero-K-Set and SD Bioline HAT are sufficient for referring positives to microscopy. Trypanolysis on DBS may discriminate HAT patients from false RDT positives. Trial registration The trial was registered under NCT03356665 in clinicaltrials.gov (November 29, 2017, retrospectively registered https://clinicaltrials.gov/ct2/show/NCT03356665) ## Background Infection with the parasite Trypanosoma brucei gambiense (T.b. gambiense) causes the chronic form of human African trypanosomiasis (HAT), also called sleeping sickness. While in Central Africa, the Democratic Republic of the Congo is responsible for about three quarters of all reported gambiense HAT patients, in West Africa, Guinea is frontrunner in the number of cases [1]. Almost all Guinean HAT cases occur along the coastline, in particular in the prefectures of Boffa, Dubreka and Forecariah [2, 3]. Despite considerable challenges, Guinea implements an efficacious HAT control program based on medical interventions supplemented with vector control. Even during the Ebola epidemic outbreak in 2014–2016, the national HAT control program managed to deploy insecticide impregnated targets in Boffa, and limited parasite transmission to humans by reducing the tsetse fly vector density [4]. Medical interventions against HAT in Guinea consist of passive and active screening, followed by treatment of confirmed HAT cases. During active screening, a specialized team visits the most affected villages and tests the whole population. An important drawback, however, is that once the disease prevalence drops, cost-effectiveness of active screening decreases as fewer cases are detected [5]. Also, as experienced in Guinea in the recent past, during epidemics of other infections, active screening may be interrupted [6, 7]. While approaching the status of elimination of HAT as a public health problem, passive screening for HAT, integrated in the existing health system, therefore increases in importance, is more resilient to interruption and more sustainable. In Guinea, passive screening was maintained at a low level during the Ebola epidemic, and was rapidly resumed in 2016 [3]. In passive screening, serological testing for HAT among individuals consulting a health centre, is initiated by the observation of symptoms or signs considered “suggestive” for HAT [3]. In such clinical suspects, an antibody detection test, usually a rapid diagnostic test (RDT), is carried out. Subjects that are RDT positive are subsequently examined microscopically for trypanosome presence, while those that test RDT negative, are considered HAT free. As neither the RDT positive predictive values (PPV) nor the sensitivities of parasite detection techniques are $100\%$, not all RDT positive suspects are parasitologically confirmed. To further discriminate potential HAT patients from RDT false positives and better target additional labour-intensive microscopic re-examinations, further reference laboratory tests can be carried out remotely on dried blood spots (DBS). While the most sensitive parasite detection techniques are routinely applied in Guinea [3, 8], for nearly all the other steps of the diagnostic chain, different options to increase or decrease suspicion for HAT are available, which, depending on their HAT diagnostic performance, influence effectiveness of screening. Although the clinical picture of gambiense HAT is relatively well documented [9, 10], the association of signs and symptoms with HAT in a healthcare seeking population has hardly been studied [11]. Furthermore, in the last decade, several RDTs have emerged for individual screening of HAT clinical suspects [12, 13]. For DBS testing, trypanolysis and enzyme-linked immunoassay (ELISA)/T.b. gambiense are available to detect antibodies against T.b. gambiense [14–16], while Trypanozoon specific DNA can be detected using the Loopamp Trypanosoma brucei Detection kit (LAMP) or m18S quantitative PCR (qPCR) [17, 18]. So far, for Guinea, the diagnostic performance of SD Bioline HAT, HAT Sero-K-Set and trypanolysis has mainly been evaluated retrospectively on stored plasma specimens or DBS [14, 19]. Within the framework of a multi-country diagnostic trial, the diagnostic performance of clinical signs and symptoms, of 3 RDTs and of serological and molecular reference laboratory tests on DBS was evaluated prospectively for diagnosis of HAT, in the context of passive screening in HAT endemic areas in Guinea. ## Study setting In Guinea, clinical suspects were prospectively recruited for the Diagnostic tools for human African trypanosomiasis elimination and clinical trials work package 2 (DiTECT-HAT-WP2) study between January 2017 and January 2020 in 14 hospitals and health posts in the prefectures of Boffa, Dubreka and Forecariah. In these 3 prefectures, the HAT prevalence expressed as number of HAT cases per 10,000 inhabitants in 2017 was respectively 2.92, 0.53 and 1.51, and decreased to 0.97, 0.33 and 0.99 in 2019 [3]. Serological screening sites (SSS) offered clinical and serological screening for HAT, and referred participants who were RDT positive to a centre for diagnosis and treatment (CDT). The CDT performed parasitological examination of RDT positives, in addition to clinical and serological screening. In Boffa prefecture, participants were recruited in Boffa hospital that acted as the CDT, while the health posts of Soumbouyadi, Tamita and Walia acted as SSS. In Dubreka prefecture, Dubreka LTO was the CDT, with 3 SSS, the health centres of Dubreka CU, Magnokhoun and Kholoya. In Forecariah prefecture, the health centre of Karakoro acted as CDT, while the health centers of Benty, Konta, Madinagbe, M’Boro and Sinkinet were SSS. The number of people looking for medical consultation in these 14 structures was around 40,000 in total in 2020 (ranging between 430 and 25,306 for individual centres). The initial sample size estimation was based on 13 health structures, performing RDTs on 15 clinical suspects each month, for 27 months, which would have resulted in 5265 inclusions. We estimated that in about $10\%$ of the clinical suspects tested, at least one RDT would be positive. The HAT prevalence among clinical suspects was estimated at $1\%$, which would result in inclusion of 53 HAT patients. ## Study protocol The study protocol is summarized in Fig. 1.Fig. 1DiTECT-HAT-WP2 study conduct and test results in Guinea. CSF Cerebrospinal fluid; DBS dried blood spot; HAT Human African trypanosomiasis; mAECT-BC mini anion exchange centrifugation on buffy coat; RDT rapid diagnostic test. * HAT confirmed by CSF examination. ND: not done. HAT ⊕: HAT positive. HAT∅: HAT free ## Inclusion criteria Individuals consulting the study SSS or CDT could be consecutively included if they had visited or resided in a HAT endemic area and presented with clinical suspicion for HAT. Clinical suspicion was defined as presence of at least one of the following clinical signs or symptoms: Recurrent fever not responding to anti-malarial medication; headache for a long duration (> 14 days); presence of enlarged lymph nodes in the neck; important weight loss; weakness; severe itching; amenorrhea, abortion, or sterility; coma; psychiatric problems (e.g., aggressiveness, apathy, mental confusion, increasing unusual hilarity, etc.); sleep disruption (nocturnal insomnia and/or excessive diurnal sleeping); motor disorders (abnormal movements, shaking, walking difficulties); convulsions; or speech disorders. Individuals were excluded from participation if they had already been treated for HAT, did not give their written informed consent or were less than 4 years old. ## Serological screening and parasitological confirmation Finger prick blood was tested with 3 RDTs, HAT Sero-K-Set (Coris Bioconcept, Gembloux, Belgium), rHAT Sero-Strip (Coris Bioconcept, Gembloux, Belgium) and SD Bioline HAT (Abbott, Yongin-si, Gyeonggi-do, the Republic of Korea) according to the instructions of the manufacturers. Clinical suspects negative in all 3 RDTs were considered HAT free, while those that were positive in at least one RDT, were considered serological suspects. Parasitological examination of serological suspects was carried out in the CDTs. If enlarged lymph nodes in the neck were present in the serological suspect, a lymph node puncture was performed, and a drop of lymph was microscopically examined under 10 × 40 magnification for presence of trypanosomes. From every serological suspect, venous blood on heparin was taken. If no lymphadenopathy was present or no parasites had been observed in lymph, 4 ml of heparinized blood was centrifuged, the buffy coat was taken and analysed for presence of trypanosomes using the mini anion exchange centrifugation technique on buffy coat (mAECT-BC) [8]. ## Dried blood spot preparation For every serological suspect two types of DBS were prepared. On a Whatman grade 4 filter paper, 16 drops of 30 µl of heparinized blood were deposited and left to dry. In parallel, 180 µl of heparinized blood were lysed for 5 min with 20 µl of $5\%$ SDS solution (Sigma Aldrich, Saint Louis, MO, USA), and 2 drops of 40 µl of lysed blood were deposited on a Whatman grade 1001 filter paper. Filter papers were dried, packed in separate envelopes, which in turn were packed in hermetic plastic bags containing silica gel. ## Patient management A lumbar puncture was carried out on parasitologically confirmed HAT patients, or if the clinician considered it appropriate, based on strong clinical suspicion. The cerebrospinal fluid (CSF) was examined for the number of white blood cells, and for presence of trypanosomes using the modified single centrifugation [20]. Patients with parasitologically confirmed HAT without CSF trypanosomes and white blood cell numbers ≤ 5/µl, were considered in first stage and HAT patients with > 5 white blood cells/µl or trypanosomes in CSF were classified as second stage. Treatment of HAT was carried out according to the treatment protocols in place at the CDTs. ## Study participants with missing data Serological suspects that could not be confirmed at the first microscopic examination, were invited for re-examination at the CDT or were re-examined by the national program. A number of RDT seropositives detected at SSS level and who did not show up at CDT were offered microscopic examination through the national sleeping sickness program (PNLTHA). ## Reference laboratory tests The DBS were sent to the Centre International de Recherche-Développement sur l’Elevage en zone Subhumide (Bobo-Dioulasso, Burkina Faso), where reference laboratory tests were performed. Test performers were not informed about the clinical and reference standard results. On DBS collected on Whatman grade 4 paper, trypanolysis and ELISA/T.b. gambiense were carried out for T.b. gambiense specific antibody detection, both targeting LiTat 1.3 and 1.5 VSG, following the methodology previously described [21]. For Trypanozoon DNA detection, m18S qPCR was carried out on DBS collected on Whatman grade 4 and if positive, followed by TgsGp-qPCR, while the lysed blood collected on Whatman grade 1001 was tested with Loopamp Trypanosoma brucei Detection Kit (LAMP, Eiken Chemical, Tokyo, Japan), according to published methodologies [18]. ## Data analysis Results obtained at the CDT were immediately entered in a digital case report form [22]. The application incorporated demographic, clinical and diagnostic data, and included pictures of positive RDTs and videos of trypanosome positive microscopy results. Results from SSS were collected on a paper case report form and entered retrospectively in the application. Data were transferred from the application to a central database and exported into a Excel spreadsheet (Microsoft Office Professionnel Plus 2016). Descriptive statistics were carried out to check for missing data and variation in each variable; categorical variables were summarized as proportions, while continuous variables were summarized with the median value and range. Regression analysis and evaluation of the diagnostic performance were based on the participants HAT status (Fig. 1). Participants with trypanosomes detected in lymph, blood or CSF were considered HAT positive. Participants who were triple RDT negative were considered HAT negative. Participants who were RDT positive but in whom no trypanosomes could be detected after microscopic examination(s), were considered HAT negative. Participants who were RDT positive, but did not undergo any parasitological examination were disregarded. Regression analysis using Stata Statistical Software (Release 14, College Station, TX: StataCorp LP) was performed to assess for associations with the HAT status and thus identify which demographic features and clinical presentations could be used as criteria to target future testing for HAT. Continuous variables were assessed for normal distribution, and the correlation between the thirteen clinical signs and symptoms was determined. Unconditional associations between HAT status and the explanatory variables (gender, age, and clinical signs and symptoms) were investigated. Subsequently, mixed logistic regression models were developed for the HAT status, with prefecture included as a random effect to account for spatial clustering within each prefecture. Backward elimination was then used to screen variables, and only statistically significant variables (P ≤ 0.05) were retained. Two-way interaction terms between all remaining variables were assessed for statistical significance. The final multivariable model included variables that were either statistically significant, or were part of a significant interaction term [23]. The intra-cluster coefficient was computed as the proportion of overall variation due to variation between groups, while interaction terms were interpreted using the coefficients [24]. As an example, the odds of a patient being HAT positive when having both enlarged lymph nodes in the neck and severe itching (compared to a patient who had neither enlarged lymph nodes in the neck nor severe itching), was determined as follows: exp [Coefficient Enlarged lymph nodes in the neck + Coefficient severe itching + Coefficient Enlarged lymph nodes in the neck X Severe itching]. The diagnostic performance of the clinical presentation (only those that were retained in mixed logistic regression), the three RDTs (individually, in parallel, and in series), and of the four reference laboratory tests was determined. The sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and accuracy for diagnosis of HAT were calculated with $95\%$ Clopper Pearson confidence intervals (GraphPad Prism 9). The Kappa agreement for combinations of RDTs and reference laboratory tests was also determined, and interpreted as poor (< 0.00), slight (0.00–0.20), fair (0.21–0.40), moderate (0.41–0.60), substantial (0.61–0.80), or almost perfect (0.81–1.00) [25]. ## Descriptive statistics of field results In total, 2353 clinical suspects were included: 707 in the prefecture of Boffa, 705 in Dubreka, and 941 in Forecariah. Of these clinical suspects, 1320 ($56.1\%$) were female and 1033 ($43.9\%$) male. Their median age was 30 years (range: 4–89). The most frequently observed clinical presentations were recurrent fever not responding to anti-malarial medication ($96\%$) and headache for a long duration ($80.3\%$), followed by weakness ($21.1\%$) (Table 1). Overall, among the 2353 study participants (Fig. 1), 122 tested positive to at least one RDT ($5.2\%$; $95\%$ CI: 4.3–$6.2\%$). Specifically, $\frac{110}{2352}$ ($4.7\%$; $95\%$ CI: 3.9–$5.6\%$) were positive to HAT Sero-K-Set; $\frac{44}{2350}$ ($1.9\%$, $95\%$ CI: 1.4–$2.5\%$) were positive to rHAT Sero-Strip; and $\frac{100}{2343}$ ($4.3\%$, $95\%$ CI: 3.5–$5.2\%$) were positive to SD Bioline HAT. Among the 122 RDT positives, 114 serological suspects were parasitologically examined (6 were lost to follow-up and 2 died before parasitology could be carried out). Forty-eight individuals were diagnosed with parasitologically confirmed HAT ($\frac{48}{2345}$, $2.0\%$; $95\%$ CI: 1.5–$2.7\%$), among whom 26 were trypanosome positive in lymph ($\frac{26}{48}$, $54.2\%$; $95\%$ CI: 39.2–$68.6\%$), and 28 in mAECT-BC ($\frac{28}{48}$, $58.3\%$; $95\%$ CI: 43.2–$72.4\%$). Out of 21 RDT positives for whom both lymph and blood were examined, eight tested positive in both body fluids. In MSC, $\frac{18}{40}$ HAT patients tested had trypanosomes in CSF ($45.0\%$; $95\%$ CI: 29.3–$61.5\%$), of which 2 had not been previously confirmed in lymph or blood ($\frac{2}{48}$, $4.2\%$; $95\%$ CI: 0.5–$14.3\%$). Three clinical suspects with positive RDTs were not parasitologically confirmed in blood or lymph and underwent lumbar puncture based on clinical suspicion, which enabled to confirm parasite presence by MSC in two. The remaining suspect (with recurrent fever not responding to anti-malarial medication, headache for a long duration, enlarged lymph nodes in the neck and motor disorders) had a white blood cell count of 265/µl, but in the absence of parasites, was not considered a HAT patient. The median CSF white blood cell count for HAT patients was 145/µl (range: 12–1086/µl). All HAT patients for whom CSF data were available ($\frac{46}{48}$) were in 2nd stage. The HAT patients included 19 females ($39.6\%$) and 29 males ($60.4\%$), and their median age was 26 years (range: 10–65).Table 1Frequency of clinical presentations in study participants and HAT patients, and association to HAT positivityVariableAll participantsn = 2353HATn = 48Univariable analysisMultivariable mixed logistic regressionFrequency (number)Frequency (number)P valueOR ($95\%$ CI)P valueOR ($95\%$ CI)Male$43.9\%$ [1033]$60.4\%$ [29]0.022.0 (1.1–3.5)0.0163.1 (1.2–7.6)AgeNANA0.020.98 (0.96–1.00)Recurrent fever not responding to anti-malarial medication$96.0\%$ [2258]$89.6\%$ [43]0.030.4 (0.1–0.9)Headache for a long duration$80.3\%$ [1889]$91.7\%$ [44]0.062.7 (0.07–7.7)Weakness$21.1\%$ [496]$54.2\%$ [26] < 0.0014.6 (2.6–8.2)Important weight loss$13.7\%$ [323]$81.3\%$ [39] < 0.00134.3 (16.3–72.6) < 0.00120.4 (7.1–58.9)Sleep disruption$9.7\%$ [229]$29.2\%$ [14] < 0.0014.0 (2.1–7.6)Enlarged lymph nodes in the neck$8.1\%$ [190]$77.1\%$ [37] < 0.00147.8 (23.9–95.6) < 0.00196.7 (20.7–452.0)Severe itching$6.4\%$ [150]$50.0\%$ [24] < 0.00118.1 (9.8–33.3) < 0.00145.9 (7.3–288.7)Amenorrhea, abortion, or sterility$6.1\%$ ($\frac{81}{1320}$)$57.9\%$ ($\frac{11}{19}$) < 0.00124.0 (9.4–61.7)Motor disorders$4.1\%$ [97]$52.1\%$ [25] < 0.00136.2 (19.2–68.2)0.074.5 (0. 9–22.5)Psychiatric problems$1.9\%$ [45]$22.9\%$ [11] < 0.00120.3 (9.5–43.4)Speech disorders$1.2\%$ [27]$0.0\%$ [0]0.98Convulsions$0.8\%$ [18]$10.4\%$ [5] < 0.00120.7 (7.0–61.0)Coma$0.2\%$ [4]$2.1\%$ [1]0.0215.7 (1.59–155.9)This table shows gender, age and the frequency of clinical symptoms and signs in clinical suspects and in HAT patients. The association to HAT positivity was first assessed in univariable analysis and after using multi-variable mixed logistic regression. HAT: Human African trypanosomiasis; OR: Odds ratio; CI: Confidence interval; only females considered, the numbers used to estimate the frequency are shown in the brackets ## HAT status of study participants For further analysis of the results, study participants were considered as true HAT positives if they were confirmed as HAT patients based on trypanosome observation during microscopy performed on blood, lymph, or CSF specimens ($$n = 48$$). Clinical suspects were considered as true HAT negatives ($$n = 2297$$) if they either (i) tested negative to all 3 RDTs ($$n = 2231$$); or (ii) were RDT positive, but parasite negative in microscopy ($$n = 66$$). The latter group included subjects (Fig. 1) who had all 4 reference laboratory tests negative ($$n = 37$$); subjects that did not undergo reference laboratory tests ($$n = 23$$); subjects that were reference laboratory test positive but in whom, upon re-examination with microscopy, no parasites could be found ($$n = 4$$); and subjects that were reference laboratory test positive but died before a second parasitological tests could be carried out ($$n = 2$$). The 8 RDT positives who were completely lost to follow-up and did not undergo any parasitology were excluded from further analyses. ## Clinical symptoms and signs associated with HAT, regression analysis The frequency of the different inclusion clinical symptoms and signs, in HAT ($$n = 48$$) and non-HAT affected study participants ($$n = 2297$$), is summarized in Fig. 2.Fig. 2Frequency of 13 clinical symptoms and signs in human African trypanosomiasis (HAT) and non-HAT affected study participants. The figure contains data for 48 HAT and 2297 non-HAT participants with the exception of * only 19 HAT and 1294 non-HAT female participants Convulsions were highly correlated with coma (Spearman's rho ρ = 0.87) and motor disorders (ρ = 0.63). Coma was also correlated with motor disorders (ρ = 0.70) and enlarged lymph nodes in the neck (ρ = 0.61). The results of the unconditional associations between the explanatory variables gender, age and clinical parameters, and the dependent variable HAT positivity, are presented in Table 1. While amenorrhea, abortion, or sterility was univariably associated with HAT positivity, it was not included in the multivariable model since it only related to female participants. The final multivariable logistic regression model for HAT patients included five explanatory variables (gender, enlarged lymph nodes in the neck, important weight loss, severe itching, and motor disorders) and two significant interaction terms (enlarged lymph nodes in the neck × severe itching and important weight loss × motor disorders). Clinical suspects presenting with enlarged lymph nodes in the neck had the highest odds (96.74) to have HAT, followed by those that presented with severe itching or important weight loss. Males had a higher odd of having HAT compared to females. Odds of clinical suspects presenting with both enlarged lymph nodes in the neck and severe itching (compared to a participant who had neither enlarged lymph nodes in the neck nor severe itching) increased to 413 ($$P \leq 0.03$$). Similarly, when both important weight loss and motor disorders were present (compared to a participant who had neither), the odds of being HAT positive increased to 2220 ($$P \leq 0.01$$). The intra-cluster correlation coefficient was 5.26 × 10–12, suggesting that spatial clustering within prefectures was negligible. ## Diagnostic performance of clinical presentation The diagnostic performance of (co-)occurrence of the 4 clinical symptoms and signs that were associated singly or in combination with HAT, namely enlarged lymph nodes in the neck, severe itching, important weight loss and motor disorders, was studied in function of the HAT status in 48 HAT and 2297 non-HAT affected study participants (Table 2). Presence of enlarged lymph nodes in the neck, and/or important weight loss and/or severe itching and/or motor disorders had $97.9\%$ sensitivity (only $\frac{1}{42}$ HAT patients did not have any of these 4 symptoms or signs) and $75.6\%$ specificity. Although the PPV of observing at least one symptom or sign remained limited to $7.7\%$, this increased to $39.3\%$ when co-existence of 2 or more of the 4 selected clinical presentations were considered. Table 2The diagnostic performance of occurrence of 4 key clinical presentations for human African trypanosomiasis (HAT) diagnosisNumber of signs or symptoms present% Sensitivity (n/N, $95\%$ CI)% Specificity (n/N, $95\%$ CI)% PPV (n/N, $95\%$ CI)% NPV (n/N, $95\%$ CI)% Accuracy (n/N, $95\%$ CI)≥ $\frac{1}{497.9}$ ($\frac{47}{48}$, 88.9–99.9)75.6 ($\frac{1737}{2297}$, 73.8–77.4)7.7 ($\frac{47}{607}$, 5.7–10.2)99.9 ($\frac{1737}{1738}$, 99.7–100)76.1 ($\frac{1784}{2345}$, 74.3–77.8) ≥ $\frac{2}{487.5}$ ($\frac{42}{48}$, 74.8–95.3)97.2 ($\frac{2232}{2297}$, 96.4–97.8)39.3 ($\frac{42}{107}$, 30.1–49.2)99.7 ($\frac{2232}{2238}$, 99.4–99.9)97.0 ($\frac{2274}{2345}$, 96.2–97.6)≥ $\frac{3}{456.3}$ ($\frac{27}{48}$, 41.2–70.5)99.7 ($\frac{2289}{2297}$, 99.3–99.9)77.1 ($\frac{27}{35}$, 59.9–89.6)99.1 ($\frac{2298}{2310}$, 98.6–99.4)98.8 ($\frac{2316}{2345}$, 98.2–99.2)$\frac{4}{418.8}$ ($\frac{9}{48}$, 9.0–32.6)100 ($\frac{2297}{2297}$, 99.8–100)100 ($\frac{9}{9}$, 66.4–100)98.3 ($\frac{2297}{2336}$, 97.7–98.8)98.3 ($\frac{2306}{2345}$, 97.7–98.8)Sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) of occurrence of presence of enlarged lymph nodes in the neck and/or important weight loss and/or severe itching, and/or motor disorders were determined for identification of HAT patients. The occurrence of at least one symptom or sign (≥ $\frac{1}{4}$) and co-occurrence (≥ $\frac{2}{4}$; ≥ $\frac{3}{4}$ or all $\frac{4}{4}$) of the 4 selected clinical presentations was counted in 48 HAT patients and 2297 non-HAT affected study participants, and proportions (n/N) with $95\%$ confidence intervals (CI) were determined ## Diagnostic performance of rapid diagnostic tests For estimation of the RDT diagnostic performance in 48 HAT and 2297 non-HAT clinical suspects, a few participants had partially missing RDT results (Table 3, Fig. 3). Of the three RDTs, HAT Sero-K-Set had the highest sensitivity ($100\%$), followed by SD Bioline HAT ($93.8\%$) and rHAT Sero-Strip ($59.6\%$). The highest specificity was observed with rHAT Sero-Strip ($99.4\%$), while HAT Sero-K-Set and SD Bioline HAT had similar specificities of $97.5\%$ and $97.9\%$ respectively. The PPV of the individual RDTs ranged between $45.2\%$ and $66.7\%$, while the NPV was between $99.2\%$ and $100\%$. Using the RDTs in parallel resulted in a high sensitivity (93.8–$100\%$), specificity (97.1–$97.7\%$) and NPV (99.9–$100\%$), while the PPV was limited (42.1–$46.4\%$). In series combinations including rHAT Sero-Strip led to low sensitivities ($59.6\%$), except for the HAT Sero-K-Set and SD Bioline HAT combination ($93.6\%$). There was moderate agreement between HAT Sero-K-Set and rHAT Sero-Strip (Kappa = 0.52; 0.43–0.62; SE = 0.02), and rHAT Sero-Strip and SD Bioline HAT (Kappa = 0.55; 0.45–0.65; SE = 0.02), while the agreement between HAT Sero-K-Set and SD Bioline HAT was almost perfect (Kappa = 0.86; 0.81–0.92; SE = 0.02).Table 3The diagnostic performance of 3 rapid diagnostic tests for diagnosis of human African trypanosomiasis (HAT)% Sensitivity (n/N, $95\%$ CI)% Specificity (n/N, $95\%$ CI)% PPV (n/N, $95\%$ CI)% NPV (n/N, $95\%$ CI)% Accuracy (n/N, $95\%$ CI)HAT Sero-K-Set100 ($\frac{47}{47}$, 92.5–100)97.5 ($\frac{2240}{2297}$, 96.8–98.1)45.2 ($\frac{47}{104}$, 35.4–55.3)100 ($\frac{2240}{2240}$, 99.8–100)97.6 ($\frac{2287}{2344}$, 96.9–98.2)rHAT Sero-Strip59.6 ($\frac{28}{47}$, 44.3–73.3)99.4 ($\frac{2281}{2295}$, 99.0–99.7)66.7 ($\frac{28}{42}$, 50.5–80.4)99.2 ($\frac{2281}{2300}$, 98.7–99.5)98.6 ($\frac{2309}{2342}$, 98.0–99.0)SD Bioline HAT93.8 ($\frac{45}{48}$, 82.8–98.7)97.9 ($\frac{2239}{2287}$, 97.2–98.4)48.4 ($\frac{45}{93}$, 37.9–59.0)99.9 ($\frac{2239}{2242}$, 99.6–100)97.8 ($\frac{2284}{2335}$, 97.1–98.4)The individual diagnostic sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV). Reference standard: Microscopic observation of trypanosomes in any body fluid. n/N: proportion; CI: Confidence intervalFig. 3Individual RDT results of human African trypanosomiasis (HAT) patients and HAT free participants. The Venn diagram shows results in the RDTs HAT Sero-K-Set, rHAT Sero-Strip and SD Bioline HAT of 48 HAT patients and 2297 HAT free participants. K° HAT Sero-K-Set not performed, S° rHAT Sero-Strip not performed, B° SD Bioline HAT not performed, HAT ⊕: HAT patients, HAT∅: HAT free participants ## Diagnostic performance of reference laboratory tests on dried blood spots Among the 48 HAT patients, $\frac{34}{48}$ had a DBS and all 4 DBS test results were available for $\frac{33}{34}$, while $\frac{1}{34}$ HAT patient missed a LAMP result. Among the 66 RDT positive HAT negatives (Fig. 1), $\frac{43}{66}$ had DBS and all 4 DBS test results were available for $\frac{42}{43}$, while $\frac{1}{43}$ missed a trypanolysis result. The individual results of all DBS are shown in Fig. 4.Fig. 4Reference laboratory test results on dried blood spots. The Venn diagram shows the results of trypanolysis, ELISA/T.b. gambiense, LAMP and m18S qPCR on DBS from 34 HAT patients (HAT ⊕) and 43 HAT negatives (HAT∅) who all tested rapid diagnostic test positive. T°trypanolysis not performed, L°LAMP not performed, ✞ died after the first parasitological examination at inclusion, including 1 HAT negative person with 265 WBC/µl. WBC White blood cell Table 4 summarizes the diagnostic performance of each individual reference laboratory test. Trypanolysis (in parallel on Trypanosoma brucei gambiense variable antigen type LiTat 1.3 and LiTat 1.5) had the highest sensitivity ($85.3\%$), followed by ELISA/T.b. gambiense ($67.6\%$). Sensitivities for m18S qPCR and LAMP were low. The highest specificity was observed for m18S qPCR ($97.7\%$), followed by ELISA/T.b. gambiense ($95.3\%$). The PPV ranged between $80.0\%$ for LAMP and $93.8\%$ for m18S qPCR, while the NPV ranged between $65.6\%$ for LAMP and $88.6\%$ for trypanolysis. There was fair agreement between LAMP and the three other reference laboratory tests [with trypanolysis: Kappa = 0.23 (0.02––0.4); SE = 0.10); with m18S qPCR: Kappa = 0.25 (0.00–0.51; SE = 0.11); with ELISA: Kappa = 0.29 (0.06–0.52; SE = 0.11)]. There was moderate agreement between m18S qPCR and both trypanolysis (Kappa = 0.42; 0.23–0.61; SE = 0.10) and ELISA/T.b.gambiense (Kappa = 0.51; 0.30–0.72; SE = 0.11). The agreement between ELISA/T.b. gambiense and trypanolysis was almost perfect (Kappa = 0.81; 0.67–0.94; SE = 0.11).Table 4The diagnostic performance of reference laboratory tests on dried blood spot for diagnosis of human African trypanosomiasisReference laboratory test% Sensitivity (n/N, $95\%$ CI)% Specificity (n/N, $95\%$ CI)% PPV (n/N, $95\%$ CI)% NPV (n/N, $95\%$ CI)% Accuracy (n/N, $95\%$ CI)Trypanolysis (in parallel)85.3 ($\frac{29}{34}$, 68.9–95.0)92.9 ($\frac{39}{42}$, 80.5–98.5)90.6 ($\frac{29}{32}$, 75.0–98.0)88.6 ($\frac{39}{44}$, 75.4–96.2)89.5 ($\frac{68}{76}$, 80.3–95.3)TL LiTat 1.379.4 ($\frac{27}{34}$, 62.1–91.3)92.9 ($\frac{39}{42}$, 80.5–98.5)90.0 ($\frac{27}{30}$, 73.5–97.9)84.8 ($\frac{39}{46}$, 71.1–93.7)86.8 ($\frac{66}{76}$, 77.1–93.5)TL LiTat 1.555.9 ($\frac{19}{34}$, 37.9–72.8)92.9 ($\frac{39}{42}$, 80.5–98.5)86.4 ($\frac{19}{22}$, 65.1–97.1)72.2 ($\frac{39}{54}$, 58.4–83.8)76.3 ($\frac{58}{76}$, 65.2–85.3)ELISA/T.b. gambiense67.6 ($\frac{23}{34}$, 49.5–82.6)95.3 ($\frac{41}{43}$, 84.2–99.4)92.0 ($\frac{23}{25}$, 74.0–99.0)78.8 ($\frac{41}{52}$, 65.3–88.9)83.1 ($\frac{64}{77}$, 72.9–90.7)m18S qPCR44.1 ($\frac{15}{34}$, 27.2–62.1)97.7 ($\frac{42}{43}$, 87.7–99.9)93.8 ($\frac{15}{16}$, 69.8–99.8)68.9 ($\frac{42}{61}$, 55.7–80.1)74.0 ($\frac{57}{77}$, 62.8–83.4)LAMP36.4 ($\frac{12}{33}$, 20.4–54.9)93.0 ($\frac{40}{43}$, 80.9–98.5)80.0 ($\frac{12}{15}$, 51.9–95.7)65.6 ($\frac{40}{61}$, 52.3–77.3)68.4 ($\frac{52}{76}$, 56.7–78.6)The individual diagnostic sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) of trypanolysis (TL), ELISA/T.b. gambiense, m18S qPCR and Loopamp Trypanosoma brucei Detection kit (LAMP) for diagnosis of HAT was determined on 77 dried blood spots collected from rapid diagnostic test positives. Reference standard: Microscopic observation of trypanosomes in any body fluid. CI: Cconfidence interval. 1 DBS missing The TgsGp-qPCR was carried out on 19 DBS only, 13 from HAT patients and 6 from RDT positive HAT negatives. Among the tested HAT patients, TgsGp-qPCR sensitivity was $38.5\%$ ($\frac{5}{13}$, $95\%$ CI: 17.7–$64.5\%$). The TgsGp-qPCR specificity was $100\%$ ($\frac{6}{6}$, $95\%$ CI: 61.0–$100.0\%$). ## Discussion The HAT prevalence observed among the study participants during the 3 years of passive screening was $2.0\%$. The overall HAT prevalence reported by the national program in passive screening in the same prefectures in 2017 and 2018 was respectively 0.98 and $0.39\%$ [3]. The DiTECT-HAT-WP2 study was set up in a small selection of experienced reference hospitals and health posts, including the 3 HAT reference centers that might attract relatively more HAT patients, which probably explains the difference. The relatively high prevalence allowed to successfully assess the sensitivity, specificity, positive and negative predictive value of clinical symptoms and signs, HAT rapid diagnostic tests and reference laboratory tests for diagnosis of HAT. Although there may be geographical and stage specific variations, the clinical picture of HAT has been described in detail [9, 10]. However, within a context of passive screening for HAT, it is important to consider the frequency of signs and symptoms in non-HAT affected individuals visiting the health infrastructure as well, for proposing criteria for HAT referral. This was well illustrated in the actual study by the criterion “presence of recurrent fever not responding to anti-malarial medication”. Fever is considered among the leading symptoms of HAT [26, 27] and has previously been reported at $97\%$ and $73.4\%$ frequency in Guinean HAT patients [28, 29]. Recurrent fever not responding to anti-malarial medication was also one of the most frequent symptoms ($89.6\%$) in our HAT patients but in univariable analysis, it was negatively associated with HAT and the odds that participants with fever would have HAT were below one, suggesting recurrent fever not responding to anti-malarial medication would be protective. This is probably an artefact given that it was common also in non-HAT affected participants ($96.1\%$). Indeed, in the multivariable analysis, both recurrent fever not responding to anti-malarial medication and having a headache for a long duration (the 2 clinical signs most frequently observed in the study participants) were not statistically significant. Among the 13 clinical symptoms and signs considered as inclusion criteria, using multivariable logistic regression, we were able to identify 4 key clinical presentations which could be used to select, among a health care seeking population in HAT endemic areas in Guinea, individuals at increased risk for HAT and which should be referred for further HAT screening. The presence of either enlarged lymph nodes in the neck, and/or severe itching, and/or important weight loss and/or motor disorders was in our study $97.9\%$ sensitive for HAT and had $75.6\%$ specificity, resulting in a PPV of $7.7\%$. A combination of at least 2 signs or symptoms increased the PPV to $39.3\%$, but resulted in a decrease in sensitivity. In particular presence of enlarged lymph nodes in the neck has previously been identified in $90\%$ and $93\%$ of HAT patients in Guinea, while occurrence of itching has been reported with frequencies of $93\%$ and $29.4\%$ [28, 29]. In the present study, enlarged lymph nodes in the neck and severe itching were present in respectively $77.1\%$ and $50.0\%$ of HAT patients, and it should be underlined that severe itching was retained as an inclusion criterion in the study at the specific request of the Guinean national HAT program. Although there might be geographical variation in the clinical presentation of HAT in different HAT endemic foci, previous independent studies on clinical presentation-based HAT diagnostic referral [11] identified sleep problems, neurological problems and weight loss as core symptoms in South Sudan, with or without oedema, swollen lymph nodes or proximity to livestock. Their diagnostic algorithms, based on these clinical presentations, had sensitivities up to $92.6\%$ and NPVs and PPVs of maximum $8.7\%$ [11]. Although itching was also significantly associated with HAT in South Sudan, it was not retained in the algorithms. In the Republic of Congo, enlarged lymph nodes, oedema, fever, headaches and itching were considered for establishing a clinical presentation based diagnostic algorithm for identifying HAT [30]. In Côte d’Ivoire, odds for having a positive RDT for HAT were increased in study participants with sleep disturbances, motor disorders, convulsions, important weight loss and psychiatric problems. In the Ivorian part of the DiTECT-HAT-WP2 study, only 2 HAT patients were identified, and the overall frequency of enlarged lymph nodes in the neck and severe itching was low (3.6 and $8.2\%$ respectively). Finally, our finding in the present study that males were at higher odds to have HAT than females, is in line with previous observations in Guinea, and has been linked to activities like rice growing, salt extraction, fishing and wood collection, which expose men more to the vector [29]. The hypothesis of unequal access to the health system [29], disfavouring women, can probably be excluded, as slightly more women were included in the present study. On the other hand, men traditionally participate less in active screening [29], which could explain why, once they start developing second stage HAT symptoms, they are more easily picked up through passive screening. The combined seroprevalence during this study was $5.2\%$, ranging from 1.9 to $4.7\%$ for the individual RDTs. As for the HAT prevalence, this was again higher than the overall seroprevalence of $1.72\%$ and $0.98\%$ previously reported in 2017 and 2018 [3]. The specificities of the 3 RDTs observed in Guinea confirm those reported in Côte d’Ivoire [21], and in other prospective evaluations in Central Africa [12, 13, 31, 32]. Despite the higher HAT prevalence in the actual study, the PPV of the 3 RDTs, ranging between $45.2\%$ and $66.7\%$ are similar to those observed in passive case detection in Guinea for 2017–2018 [3]. Sensitivity of HAT Sero-K-Set was $100\%$, confirming the high sensitivity for this test observed in prospective studies in the Democratic Republic of the Congo [12, 32]. For SD Bioline HAT, a wide variation of sensitivities has been reported from different prospective studies in Central Africa, ranging from $59.0\%$ [31] over $89.3\%$ [13] to $92.0\%$ [33]. In a retrospective study on stored plasma originating mainly from Guinean HAT patients, a sensitivity of $99.6\%$ was observed [19], although this might have been an overestimation due to selection bias using CATT/T.b. gambiense. In the present study, the $93.8\%$ sensitivity of SD Bioline HAT was close to the higher sensitivity values reported for the Democratic Republic of the Congo [33]. The sensitivity of $59.6\%$ observed with rHAT Sero-Strip in the present study was low compared to the > $97.5\%$ sensitivities obtained using stored specimens in a laboratory evaluation [34]. This was the first evaluation of the sensitivity of the rHAT Sero-Strip dipstick test under field conditions and it cannot be excluded that transport stress, the higher environmental temperatures or the humidity might have affected test stability. In parallel or in series combination of tests, with or without rHAT Sero-Strip did not improve diagnostic performance, probably because of the agreement between the HAT Sero-K-Set and SD Bioline HAT test results. Evaluation of the diagnostic performance of the parasitological tests was not an objective of our study and not all tests were systematically performed on all RDT positives, but our results confirm the relatively high sensitivity of lymph examination in Guinea [8, 29]. Indeed, $\frac{26}{48}$ ($54.2\%$) of the HAT patients had trypanosomes upon microscopic examination of the lymph node exudate. Unfortunately, DBS were missing for a relatively high number of RDT positives. The specificity of the 4 reference laboratory tests in this study was similar as for passive screening in Côte d’Ivoire [21]. For ELISA/T.b. gambiense and trypanolysis, specificity was lower than in active screening in Burkina Faso [35]. However, among the 6 DBS positives considered as non-HAT for determination of the diagnostic performance (Figs. 1 and 4), two participants died before they could be re-examined. As these participants had one negative microscopic examination, we could not determine the exact cause of death and as we cannot exclude that the positive test results were due to cross reactivity, we preferred not to exclude these subjects from the study. They had respectively $\frac{2}{4}$ and $\frac{3}{4}$ key symptoms and signs, were positive in $\frac{3}{3}$ and $\frac{2}{3}$ RDTs and were both positive in trypanolysis and ELISA/T.b. gambiense but not in the molecular tests. It cannot be confirmed nor excluded that these were real HAT patients, thus the specificity values of trypanolysis and ELISA/T.b. gambiense might have been underestimated. The sensitivities of trypanolysis and ELISA/T.b. gambiense observed in the present study were modest. It has previously been demonstrated that the sensitivity of trypanolysis and inhibition ELISA is lower on DBS compared to plasma [14, 36]. In DR Congo, trypanolysis on DBS was $95.1\%$ sensitive [37], while ELISA/T.b. gambiense was estimated to be $82.2\%$ sensitive [38]. The low sensitivities for the molecular tests LAMP and m18S qPCR were not a complete surprise. Firstly, DBS have been shown to be suboptimal for PCR and better results are obtained with nucleic acid preservation in different types of stabilisation buffers [37, 39]. Secondly, LAMP and m18S qPCR on DBS have been shown to have limited analytical sensitivity (100 and 1000 trypanosomes/ml respectively) [18], which is lower than that of mAECT-BC (10 trypanosomes/ml) [8], which was used together with lymph and CSF examination, to diagnose HAT in the present study. Finally, a prolonged or suboptimal storage of DBS could also have affected the sensitivity of both the serological and the molecular reference laboratory tests: DBS were not systematically shipped to the reference laboratory (median delay of 3 months, with a maximum delay of up to 9 months), and exposure to humidity, despite storage with silica gel, cannot be entirely excluded. Some limitations of this multi-country study have already been discussed in detail elsewhere [21], including non-inclusion of individuals without symptoms, incomplete inclusion of individuals presenting with symptoms or signs at the CDT or SSS, the assumption that individuals testing negative in all 3 RDTs are not affected by HAT without carrying out parasitological examinations, and imperfect sensitivity of parasitological techniques used as a gold standard. A number of additional limitations apply to the present study in Guinea. All the 48 HAT patients that were diagnosed suffered from stage 2 HAT. This is a known problem inherent to passive screening [3], not only in Guinea [11, 26, 27, 40]. As a result, the proposed key clinical presentations might have high sensitivity for stage 2 HAT, but their real ability to pick up stage 1 HAT patients, which may be asymptomatic or have only mild symptoms, remains to be determined. No stage 1 patients have been diagnosed in Guinea in the last 3 years. This could also be a consequence of the vector control program that has been rolled out in all endemic districts since 2017, impacting disease transmission. Furthermore, severe itching and enlarged lymph nodes in the neck in particular are relatively well known as clinical symptoms and signs of HAT in Guinea. It is possible that clinicians responsible for including participants gave more attention to these clinical presentations, compared to others which could have remained under-detected. The clinical presentation was considered as a rapid and simple decision tool for inclusion and referral to more specific HAT RDTs, and no attempts were made to quantify signs or symtoms. Some presentations, like psychiatric problems, motor disorders and speech disorders were grouped. Being relatively imprecise, we can therefore not exclude that the clinical data have been influenced by a level of subjectivity by the patient or the clinician, or by the clinician’s skills. Moreover, inclusion of participants in the study was not done by expert clinicians but by local health workers from the health centres. However, in particular for important weight loss, severe itching and motor disorders, detailed clinical data collected in therapeutic trials carried out on Guinean HAT patients could be consulted retrospectively [41]. Many DBS of RDT positives were missing, leading to relatively large confidence intervals in diagnostic test performance estimations. Health workers from the selected health centres collected the DBS and they had a lot of work in addition to the study. They may have been overloaded and forgot to take DBS. In addition, at the beginning of the study, sometimes an insufficient volume of blood was spotted and these DBS had to be discarded. Moreover, the delay between collection and analysis of DBS might have affected sensitivity of the tests. Finally, newer second generation RDTs were not available during the study, in particular SD Bioline HAT 2.0 with recombinant antigens, despite its large scale evaluation in the Democratic Republic of the Congo prior to the start of the present study [31]. Production of rHAT Sero-Strip has since been discontinued, while SD Bioline HAT with native antigens is nowadays unavailable. The actual study also has important strengths. The high HAT prevalence in Guinea allowed us to assess the association between clinical symptoms and signs and HAT in Guinea and the diagnostic performance of the combination of 4 key clinical presentations, 3 RDTs, and consequent reference laboratory tests on DBS of RDT positives. A similar study in Côte d’Ivoire [21] allowed us to associate the clinical picture with RDT positivity and to assess test specificity of RDTs and reference laboratory tests, but included only 2 HAT patients. The Guinean HAT control program paid particular attention to actively retrieving RDT positives, resulting in a limited loss to follow-up. ## Conclusions In passive case detection, we can propose to health workers and clinicians in Guinean HAT endemic areas a relatively simple set of criteria with high sensitivity for selecting individuals to be further tested using HAT RDTs, which would result in a reduction of almost $70\%$ of the HAT RDTs to be carried out. Performance of both HAT Sero-K-Set and SD Bioline HAT is sufficient for referring RDT positives for microscopic examination. Taking into account the future availability of non-toxic oral drugs to treat both stages of HAT [41], a target product profile has been established by the World Health Organization for a gambiense HAT test to identify individuals with suspected but microscopically unconfirmed g-HAT infection, eligible for treatment [42]. In HAT endemic areas in Guinea, both RDTs largely meet the minimum requirements for specificity, while HAT Sero-K-Set meets the desirable and SD Bioline HAT approaches the minimal sensitivity. Testing of DBS may discriminate individuals with false positive RDTs from true HAT patients which need to be re-examined for confirmation of HAT, and could become important to know the epidemiological HAT status of specific areas in the future. All DBS tests showed sufficient specificity, but only the antibody detection tests, and in particular trypanolysis, had sufficient sensitivity. More care should however be given to correct collection, storage, and shipment of DBS, and to minimising the delay between collection and testing of the specimens. In the future, priority should be given to assessing the diagnostic performance of new generation RDTs and of new reference laboratory tests. Inhibition ELISA could replace trypanolysis [36], increasing the feasibility of implementation of DBS testing in-country and reducing DBS storage time. New nucleic acid detection tests including SHERLOCK [43] or (RT)-qPCRs [39] should be evaluated prospectively. Decisions for the most suitable diagnostic algorithm for passive HAT case detection in HAT endemic areas in Guinea should also be guided by cost-effectiveness analysis. ## References 1. Franco JR, Cecchi G, Paone M, Diarra A, Grout L, Kadima Ebeja A. The elimination of human African trypanosomiasis: achievements in relation to WHO road map targets for 2020. PLoS Negl Trop Dis (2022.0) 16 e0010047. DOI: 10.1371/journal.pntd.0010047 2. Kagbadouno MS, Camara M, Rouamba J, Rayaisse JB, Traoré IS, Camara O. Epidemiology of sleeping sickness in Boffa (Guinea): where are the trypanosomes?. PLoS Negl Trop Dis (2012.0) 6 e1949. DOI: 10.1371/journal.pntd.0001949 3. Camara O, Biéler S, Bucheton B, Kagbadouno M, Mathu Ndung’u J, Solano P. Accelerating elimination of sleeping sickness from the Guinean littoral through enhanced screening in the post-Ebola context: a retrospective analysis. PLoS Negl Trop Dis (2021.0) 15 e0009163. DOI: 10.1371/journal.pntd.0009163 4. Kagbadouno M, Camara O, Camara M, Ilboudo H, Camara M, Rayaisse JB. Ebola outbreak brings to light an unforeseen impact of tsetse control on sleeping sickness transmission in Guinea. BioRxiv (2018.0) 5 202762. DOI: 10.1101/202762 5. Snijders R, Fukinsia A, Claeys Y, Hasker E, Mpanya A, Miaka E. Costs and outcomes of integrated human African trypanosomiasis surveillance system using rapid diagnostic tests, Democratic Republic of the Congo. Emerg Infect Dis (2021.0) 27 2144. DOI: 10.3201/eid2708.202399 6. Camara M, Ouattara E, Duvignaud A, Migliani R, Camara O, Leno M. Impact of the Ebola outbreak on. PLoS Negl Trop Dis (2017.0) 11 e0006060. DOI: 10.1371/journal.pntd.0006060 7. 7.World Health Organization. Control of Neglected Tropical Diseases. NTDs & COVID-19: https://www.who.int/teams/control-of-neglected-tropical-diseases/overview/ntds-and-covid-19. Accessed 21 Nov 2021. 8. Camara M, Camara O, Ilboudo H, Sakande H, Kaboré J, N’Dri L. Sleeping sickness diagnosis: use of buffy coats improves the sensitivity of the mini anion exchange centrifugation test. Trop Med Int Health (2010.0) 15 796. DOI: 10.1111/j.1365-3156.2010.02546.x 9. Boa YF, Traore MA, Doua F, Kouassi-Traore MT, Kouassi BE, Giordano C. The different present-day clinical picture of human African trypanosomiasis caused by. Bull Soc Pathol Exot Filiales. (1988.0) 81 427. PMID: 2848637 10. Blum J, Schmid C, Burri C. Clinical aspects of 2541 patients with second stage human African trypanosomiasis. Acta Trop (2006.0) 97 55. DOI: 10.1016/j.actatropica.2005.08.001 11. Palmer JJ, Surur EI, Goch GW, Mayen MA, Lindner AK, Pittet A. Syndromic algorithms for detection of gambiense human African trypanosomiasis in South Sudan. PLoS Negl Trop Dis (2013.0) 7 e2003. DOI: 10.1371/journal.pntd.0002003 12. Büscher P, Mertens P, Leclipteux T, Gilleman Q, Jacquet D, Mumba-Ngoyi D. Sensitivity and specificity of HAT Sero-. Lancet Glob Health (2014.0) 2 e359. DOI: 10.1016/S2214-109X(14)70203-7 13. Bisser S, Lumbala C, Nguertoum E, Kande V, Flevaud L, Vatunga G. Sensitivity and specificity of a prototype rapid diagnostic test for the detection of. PLoS Negl Trop Dis (2016.0) 10 e0004608. DOI: 10.1371/journal.pntd.0004608 14. Camara O, Camara M, Lejon V, Ilboudo H, Sakande H, Léno M. Immune trypanolysis test with blood spotted on filter paper for epidemiological surveillance of sleeping sickness. Trop Med Int Health (2014.0) 19 828. DOI: 10.1111/tmi.12316 15. Mumba D, Bohorquez E, Messina J, Kande V, Taylor SM, Tshefu AK. Prevalence of human African trypanosomiasis in the Democratic Republic of the Congo. PLoS Negl Trop Dis (2011.0) 5 e1246. DOI: 10.1371/journal.pntd.0001246 16. Inocencio da Luz R, Phanzu DM, Kiabanzawoko ON, Miaka E, Verlé P, De Weggheleire A. Feasibility of a dried blood spot strategy for serological screening and surveillance to monitor elimination of human African trypanosomiasis in the Democratic Republic of the Congo. PLoS Negl Trop Dis (2021.0) 15 e0009407. DOI: 10.1371/journal.pntd.0009407 17. Wamboga C, Matovu E, Bessell PR, Picado A, Biéler S, Ndung’u JM. Enhanced passive screening and diagnosis for gambiense human African trypanosomiasis in north-western Uganda—moving towards elimination. PLoS ONE (2017.0) 12 e0186429. DOI: 10.1371/journal.pone.0186429 18. Compaoré CAF, Ilboudo H, Kaboré J, Kaboré JW, Camara O, Bamba M. Analytical sensitivity of loopamp and quantitative real-time PCR on dried blood spots and their potential role in monitoring human African trypanosomiasis elimination. Exp Parasitol (2020.0) 219 108014. DOI: 10.1016/j.exppara.2020.108014 19. Jamonneau V, Camara O, Ilboudo H, Peylhard M, Koffi M, Sakande H. Accuracy of individual rapid tests for serodiagnosis of gambiense sleeping sickness in West Africa. PLoS Negl Trop Dis (2015.0) 9 e0003480. DOI: 10.1371/journal.pntd.0003480 20. Büscher P, Mumba Ngoyi D, Kaboré J, Lejon V, Robays J, Jamonneau V. Improved models of mini anion exchange centrifugation technique (mAECT) and modified single centrifugation (MSC) for Sleeping sickness diagnosis and staging. PLoS Negl Trop Dis (2009.0) 3 e471. DOI: 10.1371/journal.pntd.0000471 21. Koné M, Kaba D, Kaboré J, Thomas LF, Falzon LC, Koffi M. Passive surveillance of human African trypanosomiasis in Côte d’Ivoire: understanding prevalence, clinical symptoms and signs, and diagnostic test characteristics. PLoS Negl Trop Dis (2021.0) 15 e0009656. DOI: 10.1371/journal.pntd.0009656 22. Hasker E, Kwete J, Inocencio da Luz R, Mpanya A, Bebronne N, Makabuza J. Innovative digital technologies for quality assurance of diagnosis of human African trypanosomiasis. PLoS Negl Trop Dis (2018.0) 12 e0006664. DOI: 10.1371/journal.pntd.0006664 23. Dohoo I, Martin W, Stryhn H. Veterinary epidemiologic research (2003.0) 24. Chen JJ. Communicating complex information: The interpretation of statistical interaction in multiple logistic regression analysis. Am J Public Health (2003.0) 93 1376. DOI: 10.2105/AJPH.93.9.1376-a 25. Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics (1977.0) 33 159. DOI: 10.2307/2529310 26. 26.World Health Organization. Control and surveillance of human African trypanosomiasis: report of a WHO Expert Committee. Geneva: World Health Organization; 2013 https://apps.who.int/iris/handle/10665/95732. Accessed 21 Nov 2022. 27. Hasker E, Lumbala C, Mbo F, Mpanya A, Kande V, Lutumba P. Health care-seeking behaviour and diagnostic delays for human African trypanosomiasis in the Democratic Republic of the Congo: Diagnosing HAT in DRC. Trop Med Int Health (2011.0) 16 869. DOI: 10.1111/j.1365-3156.2011.02772.x 28. Camara M, Kaba D, KagbaDouno M, Sanon JR, Ouendeno FF, Solano P. Human African trypanosomiasis in the mangrove forest in Guinea: epidemiological and clinical features in two adjacent outbreak areas. Med Trop (2005.0) 65 155 29. Vanhecke C, Guevart E, Ezzedine K, Receveur MC, Jamonneau V, Bucheton B. Human African trypanosomiasis in mangrove epidemiologic area. Presentation, diagnosis and treatment in Guinea, 2005–2007. Pathol Biol (2010.0) 58 110. DOI: 10.1016/j.patbio.2009.07.033 30. 30.Jannin J, Moulia-Pelat JP, Chanfreau B, Penchenier L, Louis JP, Nzaba P, et al. African human trypanosomiasis: study of a scoring system of presumptive diagnosis in the Congo. Bull World Health Organ. 1993;71:215 (In French). 31. Lumbala C, Biéler S, Kayembe S, Makabuza J, Ongarello S, Ndung’u JM. Prospective evaluation of a rapid diagnostic test for. PLoS Negl Trop Dis (2018.0) 12 e0006386. DOI: 10.1371/journal.pntd.0006386 32. Boelaert M, Mukendi D, Bottieau E, Kalo Lilo JR, Verdonck K, Minikulu L. A phase III diagnostic accuracy study of a rapid diagnostic test for diagnosis of second-stage human African trypanosomiasis in the Democratic Republic of the Congo. EBioMedicine (2018.0) 27 11. DOI: 10.1016/j.ebiom.2017.10.032 33. Lumbala C, Bessell PR, Lutumba P, Baloji S, Biéler S, Ndung’u JM. Performance of the SD BIOLINE® HAT rapid test in various diagnostic algorithms for gambiense human African trypanosomiasis in the Democratic Republic of the Congo. PLoS ONE (2017.0) 12 e0180555. DOI: 10.1371/journal.pone.0180555 34. Büscher P, Gilleman Q, Lejon V. Rapid diagnostic test for sleeping sickness. N Engl J Med (2013.0) 368 1069. DOI: 10.1056/NEJMc1210373 35. Compaoré CFA, Kaboré J, Ilboudo H, Thomas LF, Falzon LC, Bamba M. Monitoring the elimination of. Parasite (2022.0) 29 25. DOI: 10.1051/parasite/2022024 36. Geerts M, Van Reet N, Leyten S, Berghmans R, Rock KS, Coetzer THT. Clin Infect Dis (2021.0) 73 e2477. DOI: 10.1093/cid/ciaa1264 37. Mumba Ngoyi D, Ali Ekangu R, Mumvemba Kodi MF, Pyana PP, Balharbi F, Decq M. Performance of parasitological and molecular techniques for the diagnosis and surveillance of gambiense sleeping sickness. PLoS Negl Trop Dis (2014.0) 8 e2954. DOI: 10.1371/journal.pntd.0002954 38. Hasker E, Lutumba P, Mumba D, Lejon V, Büscher P, Kande V. Diagnostic accuracy and feasibility of serological tests on filter paper samples for outbreak detection of. Am J Trop Med Hyg (2010.0) 83 374. DOI: 10.4269/ajtmh.2010.09-0735 39. Ngay Lukusa I, Van Reet N, Mumba Ngoyi D, Miaka EM, Masumu J, Patient Pyana P. Trypanosome SL-RNA detection in blood and cerebrospinal fluid to demonstrate active gambiense human African trypanosomiasis infection. PLoS Negl Trop Dis (2021.0) 15 e0009739. DOI: 10.1371/journal.pntd.0009739 40. Pepin J, Guern C, Milord F, Bokelo M. Integration of African human trypanosomiasis control in a network of multipurpose health centers. Bull World Health Organ (1989.0) 67 301. PMID: 2766452 41. Kande Betu Kumeso V, Kalonji Mutombo W, Rembry S, Valverde Mordt O, Ngolo Tete D, Prêtre A. Efficacy and safety of acoziborole in patients with human African trypanosomiasis caused by. Lancet Infect Dis (2022.0). DOI: 10.1016/S1473-3099(22)00660-0 42. 42.World Health Organization. Target product profile for a gambiense human African trypanosomiasis test to identify individuals to receive widened treatment. 2022: https://www.who.int/publications/i/item/9789240043299. Accessed 21 Nov 2022. 43. Sima N, Dujeancourt-Henry A, Perlaza BL, Ungeheuer MN, Rotureau B, Glover L. SHERLOCK4HAT: a CRISPR-based tool kit for diagnosis of human African trypanosomiasis. Biomedicine (2022.0) 85 104308 44. 44.Lejon V, Camara O, Camara M, Ilboudo H, Kaboré J, Compaoré CFA, et al. Passive case detection of human African trypanosomiasis in Guinea: symptoms and signs, rapid diagnostic test results and laboratory test results. DataSuds; 2022.
--- title: 'Beneficial effects of time and energy restriction diets on the development of experimental acute kidney injury in Rat: Bax/Bcl-2 and histopathological evaluation' authors: - Alireza Raji-Amirhasani - Mohammad Khaksari - Zahra Soltani - Shadan Saberi - Maryam Iranpour - Fatemeh Darvishzadeh Mahani - Zahra Hajializadeh - Nazanin Sabet journal: BMC Nephrology year: 2023 pmcid: PMC10026443 doi: 10.1186/s12882-023-03104-6 license: CC BY 4.0 --- # Beneficial effects of time and energy restriction diets on the development of experimental acute kidney injury in Rat: Bax/Bcl-2 and histopathological evaluation ## Abstract People’s lifestyles and, especially, their eating habits affect their health and the functioning of the organs in their bodies, including the kidneys. One’s diet influences the cells’ responses to stressful conditions such as acute kidney injury (AKI). This study aims to determine the preconditioning effects of four different diets: energy restriction (ER) diet, time restriction (TR) eating, intermittent fasting (IF), and high-fat diet (HF) on histopathological indices of the kidney as well as the molecules involved in apoptosis during AKI. Adult male rats underwent ER, TR, IF, and HF diets for eight weeks. Then, AKI was induced, and renal function indices, histopathological indices, and molecules involved in apoptosis were measured. In animals with AKI, urinary albumin excretion, serum urea, creatinine and, Bax/Bcl-2 ratio increased in the kidney, while renal eGFR decreased. ER and TR diets improved renal parameters and prevented an increase in the Bax/Bcl-2 ratio. The IF diet improved renal parameters but had no effect on the Bax/Bcl-2 ratio. On the other hand, the HF diet worsened renal function and increased the Bax/Bcl-2 ratio. Histopathological examination also showed improved kidney conditions in the ER and TR groups and more damage in the HF group. This study demonstrated that ER and TR diets have renoprotective effects on AKI and possibly cause the resistance of kidney cells to damage by reducing the Bax/Bcl-2 ratio and improving apoptotic conditions. ## Introduction Acute kidney injury (AKI), formerly known as acute renal failure (ARF) is a syndrome in which there is a sudden decline in renal function [1]. AKI is treatable, but mortality is still high (over $50\%$), so research into effective treatment to accelerate recovery and prevent AKI has received much attention [2]. AKI is associated with a rapid and reversible decrease in Glomerular Filtration Rate (GFR) and increased serum urea and creatinine over hours or days. Damage to the kidney tissue can be caused by several factors, including the kidney’s exposure to harmful substances, renal ischemia, oxidative stress, and inflammation in the kidney or urinary tract obstruction [3, 4]. In the last few decades, the prevalence of obesity has been increasing and has now reached unprecedented levels [5, 6]. Obesity is a disease associated with higher energy intake, which has genetic, metabolic, behavioral, social and environmental causes [7, 8]. Recent research has shown that energy restriction (ER) increases longevity and also minimizes functional impairment and age-related diseases [8–10]. Most studies have shown that ER leads to weight loss, but this weight loss is not sustained over time, and it still has remained as a challenge, because it is difficult to adhere to this method in the long term [11–13]. An ER diet usually includes periods of very low energy intake or fasting [14]. Therefore, ER includes different diets that are somehow associated with reducing energy consumption [14]. Animals subjected to ischemia, which already had an ER diet, were reported to have smaller infarct sizes and milder neurological damages [9, 15, 16]. This diet also caused ischemic tolerance in older rodents [8, 17]. Short-term ER preconditioning has been investigated as a possible strategy to prevent AKI induced by anticancer chemotherapeutic drugs, and it has been shown that the ER diet has protective effects against apoptosis in the kidney of DDP (cisplatin)-treated mice [18]. Short-term dietary interventions can cause resistance to pressure and stress in ischemic models of kidney and liver injury to maintain organ function [18, 19]. Beneficial metabolic effects of ER, possibly through increased SIRT1 expression and autophagy activity, delay apoptosis in renal tubular epithelial cells [18, 20]. Other diets that have been suggested for intervention are fasting or time restriction (TR) diets. In TR, a person can consume calories without restriction, but only for 8 to 10 h a day, but is in the fasting mode for the rest of the day and night [21]. Another type of diet is intermittent or periodic access to food, also known as intermittent fasting (IF), where the individual has access to food every other day [22]. Reducing food consumption (ER, TR, and IF) is different from special diet recommended by a nutritionist, such as the Mediterranean diet, because in the latter type of diet, one will consume certain substances, while in ER, TR, and IF diets, it is not specified which foods are to be eaten or which are forbidden, i.e. one can eat any food [8]. TR and IF have been shown to have effects similar to ER, increasing longevity, attenuating neurodegenerative diseases, cardiovascular disease, and increasing cerebral plasticity [8, 21, 23]. In addition, TR also lowers blood pressure and improves glucose control in humans [8, 24, 25]. Protection of the kidneys against injury by fasting diets (TR and IF) has also been reported [26–28]. It is interesting to know that in contrast to the advantages mentioned for the above three diets, many disadvantages have been reported about one of the today’s diets, namely the high-fat diet (HF) [29–32]. In today’s societies, there have been important changes in the distribution of food and its availability [33, 34]. The expansion of the western lifestyle, which is accompanied by the emergence of food preparation centers and the widespread presence of fast food establishments that produce energy-dense ready meals, has led to high-fat diets becoming common eating habits, and this type of eating is expanding both in urban and rural areas, which is worrying [33, 34]. For example, consuming too many calories [29] and high-fat foods has been shown to cause illness and enhanced vulnerability to diseases [30, 35, 36]. Long-term feeding with high-fat diet causes non-alcoholic fatty liver disease [37]. Also, high fat diet causes oxidative stress and mitochondrial dysfunction in the CNS [35]. Changes in metabolism caused by a high-fat diet aggravate brain dysfunction due to aging and also it accelerates nervous system diseases related to aging [35, 36]. Apoptosis is the programmed cell death seen in multicellular organisms. The intracellular ratio of apoptosis-promoting proteins such as Bax to anti-apoptotic proteins such as Bcl-2 determines the cell’s susceptibility to apoptosis [38, 39]. Apoptosis occurs during AKI in the kidney. In particular, apoptosis in the kidneys is detectable after ischemia, toxin exposure, inflammation, and sepsis [38]. Many of the above injuries occur simultaneously in humans in the intensive care unit [38, 40]. Renal endothelial and epithelial cells are affected by apoptosis in AKI, and this process is one of the leading causes of renal failure in AKI [38]. Previous studies have shown that dietary restrictions and their preconditioning affect apoptosis in organs, including the kidneys. For example, ER has been reported to have a protective effect against kidney damage, the mechanism of which includes exerting anti-apoptotic effects through improving oxidative conditions [18, 41]. Similar anti-apoptotic reports exist for TR [42–44] and IF [8, 45]. On the contrary, several studies have shown that the HF diet can induce apoptosis in the kidney and cause further damage during AKI [46–48]. According to the mentioned studies, manipulating the diet by changing the energy intake or the time of access to food can lead to delays in the onset and progression of diseases and lead to a long healthy life (increasing life expectancy) in most organisms [22, 49], while overeating or under-eating pose potential risks, increasing the incidence of chronic diseases and early death [49, 50]. Among the dietary interventions, the three ER, TR, and IF diets are better known than the other diets, and none of them has been proven entirely superior to the other. Besides, no study has yet compared these diets concerning their effect on AKI-induced damage. In the present study, we evaluated the effects of preconditioning by these diets and compared them with the HF diet regarding their effect on renal function indices, the severity of kidney damage measured through histopathological indices, and the role of molecules involved in apoptosis during experimental AKI. ## Animals and their grouping In this study, 60 adult male Wistar rats were used. The animals were kept in the natural environment for 12 h in the dark and 12 h in the light at a temperature of 25 °C. This study was conducted in accordance with the regulations of the Ethics Committee of Kerman University of Medical Sciences under the number IR.KMU.REC.1398.270. In this study, animals were divided into five general groups (Fig. 1), and the number of animals in each group was 12: Control group (CTL): In this group, animals had free access to food and, after two months, were divided into two subgroups: in one of the subgroups ($$n = 6$$) renal function indices were measured and after animal sacrifice (under deep anesthesia with intraperitoneal injection of pentobarbital sodium (150 mg/kg)), histopathological markers and molecules involved in apoptosis were measured in kidney tissue. In the other subgroup ($$n = 6$$), renal function indices were measured one day after the induction of acute kidney injury, and also after animal sacrifice, histopathological markers and molecules involved in apoptosis were measured in kidney tissue [51] (Fig. 1). Energy restriction group (ER): Animals in this group consumed $70\%$ of the daily food intake in the control group for two months [52, 53], and after two months, similar to the control group, they were divided into two subgroups (6 rats in each subgroup), and different factors were measured in them (Fig. 1). Time restriction group (TR): The animals in this group had free access to food for only 5 h a day for two months [22]. After two months, they were divided into two subgroups (6 rats in each subgroup), and they were treated like the control group (Fig. 1). Group of animals with intermittent or periodic access to food (IF): Animals in this group had free access to food every other day for two months (one-day food deprivation and one-day free access to food) [22, 54], and after two months, they were divided into two subgroups (6 rats in each subgroup), and they were treated like the control group (Fig. 1). High-fat diet (HF): Animals in this group received a high-fat diet for two months [55], and after two months, they were divided into two subgroups (6 rats in each subgroup) and different factors were measured in them as done in the control group (Fig. 1). Fig. 1Schematic of A animal grouping, B representation of the experimental protocol. AKI: acute kidney injury, SD: standard diet or control, ER: energy restriction, TR: time restriction feeding, IF: intermittent fasting, HF: high-fat diet, B: blood, U: urine ## Measuring the amount of food and applying the energy restriction diet To calculate the amount of food to be given to the ER group, first, the food intake in the CTL group, which had free access to food, was measured for one week, and then the average daily intake was measured. Then, $70\%$ of the daily intake in the control group, which had free access to food, was calculated and fed to the ER group for two months [53]. ## How to create a high-fat diet A combination of carbohydrates, proteins, and fats was used to make high-fat foods. The high-fat diet consisted of $58\%$ fat, $15\%$ protein, and $27\%$ carbohydrates (Royan institute, Iran), while the standard diet (Pars Animal Feed, Iran) consumed by other groups contained $6\%$ fat, $22\%$ protein, and $72\%$ carbohydrates [56]. ## Induction of acute kidney injury The animals were deprived of water for 24 h, and then a dose of hypertonic glycerol solution ($50\%$ dissolved in sterile saline) at 10 ml/kg was injected in equal portions in both lower limbs of the animals intramuscularly (IM) to induce acute kidney injury [3]. With this method, nephropathy developed rapidly, approximately 24 h after glycerol injection. Hypertonic glycerol causes rhabdomyolysis, which eventually leads to myoglobinuria, ischemia, and nephrotoxicity in the kidney [51]. In this study, urea, creatinine, and GFR (estimated by creatinine clearance) were measured to prove the induction of AKI. ## Measurement of serum and urine urea and creatinine Blood samples were collected from the choroid sinus and immediately centrifuged to isolate the serum one day before and two days after the induction of acute kidney injury to measure the above indices in serum. Also, to measure the same indicators in urine, urine was collected through the metabolic cage one day before and two days after the induction of acute kidney injury. Serum and urine urea and creatinine levels were measured using the ELISA method and according to the kit instructions (MAN, Iran). Values were reported in mg/dl [53, 57]. ## Measurement of 24-h albumin excretion Animals’ urine was collected using a metabolic cage 24 h before and 24 h after the induction of renal failure. Urine albumin concentration was also measured using the ELISA method and according to the kit instructions (MAN, Iran). Then, using the 24-h urine volume, the amount of albumin excreted in 24 h was calculated based on the following formula, and the excretory albumin levels were reported in mg/24 h [53]: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathrm{Albumin}\;\mathrm{excretion}\;\mathrm{in}\;24\;\mathrm h\:=\:(\mathrm{urine}\;\mathrm{volume}\;\mathrm{in}\;24\;\mathrm h\:\times\:\mathrm{concentration}\;\mathrm{of}\;\mathrm{urine}\;\mathrm{Albumin})$$\end{document}Albuminexcretionin24h=(urinevolumein24h×concentrationofurineAlbumin) ## Estimation of GFR Creatinine clearance was used to calculate estimated GFR (eGFR). Briefly, the animals were kept in a metabolic cage for 24 h, and their urine was collected, both before the induction of acute kidney injury and 24 h after it. Then, using a 24-h urine volume and serum and urine creatinine concentrations, eGFR was calculated based on the following formula [58, 59]:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$eGFR=\frac{\mathrm{urine}\;\mathrm{volume}\;\mathrm{in}\;24\;\mathrm h\times\mathrm{concentration}\;\mathrm{of}\;\mathrm{urine}\;\mathrm{creatinine}}{\mathrm{concentration}\;\mathrm{of}\;\mathrm{serum}\;\mathrm{creatinine}}$$\end{document}eGFR=urinevolumein24h×concentrationofurinecreatinineconcentrationofserumcreatinine Finally, eGFR calculations were corrected for body weight (BW) and expressed as ml/min/100 g BW [59]. ## Measurement of histopathological indicators in the kidney The animals were anesthetized with pentobarbital, and then their kidneys were extracted and placed in $10\%$ formalin buffer. After at least 72 h in formalin, they were washed with water, then dehydrated with alcohol, fixed in paraffin, and finally stained through the hematoxylin and eosin methods and examined microscopically [60]. A professional pathologist (who was blind to the study) determined score of damage in the kidney. The intensity of the injury was from 0 to 3 (no damage = 0, mild injury = 1, moderate injury = 2 and severe injury = 3). The injury parameters that were determined included: cellular vacuolization, congestion, tubular casts, inflammation, cellular necrosis and tubular dilation [61, 62]. ## Measurement of Bax and Bcl-2 proteins in kidney tissue After animals were sacrificed, all the animals’ kidneys were separated and placed in cold physiological saline and then homogenized using a homogenizer and centrifuged for 10 min to prepare supernatant fluid at 1000 rpm. Bax and Bcl-2 levels in kidney tissue were measured using the ELISA method according to the kit instructions (ZellBio, Germany). The values of these indices were reported in ng/mg protein [63, 64]. ## Kidney weight to body weight ratio Animal weights were measured using a digital scale (Gram Precision digital scale, Canada). Also, the kidney of all animals was isolated after sacrifice, the adipose tissue around them was cleared, and the weight of the kidneys was calculated by digital scales [65]. The kidney weight to body weight (KW/BW) ratio, was calculated by dividing left kidney weight by body weight and then was converted to percent. ## Method of calculating and analyzing data Two-way ANOVA test followed by Tukey’s test was used to compare the quantitative variables between the test groups if the assumptions were observed (normal data distribution), and the Kruskal–Wallis test was applied in case the assumptions were not fulfilled. A significance level of 0.05 was considered, and statistical analyses were performed using GraphPad Prism 8. ## Effects of diets on renal parameters in animals with and without AKI We investigated the effects of AKI and/or diets on serum urea and creatinine in the rats. Two way ANOVA showed a significant interaction between AKI and diets for serum urea [F [4,50] = 39.58, $P \leq 0.001$] and serum creatinine [F [4,50] = 32.03, $P \leq 0.001$]. In addition two-way ANOVA revealed that the main effect of AKI for serum urea [F [1,50] = 1276, $P \leq 0.001$] and serum creatinine [F [1,50] = 1011, $P \leq 0.001$]), and also the main effect of diets for serum urea [F [4,50] = 24.00, $P \leq 0.001$] and serum creatinine [F [4,50] = 32.98, $P \leq 0.001$] were significant. The post hoc Tukey test revealed, in animals without AKI, serum urea and creatinine were not significantly different from the control group in any of the groups. Twenty-four hours after AKI induction, serum urea and creatinine increased significantly in all groups with AKI, when compared to groups with similar diets but without AKI ($P \leq 0.001$). However, the increase in urea and creatinine in TR after AKI induction was smaller than their increase in the ER ($P \leq 0.01$), CTL, and HF ($P \leq 0.001$) groups and smaller than the IF group for creatinine ($P \leq 0.001$). In animals with AKI, urea showed a smaller increase in the ER and IF groups than CTL and HF groups ($P \leq 0.001$). Also, the increase in serum creatinine after AKI in the ER group was smaller than in the IF ($P \leq 0.01$), CTL, and HF ($P \leq 0.001$) groups. In the IF group, the increase in serum creatinine after AKI was smaller than in the CTL ($P \leq 0.05$) and HF ($P \leq 0.001$) groups. While Serum urea and creatinine levels after AKI in the HF group were higher than in the CTL group ($P \leq 0.001$ and $P \leq 0.05$, respectively) (Table 1).Table 1The effects of diets on renal parameters in animals with and without AKI Without AKI With AKI Groups CTLERTRIFHFCTLERTRIFHF Parameters Serum Urea (mg/dl)70.83 ± 3.3797.01 ± 5.3583.01 ± 6.0267.02 ± 4.8560.33 ± 4.18312 ± 10.51* 253.66 ± 6.88††† 204 ± 10.7###, ## 241 ± 15.22^^^ 383.5 ± 14.38&&& Serum Creatinine (mg/dl)0.45 ± 0.030.57 ± 0.010.53 ± 0.020.54 ± 0.030.61 ± 0.073.38 ± 0.14* 2.3 ± 0.15 †††, †† 1.76 ± 0.1 ###, ## 2.86 ± 0.07^^^, ^ 3.88 ± 0.17& Urine Albumin excretion (mg/24 h)0.51 ± 0.170.58 ± 0.150.41 ± 0.150.5 ± 0.180.75 ± 0.1728.16 ± 1.92* 24.5 ± 1.97†† 16.66 ± 0.8###, # 22.3 ± 0.66^^^, ^ 31.83 ± 2.5eGFR(ml/min/100 g bw)0.52 ± 0.010.48 ± 0.020.52 ± 0.020.52 ± 0.020.49 ± 0.020.1 ± 0.007* 0.18 ± 0.004† 0.26 ± 0.008###, # 0.11 ± 0.010.1 ± 0.004Body weight (g)244 ± 4.84193 ± 3.43ΩΩΩ 261 ± 5.13238 ± 5.32285 ± 5.49ΦΦΦ 245 ± 4.43192 ± 2.34257 ± 4.11240 ± 4.62286 ± 5.07Kidney weight (g)1.008 ± 0.0240.74 ± 0.023ΩΩΩ 1.001 ± 0.0420.916 ± 0.0361.188 ± 0.029ΦΦΦ 1.443 ± 0.031* 1.051 ± 0.0351.373 ± 0.0341.361 ± 0.0271.776 ± 0.043KW/BW ratio0.41 ± 0.0130.38 ± 0.0080.38 ± 0.0160.38 ± 0.0150.41 ± 0.0090.59 ± 0.01* 0.54 ± 0.016†† 0.53 ± 0.014## 0.57 ± 0.0150.62 ± 0.015Experimental groups ($$n = 6$$). Findings are reported based on Mean ± SEM CTL control, ER energy restriction, TR time restriction, IF intermittent fasting, HF high-fat diet, eGFR estimated glomerular filtration rate, KW/BW ratio Kidney weight to body weight ratio, AKI acute kidney injury * $P \leq 0.001$ VS. CTL without AKI for all parametersSerum urea in animals with AKI: ††† $P \leq 0.001$ VS. CTL and HF. ## $P \leq 0.01$ VS. ER. ### $P \leq 0.001$ VS. CTL and HF. ^^^ $P \leq 0.001$ VS. CTL and HF. &&& $P \leq 0.001$ VS. CTLSerum creatinine in animals with AKI: †† $P \leq 0.01$ VS. IF. ††† $P \leq 0.001$ VS. CTL and HF. ## $P \leq 0.01$ VS. ER. ### $P \leq 0.001$ VS. CTL, IF, and HF. ^ $P \leq 0.05$ VS. CTL. ^^^ $P \leq 0.001$ VS. HF. & $P \leq 0.05$ VS. CTLUrinary albumin excretion in animals with AKI: †† $P \leq 0.01$ VS. HF. # $P \leq 0.05$ VS. IF. ### $P \leq 0.001$ VS. CTL, ER, and HF. ^ $P \leq 0.05$ VS. CTL, ^^^ $P \leq 0.001$ VS. HFGFR in animals with AKI: † $P \leq 0.05$ VS. CTL, IF, and HF. # $P \leq 0.05$ VS. ER. ### $P \leq 0.001$ VS. IF, CTL, and HFBody weight and Kidney weight in animals without AKI: ΩΩΩ $P \leq 0.001$ VS. CTL. ΦΦΦ $P \leq 0.001$ VS. CTLKW/BW ratio in animals with AKI: †† $P \leq 0.01$ VS. HF. ## $P \leq 0.01$ VS. HF Two way ANOVA showed a significant interaction between AKI and diets for urinary albumin excretion [F [4,50] = 10.58, $P \leq 0.001$] and eGFR [F [4,50] = 10.39, $P \leq 0.001$]. In addition two-way ANOVA revealed that the main effect of AKI for urinary albumin excretion [F [1,50] = 963.1, $P \leq 0.001$] and eGFR [F [1,50] = 1213, $P \leq 0.001$], and also the main effect of diets for urinary albumin excretion [F [4,50] = 11.41, $P \leq 0.001$] and eGFR [F [4,50] = 10.57, $P \leq 0.001$] were significant. The post hoc Tukey test revealed that, none of the diets had a significant effect in urinary albumin excretion and eGFR in animals without AKI. Within 24 h after induction of AKI, urinary albumin excretion increased significantly in all groups with AKI compared to similar groups without AKI ($P \leq 0.001$). However, this increase in urinary albumin excretion was smaller in the TR group than in the IF ($P \leq 0.05$), CTL, ER, and HF ($P \leq 0.001$) groups. In addition, in animals with AKI, the increase in urinary albumin excretion was lower in the IF group than in the CTL ($P \leq 0.05$) and HF ($P \leq 0.001$) groups and smaller in the ER group than in the HF group ($P \leq 0.01$). In animals with AKI, eGFR showed a significant decrease in all groups compared to similar groups before AKI ($P \leq 0.001$). The decrease in eGFR after AKI was smaller in the TR group than in the ER ($P \leq 0.05$), CTL, IF, and HF groups ($P \leq 0.001$) and in the ER group compared to the CTL, IF, and HF groups ($P \leq 0.05$) (Table 1). Two way ANOVA showed that there was no significant interaction between AKI and diets for body weight [F [4,50] = 0.08078, $$P \leq 0.9879$$], but there was significant interaction for kidney weight [F [4,50] = 4.694, $$P \leq 0.0027$$]. In addition two-way ANOVA revealed that the main effect of AKI was not significant for body weight [F [1,50] = 0.0005306, $$P \leq 0.9817$$], but was significant for kidney weight [F [1,50] = 406.9, $P \leq 0.001$]. Also the main effect of diets was significant for both body weight [F [4,50] = 110.0, $P \leq 0.001$] and kidney weight [F [4,50] = 77.29, $P \leq 0.001$]. The post hoc Tukey test revealed that after two months of dieting, body and kidney weight decreased in the ER group and increased in the HF group compared to the CTL ($P \leq 0.001$) and didn’t change in TR and IF groups. There was a significant increase in kidney weight in all groups with AKI compared to similar groups without AKI ($P \leq 0.001$). We used the KW/BW ratio to compare renal edema among the groups. Two way ANOVA showed that there was no significant interaction between AKI and diets for KW/BW ratio [F [4,50] = 1.076, $$P \leq 0.3783$$], but the main effects of AKI and diets were significant for KW/BW ratio [F [1,50] = 408.8, $P \leq 0.001$] and [F [4,50] = 6.977, $P \leq 0.001$], respectively. The post hoc Tukey test revealed that in animals without AKI, the KW/BW ratio did not differ in study groups, but this ratio increased in animals with AKI in all groups compared to similar groups without AKI ($P \leq 0.001$). The increase of this index in the ER and TR groups was less than HF group ($P \leq 0.01$). Although the increase in this index after injury in ER and TR groups was less than CTL group, it was not statistically significant (Table 1). ## Comparison of histopathological indicators of the kidney in animals with and without AKI in the presence of different diets We investigated the effects of AKI and/or diets on histopathological indicators of the kidney. Two way ANOVA showed a significant interaction between AKI and diets for cell vacuolization, congestion, intratubular cast, tubular dilatation, inflammation, and necrosis [F [4,50] = 2.857, $$P \leq 0.0329$$], [F [4,50] = 2.863, $$P \leq 0.0326$$], [F [4,50] = 3.720, $$P \leq 0.01$$], [F [4,50] = 3.754, $$P \leq 0.0095$$], [F [4, 50] = 4.600, $$P \leq 0.0031$$] and [F [4, 50] = 6.560, $$P \leq 0.0003$$], respectively. In addition two-way ANOVA revealed that the main effect of AKI was significant for cell vacuolization, congestion, intratubular cast, tubular dilatation, inflammation, and necrosis [F [1,50] = 102.9, $P \leq 0.001$], [F [1,50] = 128.9, $P \leq 0.001$], [F [1,50] = 230.2, $P \leq 0.001$], [F [1,50] = 168.4, $P \leq 0.001$], [F [1,50] = 218.1, $P \leq 0.001$] and [F [1,50] = 228.9, $P \leq 0.001$], respectively. Also the main effect of diets was significant for cell vacuolization, congestion, intratubular cast, tubular dilatation, inflammation, and necrosis [F [4,50] = 5.089, $$P \leq 0.0016$$], [F [4,50] = 3.712, $$P \leq 0.0101$$], [F [4,50] = 2.792, $$P \leq 0.0360$$], [F [4,50] = 6.125, $P \leq 0.001$], [F [4,50] = 14.98, $P \leq 0.001$] and [F [4,50] = 10.23, $P \leq 0.001$], respectively. The post hoc Tukey test revealed that, the rates of cell vacuolization, congestion, intratubular cast, tubular dilatation, inflammation, and necrosis in the kidney, did not differ in groups without AKI, but these indices increased in animals with AKI, when compared to groups with similar diets but without AKI ($P \leq 0.001$); however, these indices in the ER and TR groups showed a smaller increase compared to the CTL group ($P \leq 0.05$). In other words, ER and TR diets could prevent the increase of injury symptoms after AKI to some extent. Also, the rate of inflammation and necrosis after AKI in the HF group was higher than in all the other groups ($P \leq 0.05$ for CTL and $P \leq 0.01$ for ER, TR and IF) (Fig. 2).Fig. 2Renal histopathological changes after AKI in experimental groups ($$n = 6$$). Findings are reported based on Mean ± SEM (magnification × 400 for vacuolization and × 100 for other indices). * $P \leq 0.001$ VS. CTL group without AKI. † $P \leq 0.05$ VS. CTL with AKI. # $P \leq 0.05$ VS. CTL with AKI. & $P \leq 0.05$ VS. CTL with AKI. && $P \leq 0.01$ VS. ER, TR and IF with AKI. CTL: control, ER: energy restriction, TR: time restriction, IF: intermittent fasting, HF: high-fat diet, AKI: acute kidney injury ## Comparison of renal Bax levels in experimental groups Two way ANOVA showed that there was no significant interaction between AKI and diets for Bax levels in the kidney [F [4,50] = 2.451, $$P \leq 0.0580$$], but the main effects of AKI and diets were significant for Bax levels [F [1,50] = 291.5, $P \leq 0.001$] and [F [4,50] = 42.36, $P \leq 0.001$], respectively. The post hoc Tukey test revealed that in animals without AKI, ER, and TR diets decreased renal Bax compared to the CTL group ($P \leq 0.05$), while HF increased this index compared to the ER, TR ($P \leq 0.001$), and IF ($P \leq 0.05$) diets. In animals with AKI, the amount of this protein increased in the kidney of all groups compared to similar groups without AKI ($P \leq 0.001$), but this increase was smaller in ER and TR groups compared to the CTL group ($P \leq 0.01$ and $P \leq 0.001$, respectively). This index was also lower than IF in the TR group after AKI ($P \leq 0.05$). In addition, the increase in Bax after AKI in the HF group was greater than in the ER, TR, and IF groups ($P \leq 0.001$) (Fig. 3).Fig. 3Renal Bax level (ng/mg protein) in experimental groups ($$n = 6$$). Findings are reported based on Mean ± SEM. Ω $P \leq 0.05$ VS. CTL without AKI, ¤ $P \leq 0.05$ VS. CTL without AKI. ΦΦΦ $P \leq 0.001$ VS. ER and TR without AKI. Φ $P \leq 0.05$ VS. IF without AKI. * $P \leq 0.001$ VS. CTL without AKI. †† $P \leq 0.01$ VS. CTL with AKI. ### $P \leq 0.001$ VS. CTL with AKI. # $P \leq 0.05$ VS. IF with AKI. &&& $P \leq 0.001$ VS. ER, TR and IF with AKI. CTL: control, ER: energy restriction, TR: time restriction, IF: intermittent fasting, HF: high-fat diet, AKI: acute kidney injury ## Comparison of renal Bcl-2 levels in experimental groups Two way ANOVA showed that there was no significant interaction between AKI and diets for Bcl-2 levels in the kidney [F [4,50] = 1.388, $$P \leq 0.2517$$], but the main effects of AKI and diets were significant for Bcl-2 levels [F [1,50] = 205.9, $P \leq 0.001$] and [F [4,50] = 46.20, $P \leq 0.001$], respectively. The post hoc Tukey test revealed that in animals without AKI, the ER and TR diets increased renal Bcl-2 compared to the CTL ($P \leq 0.05$), IF and HF groups ($P \leq 0.001$), and the HF diet decreased this protein in comparison with the CTL group ($P \leq 0.05$) (Fig. 4). In animals with AKI, the amount of this protein decreased in the kidney compared to similar groups without AKI ($P \leq 0.001$), but this decrease in ER and TR groups compared to the CTL, IF ($P \leq 0.05$, $P \leq 0.01$, respectively), and HF ($P \leq 0.001$) groups was smaller (Fig. 4).Fig. 4Renal Bcl-2 level (ng/mg protein) in experimental groups ($$n = 6$$). Findings are reported based on Mean ± SEM. Ω $P \leq 0.$ 05 VS. CTL without AKI. ΩΩΩ $P \leq 0.$ 001 VS. IF and HF without AKI. ¤ $P \leq 0.05$ VS. CTL without AKI. ¤¤¤ $P \leq 0.001$ VS. IF and HF without AKI. Φ $P \leq 0.05$ VS. CTL without AKI. * $P \leq 0.001$ VS. CTL without AKI. † $P \leq 0.05$ VS. CTL and IF with AKI. ## $P \leq 0.01$ VS. CTL and IF with AKI. &&& $P \leq 0.001$, compared to ER and TR with AKI. CTL: control, ER: energy restriction, TR: time restriction, IF: intermittent fasting, HF: high-fat diet, AKI: acute kidney injury ## Comparison of renal Bax/Bcl-2 ratio in experimental groups Two way ANOVA showed a significant interaction between AKI and diets for Bax/Bcl-2 ratio in the kidney [F [4,50] = 19.48, $P \leq 0.001$]. In addition two-way ANOVA revealed that the main effects of AKI and diets were significant for Bax/Bcl-2 ratio in the kidney [F [1,50] = 372.9, $P \leq 0.001$] and [F [4,50] = 75.42, $P \leq 0.001$], respectively. The post hoc Tukey test revealed that in animals without AKI, the renal Bax/Bcl-2 ratio in the IF group was higher than in the ER and TR groups ($P \leq 0.05$), but it was not significantly different from the CTL group. Also in animals without AKI, this index in the HF group was higher than in the ER, TR ($P \leq 0.001$), and CTL ($P \leq 0.05$) groups. In animals with AKI, this ratio increased in all groups compared to similar groups without AKI ($P \leq 0.001$), but this increase was smaller in the ER and TR groups than in the other groups ($P \leq 0.001$). Also, this index was higher after AKI in the HF group than in the CTL and IF groups ($P \leq 0.001$) (Fig. 5).Fig. 5Renal Bax/Bcl-2 ratio in experimental groups ($$n = 6$$). Findings are reported based on Mean ± SEM. Ψ $P \leq 0.05$ VS. ER and TR without AKI. Φ $P \leq 0.05$ VS. CTL without AKI. ΦΦΦ $P \leq 0.001$ VS. ER and TR without AKI. * $P \leq 0.001$ VS. CTL without AKI. ††† $P \leq 0.001$ VS. CTL, IF and HF with AKI. ### $P \leq 0.001$ VS. CTL, IF, and HF with AKI. &&& $P \leq 0.001$ VS. CTL and IF with AKI. CTL: control, ER: energy restriction, TR: time restriction, IF: intermittent fasting, HF: high-fat diet, AKI: acute kidney injury ## Discussion The results of our study, which was conducted to investigate the effects of four different diets, ER, TR, IF, and HF, on biochemical and histopathological indices and molecules affecting apoptosis in the kidney, showed that the application of these diets had different effects on the level and severity of damage symptoms before and after experimental AKI. The main findings of this study are 1- After AKI, the ER, TR, and IF diets prevented an increase in serum urea and creatinine, while HF increased them. 2- After AKI, the TR, and IF groups showed decreased urinary albumin excretion. 3- eGFR, after AKI, showed a smaller decrease in the ER and TR groups. 4- Histopathological indices after AKI showed abnormal kidney status, while these indices in the ER and TR groups showed better kidney health status. 5- Before AKI, the Bax/Bcl-2 ratio increased in the HF group, indicating apoptosis. After AKI, this ratio increased in all groups, although this increase was smaller in the ER and TR groups and greater in the HF group, indicating a worsening of apoptotic conditions in the latter group. In the present study, after AKI, renal injury indices (serum creatinine and urea, urinary albumin excretion, and eGFR) changed, while the ER, TR, and IF diets prevented the increase of serum creatinine and urea and urinary albumin excretion and a decrease in eGFR. However, the HF diet did not have a positive effect but worsened the injury indices. Consistent with our results, many studies have shown that different models of dietary restriction are effective in protecting the kidney against AKI: normalization of creatinine levels and reduction of urea and protein excretion in urine after AKI by ER [18, 66], protection of the kidney against I/R damage by fasting [26], increased renal resistance to damage by ER and TR [27], and renoprotective effects of IF in diabetes [28]. Also, previous work by our research team has shown that ER and TR diets can prevent further damage during AKI [67]. Possible mechanism(s) of the positive effects of these diets in reducing kidney damage are: improvement of mitochondrial function [26], effect on various transcription factors including FOXO3, HNF4A, HMGA1, and HSF1 [27], increased expression of SIRT1, SIRT3, and activation SOD2 [18, 67, 68], decreased TGF-β1 and mitochondrial superoxide production and increased GSH concentration [67, 69]. In line with the results of our study, it has been reported that a high-fat diet makes the kidneys more vulnerable when exposed to stress [70]. This diet also causes inflammation in the kidney [71] and further kidney damage during ischemia–reperfusion [72]. Recently, Jeon et al. reported that the HF diet increased susceptibility to ischemic AKI by increasing proinflammatory cytokines [73]. Another study has shown that a high-fat diet causes damage to the kidney structure, including dilation of the tubules and Bowman’s capsule [74]. Further reasons for the vulnerability caused by this type of diet, disruption of the oxidative balance system and an increase in proinflammatory cytokines such as TNF-α, IFN-γ, MCP-1, IL1β, and inflammatory conditions, decreased renal VEGF, increased intra-renal CD8 T cells, renin-angiotensin system imbalance, and insulin resistance, have been reported [73–77]. Histopathological results in animal with AKI indicate that the symptoms of injury are fewer in the ER and TR groups and more in the HF group than in the other groups. Consistent with the results of our study, histopathological studies of animals on the ER diet have reduced renal damage due to aging by reducing cast formation and renal inflammation [78, 79]. A study conducted by Ning et al. [ 2013] showed that the ER diet prevents dilatation of tubules during cisplatin-induced AKI [18]. Recent studies have also reported that ER reduces kidney damage, such as vacuolization, during AKI [80, 81]. Jongbloed et al. observed that fasting reduced necrosis symptoms in histopathological observations in AKI [82]. It has also been reported that the TR diet can improve histological parameters in other tissues, including the liver [83]. Contrary to our results, it has been reported that fasting does not reduce necrosis symptoms after renal ischemia–reperfusion [84]. The inconsistency between these results and ours is probably the short fasting or energy restriction periods, short-term follow-up, and the AKI model. There are reports about the harmful effects of HF on the kidneys and its higher vulnerability when exposed to stress [70], causing inflammation in the kidneys [71] and further kidney damage during ischemia–reperfusion [72]. Recently, Jeon et al. [ 2021] reported that HF could predispose kidneys to AKI, and histologically, HF exacerbated necrosis in the renal tissue [73]. In another study, Eddy et al. reported that HF caused inflammation and eventually interstitial fibrosis of kidney tissue [85]. Also, vacuolization of tubular cells [48], dilatation of tubules, further damage to glomeruli and proximal tubules [74, 86], severe inflammation, destruction of the basement membrane of epithelial cells of renal tubules, and loss of tubular epithelium [87] resulting from an HF diet have been reported in other histopathological studies. It was stated above that diets were effective in reducing apoptosis. To determine the effect of different diets on apoptosis, this study also measured Bax, Bcl-2, and the ratio of these two proteins (Bax/Bcl-2) in the kidney. The results showed that in animal without AKI, the ER and TR diets decreased Bax and increased Bcl-2 and, conversely, the HF diet had the opposite effect. However, the ratio of these two proteins remained constant in the ER and TR groups and increased in the HF group. After AKI, renal Bax levels increased, Bcl-2 decreased, and the ratio of the two also increased, indicating apoptosis in the kidney. The ER and TR diets reduced apoptosis by preventing changes in these two proteins and their proportions. However, the condition worsened in the HF diet, and the Bax/Bcl-2 ratio increased, indicating an increase in renal apoptosis in this group after AKI. Consistent with our results, it has been reported that the ER diet has a protective role against AKI by reducing the Bax/Bcl-2 ratio, ultimately reducing apoptosis [18, 66], and it has also been reported to prevent the increase in renal Bax and decrease in Bcl-2 and thus inhibit apoptosis in elder animals [41]. The effect of ER on apoptosis has also been reported for other tissues: decreased Bax/Bcl-2 ratio in the liver and delayed aging [88], increased Bcl-2 and Bcl-2/Bax ratio, and suppression of neuroapoptosis in the brain after brain injury [89], increased Bcl-2 and Bcl-2/Bax ratio in the heart, and protection of the heart against ischemia [90]. Regarding the effects of diets other than ER, it has been reported that the IF diet has a neuroprotective effect on acute spinal cord injury by reducing the Bax/Bcl-2 ratio [45]. The TR diet has been shown to have protective effects on the heart by attenuating apoptosis, regulating the autophagy process, and preserving cells [42]. Another study found that preoperative fasting was protective against renal IRI (ischemic reperfusion injury) by inhibiting apoptosis [91]. Ischemic apoptosis in the kidney is the leading cause of injury, so inhibition of the apoptotic process prevents inflammation and subsequent kidney damage [26]. Numerous studies have also reported that the HF diet can increase Bax, decrease Bcl-2, and ultimately induce apoptosis in the kidney [46–48]. It has already been reported that ER and TR diets increase the survival of cells against damage by increasing Sirt1 and improving the antioxidant system, but IF diet had no effect [67]. It can be said that among these diets, maybe TR can be associated with less difficulty for humans and is recommended. The other two diets (ER and IF), perhaps due to their limitations and difficulty, should be entered into human studies more carefully, and the individual’s health status should also be taken into account so that such diets do not cause harm. The limitations of this study include: the small number of animals in each group and the limitations of using creatinine clearance as an estimate of GFR. ## Conclusion Of the four diets used in the present study, the ER and TR diets were more effective in reducing kidney damage during AKI, although the IF diet also reduced serum urea, creatinine, and urinary albumin excretion. The renoprotective mechanism of the ER and TR diets appears to be similar, and histopathological results confirm it. Both diets also reduced the Bax/Bcl-2 ratio, possibly reducing apoptosis. On the other hand, the effects of the HF diet on histopathological indices and Bax/Bcl-2 ratio were opposite the other three diets, i.e., it worsen AKI. This study emphasizes the role of diets in pathophysiological conditions and suggests that future studies aimed at more prolonged periods of fasting or energy restriction and more extended follow-up periods may shed more light on the present findings. ## References 1. 1.Mahani FD, Khaksari M, Raji-Amirhasani A. Renoprotective effects of estrogen on acute kidney injury: the role of SIRT1. Int Urol Nephrol. 2021;53(11):2299–310. 2. Park CH, Tanaka T, Cho EJ, Park JC, Shibahara N, Yokozawa T. Glycerol-induced renal damage improved by 7-O-galloyl-D-sedoheptulose treatment through attenuating oxidative stress*. Biol Pharm Bull* (2012.0) 35 34-41. DOI: 10.1248/bpb.35.34 3. Laham SAA. The curative effects of methylsulfonylmethane against glycerol-induced acute renal failure in rats. Braz J Pharm Sci (2018.0) 5 e17442 4. 4.Raji-Amirhasani A, Khaksari M, Mahani FD, Hajializadeh Z. Activators of SIRT1 in the kidney and protective effects of SIRT1 during acute kidney injury (AKI)(effect of SIRT1 activators on acute kidney injury). Clin Exp Nephrol. 2021;25(8):807–21. 5. Wright SM, Aronne LJ. Causes of obesity. Abdom Radiol (2012.0) 37 730-732. DOI: 10.1007/s00261-012-9862-x 6. Flegal K, Carroll MD, Ogden CL, Johnson CL. Prevalence and trends in obesity among US adults (2000.0) 1999 1723-1727 7. Jebb S. Obesity: causes and consequences. Women’s Health Med (2004.0) 1 38-41. DOI: 10.1383/wohm.1.1.38.55418 8. Vemuganti R, Arumugam TV. Molecular mechanisms of intermittent fasting-induced ischemic tolerance. Cond Med (2020.0) 3 9. PMID: 34278242 9. Ciobanu O, Elena Sandu R, Tudor Balseanu A, Zavaleanu A, Gresita A, Petcu EB. Caloric restriction stabilizes body weight and accelerates behavioral recovery in aged rats after focal ischemia. Aging Cell (2017.0) 16 1394-1403. DOI: 10.1111/acel.12678 10. Wan R, Ahmet I, Brown M, Cheng A, Kamimura N, Talan M. Cardioprotective effect of intermittent fasting is associated with an elevation of adiponectin levels in rats. J Nutr Biochem (2010.0) 21 413-417. DOI: 10.1016/j.jnutbio.2009.01.020 11. Obert J, Pearlman M, Obert L, Chapin S. Popular weight loss strategies: a review of four weight loss techniques. Curr Gastroenterol Rep (2017.0) 19 1-4. DOI: 10.1007/s11894-017-0603-8 12. Liu D, Huang Y, Huang C, Yang S, Wei X, Zhang P. Calorie restriction with or without time-restricted eating in weight loss. N Engl J Med (2022.0) 386 1495-1504. DOI: 10.1056/NEJMoa2114833 13. O’Connor SG, Boyd P, Bailey CP, Shams-White MM, Agurs-Collins T, Hall K. Perspective: time-restricted eating compared with caloric restriction: potential facilitators and barriers of long-term weight loss maintenance. Adv Nutr (2021.0) 12 325-333. DOI: 10.1093/advances/nmaa168 14. Harris L, McGarty A, Hutchison L, Ells L, Hankey C. Short-term intermittent energy restriction interventions for weight management: a systematic review and meta-analysis. Obes Rev (2018.0) 19 1-13. DOI: 10.1111/obr.12593 15. Ran M, Li Z, Yang L, Tong L, Zhang L, Dong H. Calorie restriction attenuates cerebral ischemic injury via increasing SIRT1 synthesis in the rat. Brain Res (2015.0) 1610 61-68. DOI: 10.1016/j.brainres.2015.03.043 16. Varendi K, Airavaara M, Anttila J, Vose S, Planken A, Saarma M. Short-term preoperative dietary restriction is neuroprotective in a rat focal stroke model. PLoS ONE (2014.0) 9 e93911. DOI: 10.1371/journal.pone.0093911 17. Manzanero S, Gelderblom M, Magnus T, Arumugam TV. Calorie restriction and stroke. Exp Transl Stroke Med (2011.0) 3 1-13. DOI: 10.1186/2040-7378-3-8 18. Ning Y-C, Cai G-Y, Zhuo L, Gao J-J, Dong D, Cui S-Y. Beneficial effects of short-term calorie restriction against cisplatin-induced acute renal injury in aged rats. Nephron Exp Nephrol (2013.0) 124 19-27. PMID: 24401898 19. Peng W, Robertson L, Gallinetti J, Mejia P, Vose S, Charlip A. Surgical stress resistance induced by single amino acid deprivation requires Gcn2 in mice. Sci Transl Med (2012.0) 4 118ra11-ra11. DOI: 10.1126/scitranslmed.3002629 20. Kaushal GP, Kaushal V, Herzog C, Yang C. Autophagy delays apoptosis in renal tubular epithelial cells in cisplatin cytotoxicity. Autophagy (2008.0) 4 710-712. DOI: 10.4161/auto.6309 21. Moro T, Tinsley G, Bianco A, Marcolin G, Pacelli QF, Battaglia G. Effects of eight weeks of time-restricted feeding (16/8) on basal metabolism, maximal strength, body composition, inflammation, and cardiovascular risk factors in resistance-trained males. J Transl Med (2016.0) 14 1-10. DOI: 10.1186/s12967-016-1044-0 22. Di Francesco A, Di Germanio C, Bernier M, de Cabo R. A time to fast. Science (2018.0) 362 770-775. DOI: 10.1126/science.aau2095 23. Fann DY-W, Ng GYQ, Poh L, Arumugam TV. Positive effects of intermittent fasting in ischemic stroke. Exp Gerontol. (2017.0) 89 93-102. DOI: 10.1016/j.exger.2017.01.014 24. Jamshed H, Beyl RA, Della Manna DL, Yang ES, Ravussin E, Peterson CM. Early time-restricted feeding improves 24-hour glucose levels and affects markers of the circadian clock, aging, and autophagy in humans. Nutrients (2019.0) 11 1234. DOI: 10.3390/nu11061234 25. Sutton EF, Beyl R, Early KS, Cefalu WT, Ravussin E, Peterson CM. Early time-restricted feeding improves insulin sensitivity, blood pressure, and oxidative stress even without weight loss in men with prediabetes. Cell Metab (2018.0) 27 1212-21.e3. DOI: 10.1016/j.cmet.2018.04.010 26. Verweij M, Sluiter W, van den Engel S, Jansen E, Ijzermans JN, de Bruin RW. Altered mitochondrial functioning induced by preoperative fasting may underlie protection against renal ischemia/reperfusion injury. J Cell Biochem. (2013.0) 114 230-7. DOI: 10.1002/jcb.24360 27. Jongbloed F, Saat TC, Verweij M, Payán-Gómez C, Hoeijmakers JH, van den Engel S. A signature of renal stress resistance induced by short-term dietary restriction, fasting, and protein restriction. Sci Rep (2017.0) 7 1-12. DOI: 10.1038/srep40901 28. Tikoo K, Tripathi DN, Kabra DG, Sharma V, Gaikwad AB. Intermittent fasting prevents the progression of type I diabetic nephropathy in rats and changes the expression of Sir2 and p53. FEBS Lett (2007.0) 581 1071-1078. DOI: 10.1016/j.febslet.2007.02.006 29. Einer C, Leitzinger C, Lichtmannegger J, Eberhagen C, Rieder T, Borchard S. A high-calorie diet aggravates mitochondrial dysfunction and triggers severe liver damage in Wilson disease rats. Cell Mol Gastroenterol Hepatol (2019.0) 7 571-596. DOI: 10.1016/j.jcmgh.2018.12.005 30. Declèves A-E, Mathew AV, Cunard R, Sharma K. AMPK mediates the initiation of kidney disease induced by a high-fat diet. J Am Soc Nephrol (2011.0) 22 1846-1855. DOI: 10.1681/ASN.2011010026 31. McCabe LR, Irwin R, Tekalur A, Evans C, Schepper JD, Parameswaran N. Exercise prevents high fat diet-induced bone loss, marrow adiposity and dysbiosis in male mice. Bone (2019.0) 118 20-31. DOI: 10.1016/j.bone.2018.03.024 32. Zhang L, Du J, Yano N, Wang H, Zhao YT, Dubielecka PM. Sodium butyrate protects against high fat diet-induced cardiac dysfunction and metabolic disorders in type II diabetic mice. J Cell Biochem (2017.0) 118 2395-2408. DOI: 10.1002/jcb.25902 33. Assifi MM, Eibl G. Western diet-induced pancreatic cancer (2012.0) 327-338 34. Popkin BM. Nutrition transition and the global diabetes epidemic. Curr DiabRep (2015.0) 15 1-8 35. Langley MR, Yoon H, Kim HN, Choi C-I, Simon W, Kleppe L. High fat diet consumption results in mitochondrial dysfunction, oxidative stress, and oligodendrocyte loss in the central nervous system. Biochim Biophys Acta Mol Basis Dis (2020.0) 1866 165630. DOI: 10.1016/j.bbadis.2019.165630 36. Bruce-Keller AJ, Keller JN, Morrison CD. Obesity and vulnerability of the CNS. Biochim Biophys Acta Mol Basis Dis (2009.0) 1792 395-400. DOI: 10.1016/j.bbadis.2008.10.004 37. Velázquez KT, Enos RT, Bader JE, Sougiannis AT, Carson MS, Chatzistamou I. Prolonged high-fat-diet feeding promotes non-alcoholic fatty liver disease and alters gut microbiota in mice. World J Hepatol (2019.0) 11 619. DOI: 10.4254/wjh.v11.i8.619 38. Havasi A, Borkan SC. Apoptosis and acute kidney injury. Kidney Int (2011.0) 80 29-40. DOI: 10.1038/ki.2011.120 39. Korsmeyer SJ. BCL-2 gene family and the regulation of programmed cell death. Can Res (1999.0) 59 1693s-s1700 40. Klenzak J, Himmelfarb J. Sepsis and the kidney. Crit Care Clin (2005.0) 21 211-222. DOI: 10.1016/j.ccc.2005.01.002 41. Lee JH, Jung KJ, Kim JW, Kim HJ, Yu BP, Chung HY. Suppression of apoptosis by calorie restriction in aged kidney. Exp Gerontol (2004.0) 39 1361-1368. DOI: 10.1016/j.exger.2004.06.015 42. Abas E, Sabry MM. Intermittent fasting attenuates apoptosis, modulates autophagy and preserves telocytes in doxorubicin induced cardiotoxicity in albino rats: a histological study. Egyptian Journal of Histology (2020.0) 43 663-683 43. Upadhyay A, Anjum B, Godbole NM, Rajak S, Shukla P, Tiwari S. Time-restricted feeding reduces high-fat diet associated placental inflammation and limits adverse effects on fetal organ development. Biochem Biophys Res Commun (2019.0) 514 415-421. DOI: 10.1016/j.bbrc.2019.04.154 44. Sabet N, Soltani Z, Khaksari M, Iranpour M, Afshar RM, Mehdiabadi FM. The effect of two types of diet on apoptosis indexes, lipid profile and histopathological outcome in acute kidney injury during exercise. BMC Nephrol (2022.0) 23 1-15. DOI: 10.1186/s12882-022-02938-w 45. Yuan W, Zhu Y. 141. Autophagy induction contributes to the neuroprotective effect of intermittent fasting in experimental spinal cord injury. Spine J (2019.0) 19 S68. DOI: 10.1016/j.spinee.2019.05.155 46. Costa MC, Lima TFO, Arcaro CA, Inacio MD, Batista-Duharte A, Carlos IZ. Trigonelline and curcumin alone, but not in combination, counteract oxidative stress and inflammation and increase glycation product detoxification in the liver and kidney of mice with high-fat diet-induced obesity. J Nutr Biochem (2020.0) 76 108303. DOI: 10.1016/j.jnutbio.2019.108303 47. 47.Eleazu C, Suleiman JB, Othman ZA, Zakaria Z, Nna VU, Hussain NHN, et al. Bee bread attenuates high fat diet induced renal pathology in obese rats via modulation of oxidative stress, downregulation of NF-kB mediated inflammation and Bax signalling. Arch Physiol Biochem. 2022;128(4):1088–104. 48. Guo H, Li H, Wang B, Ding W, Ling L, Yang M. Protective effects of glucagon-like peptide-1 analog on renal tubular injury in mice on high-fat diet. Cell Physiol Biochem (2017.0) 41 1113-1124. DOI: 10.1159/000464118 49. Panda S. Circadian physiology of metabolism. Science (2016.0) 354 1008-1015. DOI: 10.1126/science.aah4967 50. Mattson MP, Allison DB, Fontana L, Harvie M, Longo VD, Malaisse WJ. Meal frequency and timing in health and disease. Proc Natl Acad Sci (2014.0) 111 16647-16653. DOI: 10.1073/pnas.1413965111 51. Sun X, Luan Q, Qiu S. Valsartan prevents glycerol-induced acute kidney injury in male albino rats by downregulating TLR4 and NF-κB expression. Int J Biol Macromol (2018.0) 119 565-571. DOI: 10.1016/j.ijbiomac.2018.07.149 52. Zhaleh F, Amiri F, Mohammadzadeh-Vardin M, Bahadori M, Harati MD, Roudkenar MH. Nuclear factor erythroid-2 related factor 2 overexpressed mesenchymal stem cells transplantation, improves renal function, decreases injuries markers and increases repair markers in glycerol-induced Acute kidney injury rats. Iran J Basic Med Sci (2016.0) 19 323. PMID: 27114803 53. Kume S, Uzu T, Horiike K, Chin-Kanasaki M, Isshiki K, Araki SI, Sugimoto T, Haneda M, Kashiwagi A, Koya D. Calorie restriction enhances cell adaptation to hypoxia through Sirt1-dependent mitochondrial autophagy in mouse aged kidney. J Clin Investigation (2010.0) 120 1043-55. DOI: 10.1172/JCI41376 54. Kouda K, Iki M. Beneficial effects of mild stress (hormetic effects): dietary restriction and health. J Physiol Anthropol (2010.0) 29 127-132. DOI: 10.2114/jpa2.29.127 55. Strakovsky RS, Zhang X, Zhou D, Pan YX. Gestational high fat diet programs hepatic phosphoenolpyruvate carboxykinase gene expression and histone modification in neonatal offspring rats. J Physiol (2011.0) 589 2707-2717. DOI: 10.1113/jphysiol.2010.203950 56. Farhadi Z, Khaksari M, Azizian H, Dabiri S, Fallah H, Nozari M. Aging is associated with loss of beneficial effects of estrogen on leptin responsiveness in mice fed high fat diet: Role of estrogen receptor α and cytokines. Mech Ageing Dev (2020.0) 186 111198. DOI: 10.1016/j.mad.2019.111198 57. Zhang Y, Du Y, Yu H, Zhou Y, Ge F. Protective Effects of Ophiocordyceps lanpingensis on Glycerol-Induced Acute Renal Failure in Mice. J Immunol Res (2017.0) 2017 2012585. DOI: 10.1155/2017/2012585 58. Hocher B, Dembowski C, Slowinski T, Friese ST, Schwarz A, Siren A-L. Impaired sodium excretion, decreased glomerular filtration rate and elevated blood pressure in endothelin receptor type B deficient rats. J Mol Med (2001.0) 78 633-641. DOI: 10.1007/s001090000158 59. Erdem A, Gündogan NÜ, Usubütün A, Kılınç K, Erdem ŞR, Kara A. The protective effect of taurine against gentamicin-induced acute tubular necrosis in rats. Nephrol Dial Transplant (2000.0) 15 1175-1182. DOI: 10.1093/ndt/15.8.1175 60. Atwa A, Hegazy R, Shaffie N, Yassin N, Kenawy S. Protective Effects of Vasodilatory Βeta-Blockers Carvedilol and Nebivolol against Glycerol Model of Rhabdomyolysis-Induced Acute Renal Failure in Rats. Open Access Maced J Med Sci (2016.0) 4 329. DOI: 10.3889/oamjms.2016.082 61. Su H, Yang M, Wan C, Yi L-X, Tang F, Zhu H-Y. Renal histopathological analysis of 26 postmortem findings of patients with COVID-19 in China. Kidney Int (2020.0) 98 219-227. DOI: 10.1016/j.kint.2020.04.003 62. Mahani FD, Khaksari M, Iranpour M, Soltani Z, Raji-Amirhasani A, Hajializadeh Z. Effects of caloric and time restriction diets on kidney health in rat model of postmenopausal acute kidney injury: An apoptosis and histopathological study. Iran J Basic Med Sci (2022.0) 25 390. PMID: 35656185 63. Abdel-Magied N, Elkady AA. Possible curative role of curcumin and silymarin against nephrotoxicity induced by gamma-rays in rats. Exp Mol Pathol (2019.0) 111 104299. DOI: 10.1016/j.yexmp.2019.104299 64. Hao Y, Huang J, Gu Y, Liu C, Li H, Liu J. Metallothionein deficiency aggravates depleted uranium-induced nephrotoxicity. Toxicol Appl Pharmacol (2015.0) 287 306-315. DOI: 10.1016/j.taap.2015.06.019 65. Khaksari M, Mahmmodi R, Shahrokhi N, Shabani M, Joukar S, Aqapour M. The effects of shilajit on brain edema, intracranial pressure and neurologic outcomes following the traumatic brain injury in rat. Iran J Basic Med Sci (2013.0) 16 858. PMID: 23997917 66. Estrela GR, Wasinski F, Batista RO, Hiyane MI, Felizardo RJ, Cunha F. Caloric restriction is more efficient than physical exercise to protect from cisplatin nephrotoxicity via PPAR-alpha activation. Front Physiol (2017.0) 8 116. DOI: 10.3389/fphys.2017.00116 67. Raji-Amirhasani A, Khaksari M, Shahrokhi N, Soltani Z, Nazari-Robati M, Mahani FD. Comparison of the effects of different dietary regimens on susceptibility to experimental acute kidney injury: the roles of SIRT1 and TGF-β1. Nutrition (2022.0) 96 111588. DOI: 10.1016/j.nut.2022.111588 68. Qiu X, Brown K, Hirschey MD, Verdin E, Chen D. Calorie restriction reduces oxidative stress by SIRT3-mediated SOD2 activation. Cell Metab (2010.0) 12 662-667. DOI: 10.1016/j.cmet.2010.11.015 69. Stankovic M, Mladenovic D, Ninkovic M, Vucevic D, Tomasevic T, Radosavljevic T. Effects of caloric restriction on oxidative stress parameters. Gen Physiol Biophys (2013.0) 32 277-283. DOI: 10.4149/gpb_2013027 70. Zhang N, Li Z, Xu K, Wang Y, Wang Z. Resveratrol protects against high-fat diet induced renal pathological damage and cell senescence by activating SIRT1. Biol Pharm Bull (2016.0) 39 1448-1454. DOI: 10.1248/bpb.b16-00085 71. Sleeman P, Patel NN, Lin H, Walkden GJ, Ray P, Welsh GI. High fat feeding promotes obesity and renal inflammation and protects against post cardiopulmonary bypass acute kidney injury in swine. Crit Care (2013.0) 17 1-15. DOI: 10.1186/cc13092 72. Johnson JB, Summer W, Cutler RG, Martin B, Hyun D-H, Dixit VD. Alternate day calorie restriction improves clinical findings and reduces markers of oxidative stress and inflammation in overweight adults with moderate asthma. Free Radical Biol Med (2007.0) 42 665-674. DOI: 10.1016/j.freeradbiomed.2006.12.005 73. Jeon J, Lee K, Yang KE, Lee JE, Kwon GY, Huh W. Dietary Modification Alters the Intrarenal Immunologic Micromilieu and Susceptibility to Ischemic Acute Kidney Injury. Front Immunol (2021.0) 12 688. DOI: 10.3389/fimmu.2021.621176 74. Shamseldeen AM, Ali Eshra M, Ahmed Rashed L, FathyAmer M, Elham Fares A, Samir KS. Omega-3 attenuates high fat diet-induced kidney injury of female rats and renal programming of their offsprings. Arch Physiol Biochem (2019.0) 125 367-377. DOI: 10.1080/13813455.2018.1471511 75. Noeman SA, Hamooda HE, Baalash AA. Biochemical study of oxidative stress markers in the liver, kidney and heart of high fat diet induced obesity in rats. Diabetol Metab Syndr (2011.0) 3 1-8. DOI: 10.1186/1758-5996-3-17 76. Arany I, Hall S, Reed DK, Reed CT, Dixit M. Nicotine enhances high-fat diet-induced oxidative stress in the kidney. Nicotine Tob Res (2016.0) 18 1628-1634. DOI: 10.1093/ntr/ntw029 77. Bunbupha S, Apaijit K, Maneesai P, Prasarttong P, Pakdeechote P. Nobiletin ameliorates high-fat diet-induced vascular and renal changes by reducing inflammation with modulating AdipoR1 and TGF-β1 expression in rats. Life Sci (2020.0) 260 118398. DOI: 10.1016/j.lfs.2020.118398 78. Loupal G, Url A, Skalicky M, Viidik A. Physical exercise retards the development of chronic nephropathy in the ageing rat as efficiently as food restriction does. Gerontology (2005.0) 51 83-93. DOI: 10.1159/000082193 79. Keenan KP, Coleman JB, Mccoy CL, Hoe C-M, Soper KA, Laroque P. Chronic nephropathy in ad libitum overfed Sprague-Dawley rats and its early attenuation by increasing degrees of dietary (caloric) restriction to control growth. Toxicol Pathol (2000.0) 28 788-798. DOI: 10.1177/019262330002800604 80. 80.Chen G, Geng Y, Liu C. A low-protein calorie-restricted diet attenuates renal injury and facilitates podocyte autophagy in type 2 diabetic rats. Am Diabetes Assoc. 2018;67(Supplement_1):514. 81. Wang S-Y, Wang W-J, Liu J-Q, Song Y-H, Li P, Sun X-F. Methionine restriction delays senescence and suppresses the senescence-associated secretory phenotype in the kidney through endogenous hydrogen sulfide. Cell Cycle (2019.0) 18 1573-1587. DOI: 10.1080/15384101.2019.1618124 82. Jongbloed F, De Bruin RW, Pennings JL, Payán-Gómez C, Van Den Engel S, Van Oostrom CT. Preoperative fasting protects against renal ischemia-reperfusion injury in aged and overweight mice. PLoS ONE (2014.0) 9 e100853. DOI: 10.1371/journal.pone.0100853 83. Beshyah SA, Badi A, El-Ghul AM, Gabroun A, Dougman KK, Eledrisi MS. The year in “Ramadan Fasting and Health”(2018): a narrative review. Ibnosina J Med Biomed Sci (2019.0) 11 151. DOI: 10.4103/ijmbs.ijmbs_77_19 84. Klooster A. Thiamine, fasting and the kidney (2013.0) 85. Eddy AA, Liu E, McCulloch L. Interstitial fibrosis in hypercholesterolemic rats: role of oxidation, matrix synthesis, and proteolytic cascades. Kidney Int (1998.0) 53 1182-1189. DOI: 10.1046/j.1523-1755.1998.00889.x 86. Szeto HH, Liu S, Soong Y, Alam N, Prusky GT, Seshan SV. Protection of mitochondria prevents high-fat diet–induced glomerulopathy and proximal tubular injury. Kidney Int (2016.0) 90 997-1011. DOI: 10.1016/j.kint.2016.06.013 87. Gunebakan E, Yalcin E, CiklerDulger E, Yigitbasi A, Ates N, Caglayan A. Short-term diet restriction but not alternate day fasting prevents cisplatin-induced nephrotoxicity in mice. Biomedicines (2020.0) 8 23. DOI: 10.3390/biomedicines8020023 88. Abozaid OA, Elshawi OE, Mohammad MA, Moawed FS. Caloric restriction combined with vitamin c and Metformin regulates inflammatory mediators through the JNK/P53/FOXO3 pathway in aged rats journal of biology and today’s world vol. 10. J Biol (2021.0) 10 001-7 89. Lončarević-Vasiljković N, Milanović D, Pešić V, Tešić V, Brkić M, Lazić D. Dietary restriction suppresses apoptotic cell death, promotes Bcl-2 and Bcl-xl mRNA expression and increases the Bcl-2/Bax protein ratio in the rat cortex after cortical injury. Neurochem Int (2016.0) 96 69-76. DOI: 10.1016/j.neuint.2016.02.017 90. Peart JN, See Hoe L, Pepe S, Johnson P, Headrick JP. Opposing effects of age and calorie restriction on molecular determinants of myocardial ischemic tolerance. Rejuvenation Res (2012.0) 15 59-70. DOI: 10.1089/rej.2011.1226 91. Saat TC, van der Pluijm I, Ridwan Y, Van Damme-van den Engel S, van Heijningen PM, Clahsen-vanGroningen MC. Pre-operative fasting provides long term protection against chronic renal damage induced by Ischaemia reperfusion injury in wild type and aneurysm prone fibulin-4 mice. Eur J Vasc Endovasc Surg (2020.0) 60 905-15. DOI: 10.1016/j.ejvs.2020.08.013
--- title: 'Association of smoking, alcohol, and coffee consumption with the risk of ovarian cancer and prognosis: a mendelian randomization study' authors: - Sicong Liu - Songwei Feng - Furong Du - Ke Zhang - Yang Shen journal: BMC Cancer year: 2023 pmcid: PMC10026459 doi: 10.1186/s12885-023-10737-1 license: CC BY 4.0 --- # Association of smoking, alcohol, and coffee consumption with the risk of ovarian cancer and prognosis: a mendelian randomization study ## Abstract ### Objective Currently, the association between smoking, alcohol, and coffee intake and the risk of ovarian cancer (OC) remains conflicting. In this study, we used a two-sample mendelian randomization (MR) method to evaluate the association of smoking, drinking and coffee consumption with the risk of OC and prognosis. ### Methods Five risk factors related to lifestyles (cigarettes per day, smoking initiation, smoking cessation, alcohol consumption and coffee consumption) were chosen from the Genome-Wide Association Study, and 28, 105, 10, 36 and 36 single-nucleotide polymorphisms (SNPs) were obtained as instrumental variables (IVs). Outcome variables were achieved from the Ovarian Cancer Association Consortium. Inverse-variance-weighted method was mainly used to compute odds ratios (OR) and $95\%$ confidence intervals (Cl). ### Results The two-sample MR analysis supported the causal association of genetically predicted smoking initiation (OR: 1.15 per SD, $95\%$CI: 1.02–1.29, $$P \leq 0.027$$) and coffee consumption (OR: 1.40 per $50\%$ increase, $95\%$CI: 1.02–1.93, $$P \leq 0.040$$) with the risk of OC, but not cigarettes per day, smoking cessation, and alcohol consumption. Subgroup analysis based on histological subtypes revealed a positive genetical predictive association between coffee consumption and endometrioid OC (OR: 3.01, $95\%$CI: 1.50–6.04, $$P \leq 0.002$$). Several smoking initiation-related SNPs (rs7585579, rs7929518, rs2378662, rs10001365, rs11078713, rs7929518, and rs62098013), and coffee consumption-related SNPs (rs4410790, and rs1057868) were all associated with overall survival and cancer-specific survival in OC. ### Conclusion Our findings provide the evidence for a favorable causal association of genetically predicted smoking initiation and coffee consumption with OC risk, and coffee consumption is linked to a greater risk of endometrioid OC. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12885-023-10737-1. ## Introduction Ovarian cancer (OC) is the seventh most common malignancy in women worldwide, with the highest mortality rate among gynecological tumors [1]. According to statistics, the 5-year survival rate of patients with OC ranges from 30 to $50\%$ [2]. In 2022, approximately 19,880 people will be diagnosed with OC, and roughly 12,810 will die in the United States [3]. Despite awareness of OC, curative and survival trends have not dramatically changed because there still exists a challenge for early diagnosis. Until now, the extract pathogenesis of OC development remains unknown, which may be mediated by a variety of factors including genetics, environment, and lifestyles. A previous study has shown an unexpected phenomenon that several lifestyle behaviors, such as smoking, drinking alcohol, and coffee consumption, are significantly associated with increased OC risk [4]. Meanwhile, smoking and alcohol consumption have positive associations with a poor prognosis of OC, probably because these behaviors themselves have a detrimental effect on survival [5, 6]. However, some studies have discovered a neutral or poor relationship between smoking, alcohol, and OC [7–9]. Similarly, the relationship between coffee intake and OC is also contradictory [10, 11]. Due to the possible residual confounders and lack of high-quality randomized controlled trials, whether there is a causal relationship between smoking, alcohol, coffee intake and OC needs to be investigated urgently. Mendelian randomization (MR) is a method for determining whether a certain exposure has a causal effect on an outcome [12]. To reduce confounders and reversed causation in observational data, MR design utilizes single-nucleotide polymorphisms (SNPs) as instrumental variants (IVs) for risk factors [13]. Since MR relies on random assignment of alleles during meiosis, it is less affected by confounding factors and can reverse causality. Summary-level data from genome-wide association studies (GWAS) are easier to be obtained and typically large for two-sample MR design, which can enhance the genetic interpretation of IVs on exposure and improve the accuracy and reliability of analysis results [14]. In the present study, we evaluated the association between smoking, drinking and coffee consumption and the risk of OC using a two-sample MR method, aiming at determining whether these lifestyles have a causal rather than pleiotropic impact on OC. Additionally, we also assessed the association of these genetically predicted exposures with the OC prognosis. ## Study design In this MR study, the SNPs were retrieved from a number of published GWAS to determine the causal relationship between exposures and outcomes. Three important assumptions needed to be proven in order to guarantee an efficient MR analysis process: [1] The SNPs were linked with smoking, drinking, and coffee consumption; [2] The SNPs only impacted OC via smoking, drinking, and coffee consumption; [3] The SNPs were entirely unconnected with any possible confounding variables that affected smoking, drinking, and coffee consumption as well as OC. The assumptions of the IVs are shown in Fig. 1.Fig. 1Genetic instrument construction, data sources, and analysis plan on the association between lifestyle factors and ovarian cancer. SNP: single nucleotide polymorphisms ## Genetic Instrument Selection and Data Sources Independent SNPs were chosen from a large meta-analysis including 33 GWAS and a genome-wide meta-analysis involving 28 studies, respectively [15, 16]. They were associated with the number of cigarettes per day ($$n = 337$$,334), smoking initiation (whether an individual had ever smoked regularly, $$n = 1$$,232,091), smoking cessation (current versus former smokers, $$n = 547$$,219), drinks per week ($$n = 66$$,450), and coffee consumption ($$n = 375$$,833). Instruments were chosen at the genome-wide significance threshold ($P \leq 5$ × 10–8) for each exposure trait. The minor allele frequency (MAF) threshold was set at 0.3. We clumped linkage disequilibrium based on European ancestry reference data (1000 Genomes Project, r2 = 0.01, clump window = 10,000 kb) to establish the independence among the SNPs used. Palindromic SNPs were removed from the instrumental variables. We chose independent SNPs for each exposure trait that were significant at the genome-wide level in each GWAS ($$P \leq 5$$ × 10–8). For each selected SNP, we identified the pleiotropy by Phenoscanner (http://www.phenoscanner.medschl.cam.ac.uk/). Pleiotropic SNPs were included in the analysis, and subsequently excluded if sensitivity analysis revealed horizontal pleiotropy. To avoid bias caused by weak instrumental variables, we calculated the F statistic of each SNP. Generally, F > 10 often indicates there is no weak instrument bias. Finally, 189 independent SNPs were chosen as the IVs for MR analysis, including 25 SNPs for cigarettes per day, 88 SNPs for smoking initiation, 7 SNPs for smoking cessation, 37 SNPs for alcohol drinking and 32 SNPs for coffee consumption. The specifics of instrument selection and the corresponding genome-wide association meta-analyses are shown in Table 1 and the raw information of the associations of selected SNPs with lifestyle behaviors are given in Supplementary Table S1.Table 1Data sources of instrumental variablesTraitsNo. of participantsAncestryUnit for each factorNo. of variants includedNumber of SNPs availableaNumber of SNPs usedbPubMed IDCigarettes per day337,334EuropeanSD increase in the number of cigarettes per day11,991,6012,1292530,643,251Smoking initiation1,232,091EuropeanPrevious smoking compared with never smoking11,792,2887,8468830,643,251Smoking cessation547,219EuropeanCurrent smokers Compared with former smokers12,186,231223730,643,251Alcohol consumption66,450EuropeanSD increase in log-transformed alcoholic drinks/week11,976,7065,1963730,643,251Coffee consumption375,833European$50\%$ change7,875,3182,9963231,046,077Ovarian cancer66, 450European----28,346,442“a” corresponds to the number of SNPs available at the genome-wide significance level ($P \leq 5$ × 10–8). “ b” corresponds to the number of SNPs (or linkage disequilibrium proxies) available in ovarian cancer datasetsSNP Single nucleotide polymorphism, SD Standard deviation We used GWAS summary data for the overall OCs and subtypes from Ovarian Cancer Association Consortium (OCAC) [17], an international collaboration with participants of European ancestry recruited from 14 countries, to determine whether genetically predicted smoking, drinking, and coffee consumption was associated with the risk of OC. The studies included 66,450 samples from 7 different genotyping projects, while the OCAC OncoArray data contained 63 genotyping project/case–control combinations. Individuals in the OCAC were disqualified if $5\%$ or more of the genotyping calls were missing. The summary data (25,509 cases; 40,941 controls) were used to analyze the connection of these SNPs with the risk of total OCs. 13,037 high-grade serous, 1,012 low-grade serous, 1,366 clear cell, 2,810 endometrioid, and 1,417 invasive mucinous OC samples were available for the subgroup analyses. Our study only utilized the results of published GWAS. All summary data were downloaded from the IEU OpenGWAS project (https://gwas.mrcieu.ac.uk/). ## Statistical analysis To identify the genetic relationships from each separate GWAS dataset, we employed a two-sample MR design. Inverse variance weighted (IVW) with random effects served as the primary method of statistical analysis. The advantage of the random effects model is that it accounts for the variations in the effect sizes of the chosen SNPs on the exposed phenotypes [18]. The IVW approach, which is used on the presumption that all SNPs are valid IVs and independent of each other [13], is for meta-summarizing the effects of various loci in MR analysis of multiple SNPs. To prevent the influence of unidentified and immeasurable confounders, MR regression, weighted median, weighted mode, and simple mode were utilized as supplemental analyses. The MR-Egger regression can identify and correct the potential pleiotropy [19], but may reduce statistical power. The weighted median method provides a consistent estimate of causality, even though more than half of the instrument weights originate from invalid IVs [20]. Weighted model method weights the contribution of each variant to the clustering by the inverse variance associated with its results [21]. The simple model method groups the SNPs depending on the similarity of MR associations. MR estimates are unbiased when the maximally clustered SNPs are valid instruments [22]. Using MR-Egger regression and weighted medians, sensitivity analysis was performed to clarify possible breaches of the assumptions related to the instrumental variable. The P value for the intercept in MR-Egger regression was employed as an index of pleiotropy [20]. MR-PRESSO detects the presence of horizontal pleiotropy, removes possible outliers and estimates the corrected results, testing for differences between pre-corrected and post-corrected results [23]. Cochran Q values were used to assess the heterogeneity among selected IVs for each exposure. In order to determine whether a particular genetic variation was responsible for the causal connection, the "leave-one-out" method for sensitivity analysis was used. Based on 5 histological subtypes of OCAC, we evaluated the association between the risk factors and the risk of OC subtypes. Additionally, we also evaluated the causative impact of selected SNPs on the prognosis of OC, in accordance with a MR analytic framework basing on SUrvival related cancer Multi-omics database via MEndelian Randomization (SUMMER, http://njmu-edu.cn:3838/SUMMER/) [24]. All statistical analyses were carried out using the package “TwoSampleMR" and “MR-PROESSO” in R software 4.2.0, and all P values were two-sided. ## Association of smoking, alcohol and coffee consumption with the overall risk of OC The F statistics for the five risk factors analyzed in this MR study varied from 41 to 642, indicating that there may not have been any weak instrument biases in our analyses. There were no associations of genetic liability to cigarettes per day, smoking cessation and alcohol consumption with the overall risk of OC. However, two-sample MR showed an adverse effect of smoking initiation and coffee consumption on the overall risk of OC (OR: 1.15, $95\%$CI: 1.02–1.29, $$P \leq 0.027$$; OR: 1.40, $95\%$CI: 1.02–1.93, $$P \leq 0.040$$; Fig. 2, Table 2).Fig. 2Association of smoking, alcohol and coffee consumption with the overall risk of ovarian cancer in the Mendelian randomization analysis. OR, odds ratio; CI, confidence intervalTable 2Mendelian randomization estimates between phenotypes and ovarian cancerExposuresOutcomesMethodOR ($95\%$CI)P valueCigarettes per dayOverall ovarian cancerInverse-variance weighted0.99 (0.90–1.08)0.755MR Egger1.19 (1.01–1.40)0.051Weighted median1.09 (0.95–1.24)0.211Weighted mode1.11(0.97–1.27)0.148Simple mode1.07 (0.80–1.43)0.671Smoking initiationOverall ovarian cancerInverse-variance weighted1.15 (1.02–1.29)0.027MR Egger1.12 (0.61–2.05)0.71Weighted median1.14 (0.96–1.35)0.133Weighted mode1.17 (0.84–1.64)0.349Simple mode1.20 (0.78–1.84)0.41Smoking cessationOverall ovarian cancerInverse-variance weighted0.95 (0.76–1.19)0.665MR Egger1.00 (0.50–2.00)0.991Weighted median0.97 (0.72–1.29)0.814Weighted mode0.97 (0.68–1.38)0.865Simple mode0.92 (0.60–1.40)0.695Alcohol consumptionOverall ovarian cancerInverse-variance weighted0.79 (0.55–1.11)0.175MR Egger0.66 (0.38–1.13)0.136Weighted median0.74 (0.52–1.04)0.081Weighted mode0.74 (0.53–1.05)0.103Simple mode0.88 (0.39–1.96)0.747Coffee consumptionOverall ovarian cancerInverse-variance weighted1.40 (1.02–1.93)0.04MR Egger0.77 (0.40–1.50)0.449Weighted median1.44 (0.87–2.41)0.158Weighted mode1.47 (0.83–2.61)0.2Simple mode1.77 (0.60–5.19)0.308 The complementary estimates further demonstrated this consistency in detecting smoking initiation (Table 3). MR-PRESSO did not show a horizontal pleiotropy. There may be no pleiotropy since the P value for the MR-Egger intercept was over 0.05 (Table 3). No heterogeneity was detected and the leave-one-out sensitivity test indicated that the results of the MR analysis were robust. Table 3Pleiotropy and heterogeneity of five phenotypes in ovarian cancerExposuresPleiotropy (P value)Heterogeneity (P value)Cigarettes per day0.0120.505Smoking initiation0.9460.277Smoking cessation0.8750.793Alcohol consumption0.3971.26E-05Coffee consumption0.0530.809 ## Association of smoking initiation and coffee consumption with OC subtypes Based on histological subtypes, a subgroup analysis of OC was performed in terms of smoking initiation and coffee consumption. The results showed that coffee consumption was strongly associated with a high risk of endometrioid OC (OR: 3.01, $95\%$CI: 1.50–6.04, $$P \leq 0.002$$), but not other OC subtypes (Table 4). Additionally, no association was observed between genetically predicted smoke initiation and all OC subtypes. Table 4Causal estimates for the association between smoking initiation and coffee consumption and ovarian cancer subtypesExposuresHistological subtypesOR ($95\%$CI)P valueSmoking initiationHigh-grade serous1.15 (1.00–1.33)0.057Smoking initiationLow grade serous1.43 (0.93–2.20)0.104Smoking initiationClear cell1.24 (0.87–1.76)0.231Smoking initiationEndometrioid1.13 (0.86–1.48)0.369Smoking initiationInvasive mucinous0.92 (0.65–1.32)0.667Coffee consumptionHigh-grade serous1.23 (0.84–1.81)0.28Coffee consumptionLow grade serous1.00 (0.28–3.53)0.998Coffee consumptionClear cell2.02 (0.73–5.61)0.178Coffee consumptionEndometrioid3.01 (1.50–6.04)0.002Coffee consumptionInvasive mucinous1.15 (0.44–3.05)0.77 ## Effect of smoking initiation and coffee consumption on the prognosis of OC We evaluated the effect of the chosen SNPs on the prognosis of OC, which were discovered to be linked with a greater probability of developing OC using MR analysis. Shorter overall survival (OS) for OC was positively correlated with SNPs rs7585579 (HR: 1.25, $$P \leq 0.031$$), rs7929518 (HR: 1.45, $$P \leq 0.004$$), and rs2378662 (HR: 1.26, $$P \leq 0.030$$) related to smoking initiation, whereas SNPs rs10001365 (HR: 0.74, $$P \leq 0.007$$) and rs11078713 (HR: 0.80, $$P \leq 0.033$$) related to smoking initiation had the opposite effect. Additionally, SNPs rs7585579 (HR: 1.30, $$P \leq 0.018$$), rs7929518 (HR: 1.42, $$P \leq 0.013$$), rs2378662 (HR: 1.26, $$P \leq 0.037$$) and rs62098013 (HR: 1.24, $$P \leq 0.045$$) associated with smoking initiation were also identified to be linked with shorter cancer-specific survival (CSS), whereas rs10001365 (HR: 0.75, $$P \leq 0.016$$) was significantly associated with longer CSS in OC. Notably, SNPs rs4410790 (HR: 1.31, $$P \leq 0.007$$) and rs1057868 (HR: 1.28, $$P \leq 0.022$$) related to coffee consumption were both correlated with poor OS in OC, and SNP rs4410790 (HR: 1.30, $$P \leq 0.016$$) was in the association with poor CSS in OC (Table 5; Figs. 3 and 4).Table 5Effect of smoking initiation and coffee consumption on overall survival and cancer-specific survival in all ovarian cancersSNPsExposuresHR_OSSE-OSP_value_OSHR_CSSSE-CSSP_value_CSSrs7585579Smoking initiation1.250.10.0311.30.110.018rs10001365Smoking initiation0.740.110.0070.750.120.016rs7929518Smoking initiation1.450.130.0041.420.140.013rs2378662Smoking initiation1.260.10.031.260.110.037rs11078713Smoking initiation0.80.10.0330.850.110.126rs62098013Smoking initiation1.20.10.081.240.110.045rs4410790Coffee consumption1.310.10.0071.30.110.016rs1057868Coffee consumption1.280.110.0221.140.120.261Abbreviations: SNP Single nucleotide polymorphism, HR Hazard ratio, OS Overall survival, CSS Cancer-specific survivalFig. 3Kaplan–Meier plots of the effect of smoking initiation on overall survival and cancer-specific survival in ovarian cancer. Association between A rs7585579, B rs10001365, C rs7929518, D rs2378662, E rs11078713 and overall survival in ovarian cancer. Association between F rs7585579, G rs2378662, H rs10001365, I rs62098013, J rs7929518 and cancer-specific survival in ovarian cancerFig. 4Kaplan–Meier plots of the effect of coffee consumption on overall survival and cancer-specific survival in ovarian cancer. Association between A rs4410790, B rs1057868 and overall survival in ovarian cancer. Association between C rs4410790 and cancer-specific survival in ovarian cancer ## Discussion The prevention of OC remains a big challenge at present. The results of this two-sample MR analysis suggested that smoking initiation and coffee consumption were associated with an increased risk of OC. The MR analysis failed to provide the evidence for a causal effect of alcohol drinking, cigarettes per day and smoking cessation. The subgroup analysis based on OC subtypes further revealed that coffee consumption was associated with a higher risk of developing endometrioid OC. Notably, several SNPs related to smoking initiation and coffee consumption were found to relate to the OS and CSS of OC. Smoking is a predictor of cancer incidence and related to worse long-term outcomes [25, 26]. Our results offered a modest evidence for a positive causal relationship between smoking initiation and OC risk according to the IVW analysis, and there was no relationship between cigarettes per day or smoking cessation and OC risk. According to a recent research involving 1,279 participants, the risk of OC-specific mortality has risen by $19\%$ and $21\%$ in patients with pre-diagnosis and post-diagnosis smoking compared with never smokers, respectively [27]. By comparison to women who had never smoked, the overall prevalence of OC was only marginally higher among current smokers. Smoking initiation may be a causative risk for the total OC, which was supported by this MR study based on the summary data from 66,450 women. There is also evidence suggesting that women who had never smoked previously were more likely to develop mucinous OC than those who did [9]. The current research, however, could not corroborate this finding linking smoking with invasive mucinous OC. The precise mechanisms of how smoking contributes to the development of OC are not fully understood. There are various carcinogens in cigarette smoke, including N-nitrosamines, aromatic amines, 1,3-butadiene, and benzene [28, 29]. Nicotine in tobacco is a cancer promoter, and chronic smoking may promote cancer cell proliferation, epithelial-mesenchymal transition and angiogenesis [30–33], as well as cause OC to develop a more aggressive phenotype for promoting metastatic spread [34]. Smoking may enhance the pro-inflammatory cytokines and chemokines in the tumor environment, consequently increasing the likelihood of treatment resistance [35]. Smokers are more prone to engage in unhealthy lifestyles including obesity and alcohol use, which may have a negative impact on the prognosis of OC, even though alcohol consumption is not linked with OC in this MR analysis. Currently, smoking cessation has been described as a key method in preventing a variety of cancers [36]. Our finding shows that there is no relationship between alcohol consumption and OC, which is consistent with previously published studies. A meta-analysis that included 16,554 ovarian cancer patients found that alcohol consumption was not associated with OC risk [37]. In addition, Cook et al. [ 38] suggested that alcohol consumption consistent with guidelines did not increase the risk of epithelial OC, but higher wine consumption was associated with a lower risk of ovarian cancer. The biological mechanisms underlying the relationship between alcohol and OC are currently unclear. Alcohol consumption may lead to increased cumulative estrogen exposure, leading to the development of oc through epithelial cell genotoxicity and mitosis [39, 40]. Meanwhile, acetaldehyde, an oxidized metabolite of alcohol, can be carcinogenic [41]. On the contrary, polyphenols, flavonoids, and resveratrol found in alcohol such wine and red wine had anti-inflammatory and antioxidant effects, which may decrease OC risk [42–44]. There is a correlation between coffee consumption and the risk of developing OC, although there is no statistically significant relationship, according to a systematic review of 15 OC studies including 5,021 individuals [45]. Previous studies have demonstrated that caffeine intake among premenopausal women is associated with increased OC risk, whereas there is no or little association among postmenopausal women [36, 46]. In our two-sample MR analysis, coffee drinking was positively related to the risk of OC. However, several studies showed that drinking coffee did not increase or lower the incidence of OC [47–49]. Additionally, we also found the association between coffee drinking as a risk factor and the risk of endometrioid OC. Ong et al. discovered no evidence indicative of a causal relationship between genetically predicted coffee or caffeine concentrations and epithelial OC risk [50], which was conflicting with our two-sample MR research. There were several reasons that may explain this phenomenon. First, the database we chose was the latest GWAS database with a larger sample size, thus a large variety of IVs may be chosen, consequently increasing the potential of the association between genetically interpreted coffee consumption and OC risk. In our two-sample MR analysis, a total of 28 significant SNPs were selected for coffee consumption, significantly more than those in the prior study [51]. Second, the Ong’s study only examined the connection between coffee intake and epithelial OC, but in our study, the potential causal relationship between coffee consumption and OC subtypes was investigated. Interestingly, we found a strong association between coffee consumption and endometrioid OC risk. Acrylamide, which is produced during roasting coffee beans at high temperatures, may be the mechanism causing this outcome [52]. High acrylamide intake may be scientifically conceivable as a potential risk factor for OC [53]. In addition, caffeine is able to inhibit aromatase activity and increase the secretion of sex hormone-binding globulin altering the hormonal milieu [54, 55]. The hormonal alterations synergize with coelomic metaplasia, proliferation of progenitor stem cells, or retrograde menstruation of endometrial cells, leading to implantation and proliferation of ectopic endometrial cells and increasing the risk of endometrioid ovarian cancer [56]. It is well known that obesity is a key risk factor for OC. In our study, robust evidences points to favorable causal association of genetically predicted smoking initiation and coffee consumption with OC risk. Caffeine is the main ingredient in coffee [50]. It have been shown that in obese people the apparent distribution of caffeine increases by $60\%$, but does not affect caffeine clearance [57]. Biologically active compounds in coffee, such as chlorogenic acid, caffeine, have shown to be associated with anti-obesity benefits [58]. Nicotine in tobacco can increase energy consumption and inhibit appetite in a short period of time, but people who smoke more have a higher BMI than light smokers, probably because heavy smokers are accompanied by unhealthy behaviors such as poor diet, alcohol abuse and low physical activity [59, 60]. Also, smoking can promote visceral fat accumulation and insulin resistance and hyperinsulinemia, increasing the risk of obesity [59]. The IVW method's statistical power is much greater than that of other MR methods, particularly MR-Egger [61]. In our study, IVW was used as a main approach to screen the results in MR. To ensure the robust findings, we also performed a sensitivity analysis. Taken together, our findings supported the hypothesis that smoking and coffee consumption could increase the risk of OC, thus the strategies to reduce the exposure of these two factors was worthy of attention to decrease the risk of OC. Regular smoking and coffee cessation campaigns should be conducted among the female population to lower the incidence of OC. The present study has several strengths and limitations. First, the majority of studies on smoking, drinking, and coffee utilized self-reported consumption, which was easy to cause the bias. This MR study, however, examined summary statistics of several behaviors from a large dataset. In our two-sample MR investigation, genetic variation may examine the possible causative influence of exposures on OC without being biased or confounded by confounding or reserve causation [61]. Second, we explored the association of a few significant SNPs with OS and CSS in OC and made Kaplan–Meier plot diagrams to show. Additionally, all of the studies included communities with a predominance of European ancestry, which reduced the possibility of population stratification-related bias, but may not be generalizable to other groups. Pleiotropy was ruled out, but there might still be alternative mechanisms through which SNPs and OC were related. Importantly, we cannot rule out a genetic link between OC, smoking initiation, and coffee consumption. Notably, the menopausal status (premenopausal or postmenopausal) of OC patients was not stratified in this research, so it was uncertain to determine whether the effect of smoking and coffee intake on the OC risk was under the influence of menopausal status. Another limitation was that a two-sample MR design was unable to evaluate the reverse causality across these exposures on OC. ## Conclusions This two-sample MR study provided the evidence for favorable causal association between genetically predicted smoking initiation and coffee consumption and OC risk. Meanwhile, coffee consumption was linked to a greater risk of endometrioid OC according to histological subgroup analysis. In the future, clinicians can collect peripheral blood or tumor tissue from OC patients for SNP testing. They can use several methods for SNP detection, such as sequencing, TaqMan probes, gene microarrays, and mass spectrometry. Based on our results, the risk of OC and survival outcomes can be better identified. ## Supplementary Information Additional file 1: Table S1. The causal effect estimates of the associations between genetic instrumental variables for lifestyle behaviors and risk of ovarian cancers. ## References 1. Lheureux S, Gourley C, Vergote I. M Oza A. Epithelial ovarian cancer The Lancet (2019.0) 393 1240-1253 2. Allemani C, Matsuda T, Di Carlo V, Harewood R, Matz M, Nikšić M. Global surveillance of trends in cancer survival 2000–14 (CONCORD-3): analysis of individual records for 37 513 025 patients diagnosed with one of 18 cancers from 322 population-based registries in 71 countries. Lancet (2018.0) 391 1023-1075. DOI: 10.1016/S0140-6736(17)33326-3 3. Siegel Rebecca L, Miller Kimberly D, Fuchs Hannah E. Cancer statistics 2022. CA Cancer J Clin (2022.0) 72 7-33. DOI: 10.3322/caac.21708 4. Tanha K, Mottaghi A, Nojomi M. Investigation on factors associated with ovarian cancer: an umbrella review of systematic review and meta-analyses. J Ovarian Res (2021.0) 14 153. DOI: 10.1186/s13048-021-00911-z 5. Wang T, Townsend MK, Vinci C. Early life exposure to tobacco smoke and ovarian cancer risk in adulthood. Int J Epidemiol. (2021.0) 50 965-974. DOI: 10.1093/ije/dyab018 6. Chang ET, Canchola AJ, Lee VS, Clarke CA, Purdie DM, Reynolds P. Wine and other alcohol consumption and risk of ovarian cancer in the California Teachers Study cohort. Cancer Causes Control (2007.0) 18 91-103. DOI: 10.1007/s10552-006-0083-x 7. Huang Y-H, Li J, Luan H, Huang S-S, Li Y. Association between alcohol consumption and the risk of ovarian cancer: a meta-analysis of prospective observational studies. BMC Public Health (2015.0) 15 223. DOI: 10.1186/s12889-015-1355-8 8. Rota M, Pasquali E, Scotti L, Pelucchi C, Tramacere I, Islami F. Alcohol drinking and epithelial ovarian cancer risk. A systematic review and meta-analysis. Gynecol Oncol (2012.0) 125 758-63. DOI: 10.1016/j.ygyno.2012.03.031 9. Beral V, Gaitskell K, Hermon C, Moser K, Reeves G. Ovarian cancer and smoking: individual participant meta-analysis including 28,114 women with ovarian cancer from 51 epidemiological studies. Lancet Oncol (2012.0) 13 946-956. DOI: 10.1016/S1470-2045(12)70322-4 10. Salari-Moghaddam A, Milajerdi A, Surkan PJ, Larijani B, Esmaillzadeh A. Caffeine, type of coffee, and risk of ovarian cancer: a dose-response meta-analysis of prospective studies. J Clin Endocrinol Metab (2019.0) 104 5349-5359. DOI: 10.1210/jc.2019-00637 11. Gunter MJ, Murphy N, Cross AJ, Dossus L, Dartois L. Coffee drinking and mortality in 10 European countries: a multinational cohort study. Ann Intern Med. (2017.0) 167 236-247. DOI: 10.7326/M16-2945 12. Smith G, Ebrahim S. ‘Mendelian randomization’: can genetic epidemiology contribute to understanding environmental determinants of disease?. Int J Epidemiol (2003.0) 32 1-22. DOI: 10.1093/ije/dyg070 13. Burgess S, Butterworth A, Thompson S. Mendelian randomization analysis with multiple genetic variants using summarized data. Genet Epidemiol (2013.0) 37 658-665. DOI: 10.1002/gepi.21758 14. Hwang LD, Lawlor DA, Freathy RM, Evans DM, Warrington NM. Using a two-sample Mendelian randomization design to investigate a possible causal effect of maternal lipid concentrations on offspring birth weight. Int J Epidemiol (2019.0) 48 1457-1467. DOI: 10.1093/ije/dyz160 15. Liu M, Jiang Y, Wedow R, Li Y, Brazel DM, Chen F. Association studies of up to 1. 2 million individuals yield new insights into the genetic etiology of tobacco and alcohol use. Nat Genet (2019.0) 51 237-44. DOI: 10.1038/s41588-018-0307-5 16. Zhong VW, Kuang A, Danning RD, Kraft P, Dam RM, Chasman D. A genome-wide association study of bitter and sweet beverage consumption. Hum Mol Genet (2019.0) 28 2449-2457. DOI: 10.1093/hmg/ddz061 17. Phelan CM, Kuchenbaecker KB, Tyrer JP, Kar SP, Lawrenson K, Winham SJ. Identification of 12 new susceptibility loci for different histotypes of epithelial ovarian cancer. Nat Genet (2017.0) 49 680-691. DOI: 10.1038/ng.3826 18. Nikolakopoulou A, Mavridis D, Salanti G. How to interpret meta-analysis models: fixed effect and random effects meta-analyses. Evid Based Ment Health (2014.0) 17 64. DOI: 10.1136/eb-2014-101794 19. Bowden J, Davey Smith G, Burgess S. Mendelian randomization with invalid instruments: effect estimation and bias detection through Egger regression. Int J Epidemiol (2015.0) 44 512-525. DOI: 10.1093/ije/dyv080 20. Bowden J, Davey Smith G, Haycock PC, Burgess S. Consistent estimation in mendelian randomization with some invalid instruments using a weighted median estimator. Genet Epidemiol (2016.0) 40 304-314. DOI: 10.1002/gepi.21965 21. Hartwig FP, Davey Smith G, Bowden J. Robust inference in summary data Mendelian randomization via the zero modal pleiotropy assumption. Int J Epidemiol (2017.0) 46 1985-1998. DOI: 10.1093/ije/dyx102 22. Hemani G, Zheng J, Elsworth B, Wade KH, Haberland V, Baird D. The MR-Base platform supports systematic causal inference across the human phenome. Elife (2018.0) 7 1-29. DOI: 10.7554/eLife.34408 23. Verbanck M, Chen CY, Neale B, Do R. Detection of widespread horizontal pleiotropy in causal relationships inferred from Mendelian randomization between complex traits and diseases. Nat Genet (2018.0) 50 693-698. DOI: 10.1038/s41588-018-0099-7 24. 24.Xin J, Gu D, Chen S, Ben S, Li H, Zhang Z, et al. SUMMER: a Mendelian randomization interactive server to systematically evaluate the causal effects of risk factors and circulating biomarkers on pan-cancer survival. Nucleic Acids Res. 2023;51(D1):D1160-D1167. 25. Sasco AJ, Secretan MB, Straif K. Tobacco smoking and cancer: a brief review of recent epidemiological evidence. Lung Cancer (2004.0) 45 S3-S9. DOI: 10.1016/j.lungcan.2004.07.998 26. Warren GW, Kasza KA, Reid ME, Cummings KM, Marshall JR. Smoking at diagnosis and survival in cancer patients. Int J Cancer (2013.0) 132 401-410. DOI: 10.1002/ijc.27617 27. Wang T, Townsend MK, Simmons V, Terry KL, Matulonis UA, Tworoger SS. Prediagnosis and postdiagnosis smoking and survival following diagnosis with ovarian cancer. Int J Cancer (2019.0) 147 736-746. DOI: 10.1002/ijc.32773 28. Grando SA. Connections of nicotine to cancer. Nat Cancer (2014.0) 14 419-429. DOI: 10.1038/nrc3725 29. Hecht SS. Tobacco carcinogens, their biomarkers and tobacco-induced cancer. Nat Cancer (2003.0) 3 733-744. DOI: 10.1038/nrc1190 30. Schaal C, Chellappan SP. Nicotine-mediated cell proliferation and tumor progression in smoking-related cancers. Molecular cancer research : MCR (2014.0) 12 14-23. DOI: 10.1158/1541-7786.MCR-13-0541 31. Schuller HM. Regulatory role of the alpha7nAChR in cancer. Curr Drug Targets (2012.0) 13 680-687. DOI: 10.2174/138945012800398883 32. Singh S, Pillai S, Chellappan S. Nicotinic Acetylcholine Receptor Signaling in Tumor Growth and Metastasis. Journal of Oncology (2011.0) 2011 456743. DOI: 10.1155/2011/456743 33. Jeon SY, Go RE, Heo JR, Kim CW, Hwang KA, Choi KC. Effects of cigarette smoke extracts on the progression and metastasis of human ovarian cancer cells via regulating epithelial-mesenchymal transition. Reprod Toxicol (2016.0) 65 1-10. DOI: 10.1016/j.reprotox.2016.06.012 34. Fortner RT, Poole EM, Wentzensen NA. Ovarian cancer risk factors by tumor aggressiveness: an analysis from the ovarian cancer cohort consortium. Int J Cancer (2018.0) 145 58-69. DOI: 10.1002/ijc.32075 35. Clendenen TV, Lundin E, Zeleniuch-Jacquotte A, Koenig KL, Berrino F, Lukanova A. Circulating inflammation markers and risk of epithelial ovarian cancer. Cancer Epidemiol Biomarkers Prev (2011.0) 20 799-810. DOI: 10.1158/1055-9965.EPI-10-1180 36. Kuper H, Titus-Ernstoff L, Harlow BL, Cramer DW. Population based study of coffee, alcohol and tobacco use and risk of ovarian cancer. Int J Cancer (2000.0) 88 313-318. DOI: 10.1002/1097-0215(20001015)88:2<313::AID-IJC26>3.0.CO;2-5 37. Rota M, Pasquali E, Scotti L. Alcohol drinking and epithelial ovarian cancer risk. A systematic review and meta-analysis. Gynecol Oncol. (2012.0) 125 758-763. DOI: 10.1016/j.ygyno.2012.03.031 38. Cook LS, Leung AC, Swenerton K. Adult lifetime alcohol consumption and invasive epithelial ovarian cancer risk in a population-based case-control study. Gynecol Oncol (2016.0) 140 277-284. DOI: 10.1016/j.ygyno.2015.12.005 39. Ho SM. Estrogen, progesterone and epithelial ovarian cancer. Reprod Biol Endocrinol. (2003.0) 1 73. DOI: 10.1186/1477-7827-1-73 40. Singletary KW, Gapstur SM. Alcohol and breast cancer: review of epidemiologic and experimental evidence and potential mechanisms. JAMA (2001.0) 286 2143-2151. DOI: 10.1001/jama.286.17.2143 41. Secretan B, Straif K, Baan R, Grosse Y, El Ghissassi F, Bouvard V. A review of human carcinogens—Part E: tobacco, areca nut, alcohol, coal smoke, and salted fish. Lancet Oncol (2009.0) 10 1033-1034. DOI: 10.1016/S1470-2045(09)70326-2 42. Arranz S, Chiva-Blanch G, Valderas-Martínez P, Medina-Remón A, Lamuela-Raventós RM, Estruch R. Wine, beer, alcohol and polyphenols on cardiovascular disease and cancer. Nutrients (2012.0) 4 759-781. DOI: 10.3390/nu4070759 43. Udenigwe CC, Ramprasath VR, Aluko RE, Jones PJ. Potential of resveratrol in anticancer and anti-inflammatory therapy. Nutr Rev (2008.0) 66 445-454. DOI: 10.1111/j.1753-4887.2008.00076.x 44. Fernandez-Panchon MS, Villano D, Troncoso AM, Garcia-Parrilla MC. Antioxidant activity of phenolic compounds: from in vitro results to in vivo evidence. Crit Rev Food Sci Nutr (2008.0) 48 649-671. DOI: 10.1080/10408390701761845 45. Zhao LG, Li ZY, Feng GS, Ji XW, Tan YT, Li HL. Coffee drinking and cancer risk: an umbrella review of meta-analyses of observational studies. BMC Cancer (2020.0) 20 1-12 46. Kotsopoulos J, Vitonis AF, Terry KL, DeVivo I, Cramer DW, Hankinson SE, Tworoger SS. Coffee intake, variants in genes involved in caffeine metabolism, and the risk of epithelial ovarian cancer. Cancer Causes Control. (2009.0) 20 335-344. DOI: 10.1007/s10552-008-9247-1 47. Braem MGM, Onland-Moret NC, Schouten LJ, Tjønneland A, Hansen L. Coffee and tea consumption and the risk of ovarian cancer: a prospective cohort study and updated meta-analysis. Am J Clin Nutr. (2012.0) 95 1172-81. DOI: 10.3945/ajcn.111.026393 48. Hashibe M, Galeone C, Buys SS, Gren L. Coffee, tea, caffeine intake, and the risk of cancer in the PLCO cohort. Br J Cancer (2015.0) 113 809-816. DOI: 10.1038/bjc.2015.276 49. Gosvig CF, Kjaer SK, Blaakær J, Høgdall E, Høgdall C. Coffee, tea, and caffeine consumption and risk of epithelial ovarian cancer and borderline ovarian tumors: results from a Danish case-control study. Acta Oncol (Stockholm, Sweden) (2015.0) 54 1144-51. DOI: 10.3109/0284186X.2014.1001035 50. Heckman MA, Weil J, Gonzalez de Mejia E. Caffeine (1, 3, 7-trimethylxanthine) in foods: a comprehensive review on consumption, functionality, safety, and regulatory matters. J Food Sci (2010.0) 75 R77-87. DOI: 10.1111/j.1750-3841.2010.01561.x 51. Ong JS, Hwang LD, Cuellar Partida G, Martin NG, Chenevix Trench G, Quinn MCJ. Assessment of moderate coffee consumption and risk of epithelial ovarian cancer: a mendelian randomization study. Int J Epidemiol (2018.0) 47 450-459. DOI: 10.1093/ije/dyx236 52. Mojska H, Gielecińska I. Studies of acrylamide level in coffee and coffee substitutes: influence of row material and manufacturing conditions. Rocz Panstw Zakl Hig (2013.0) 64 173-181. PMID: 24325083 53. Pelucchi C, Bosetti C, Galeone C, La Vecchia C. Dietary acrylamide and cancer risk: an updated meta-analysis. Int J Cancer (2014.0) 136 2912-2922. DOI: 10.1002/ijc.29339 54. Kotsopoulos J, Eliassen AH, Missmer SA, Hankinson SE, Tworoger SS. Relationship between caffeine intake and plasma sex hormone concentrations in premenopausal and postmenopausal women. Cancer (2009.0) 115 2765-2774. DOI: 10.1002/cncr.24328 55. Nagata C, Kabuto M, Shimizu H. Association of coffee, green tea, and caffeine intakes with serum concentrations of estradiol and sex hormone-binding globulin in premenopausal Japanese women. Nutr Cancer (1998.0) 30 21-24. DOI: 10.1080/01635589809514635 56. Sadeu J, Hughes CL, Agarwal S, Foster WG. Alcohol, drugs, caffeine, tobacco, and environmental contaminant exposure: Reproductive health consequences and clinical implications. Crit Rev Toxicol (2010.0) 40 633-652. DOI: 10.3109/10408444.2010.493552 57. Kamimori GH, Somani SM, Knowlton RG, Perkins RM. The effects of obesity and exercise on the pharmacokinetics of caffeine in lean and obese volunteers. Eur J Clin Pharmacol (1987.0) 31 595-600. DOI: 10.1007/BF00606637 58. Higdon JV, Frei B. Coffee and health: a review of recent human research. Crit Rev Food Sci Nutr (2006.0) 46 101-123. DOI: 10.1080/10408390500400009 59. Song M, Giovannucci E. Estimating the Influence of Obesity on Cancer Risk: Stratification by Smoking Is Critical. J Clin Oncol. (2016.0) 34 3237-9. DOI: 10.1200/JCO.2016.67.6916 60. Lin Z, Deng Y, Pan W. Combining the strengths of inverse-variance weighting and Egger regression in Mendelian randomization using a mixture of regressions model. PLoS Genet (2021.0) 17 e1009922. DOI: 10.1371/journal.pgen.1009922 61. Lawlor DA, Harbord RM, Sterne JA, Timpson N, Davey SG. Mendelian randomization: Using genes as instruments for making causal inferences in epidemiology. Stat Med (2008.0) 27 1133-1163. DOI: 10.1002/sim.3034
--- title: Whole-body vibration ameliorates glial pathological changes in the hippocampus of hAPP transgenic mice, but does not affect plaque load authors: - Tamas Oroszi - Eva Geerts - Reuben Rajadhyaksha - Csaba Nyakas - Marieke J. G. van Heuvelen - Eddy A. van der Zee journal: 'Behavioral and Brain Functions : BBF' year: 2023 pmcid: PMC10026461 doi: 10.1186/s12993-023-00208-9 license: CC BY 4.0 --- # Whole-body vibration ameliorates glial pathological changes in the hippocampus of hAPP transgenic mice, but does not affect plaque load ## Abstract ### Background Alzheimer’s disease (AD) is the core cause of dementia in elderly populations. One of the main hallmarks of AD is extracellular amyloid beta (Aβ) accumulation (APP-pathology) associated with glial-mediated neuroinflammation. Whole-Body Vibration (WBV) is a passive form of exercise, but its effects on AD pathology are still unknown. ### Methods Five months old male J20 mice ($$n = 26$$) and their wild type (WT) littermates ($$n = 24$$) were used to investigate the effect of WBV on amyloid pathology and the healthy brain. Both J20 and WT mice underwent WBV on a vibration platform or pseudo vibration treatment. The vibration intervention consisted of 2 WBV sessions of 10 min per day, five days per week for five consecutive weeks. After five weeks of WBV, the balance beam test was used to assess motor performance. Brain tissue was collected to quantify Aβ deposition and immunomarkers of astrocytes and microglia. ### Results J20 mice have a limited number of plaques at this relatively young age. Amyloid plaque load was not affected by WBV. Microglia activation based on IBA1-immunostaining was significantly increased in the J20 animals compared to the WT littermates, whereas CD68 expression was not significantly altered. WBV treatment was effective to ameliorate microglia activation based on morphology in both J20 and WT animals in the Dentate Gyrus, but not so in the other subregions. Furthermore, GFAP expression based on coverage was reduced in J20 pseudo-treated mice compared to the WT littermates and it was significantly reserved in the J20 WBV vs. pseudo-treated animals. Further, only for the WT animals a tendency of improved motor performance was observed in the WBV group compared to the pseudo vibration group. ### Conclusion In accordance with the literature, we detected an early plaque load, reduced GFAP expression and increased microglia activity in J20 mice at the age of ~ 6 months. Our findings indicate that WBV has beneficial effects on the early progression of brain pathology. WBV restored, above all, the morphology of GFAP positive astrocytes to the WT level that could be considered the non-pathological and hence “healthy” level. Next experiments need to be performed to determine whether WBV is also affective in J20 mice of older age or other AD mouse models. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12993-023-00208-9. ## Introduction Alzheimer’s disease (AD) has been identified as the most prevalent type of dementia [1]. The major pathological hallmarks of AD are extracellular amorphous amyloid plaques aggregated of misfolded amyloid beta peptides and intracellular neurofibrillary tangles consisting of hyperphosphorylated TAU proteins [1, 2]. However, there is growing evidence that the involvement of glial-mediated neuroinflammation may also play an important role in the pathogenesis of AD [3–8]. Given the high complexity and serious consequences of AD, there has been a broad range of scientific inquiries for pharmacological, as well as non-pharmacological treatment strategies [9, 10]. Ample meta-analytical reviews have summarized the effects of regular physical activity (PA) on the symptoms and progression of AD [11–13]. Although it is important to note that these studies often demonstrated inconsistent results and do not show unanimity, a growing body of evidence support the potential efficiency and value of regular and moderated PA to prevent and/or slow the progression of AD. Further, regular PA promotes neurogenesis, synaptic plasticity and angiogenesis via increases in the level of neurotransmitters and neurotrophic factors [12, 13]. The findings of these studies indicate that amyloid plaque formation and AD-related neuroinflammation could be reduced by these factors. Active exercise may therefore be a viable intervention to trigger anti-inflammatory effects to ameliorate amyloid plaque formation. Pharmacological and non-pharmacological AD treatments recently focus more on the function of astrocytes, although their role entails a complex balance between neurotoxic and neuroprotective effects depending on the disease stage and microenvironmental factors (Rodríguez-Giraldo et al. and references therein [14]). Moreover, there is a growing need for alternative exercise strategies to support populations who are unable and/or unmotivated to perform sufficient PA due to their limited cognitive and/or motor capabilities. Whole body vibration (WBV), a form of passive exercise using mechanical vibration platforms, may provide an alternative for PA. Benefits of WBV in older populations are reflected by improved general fitness, mobility and balance [15, 16]. In rodents, WBV is able to increase neuromuscular dynamics and muscle strength [17], to improve adipose tissue dysfunction and glucose metabolism [18], and to promote muscle healing [19]. Recent animal studies have shown that vibration stimulates hippocampal functioning reflected in improved modulation of synaptic and/or neural plasticity and spatial memory, and in alleviated pathological changes of glial cells [20–25]. To the best of our knowledge, and confirmed by a recent review [26], the therapeutic effects of WBV have never been investigated in the context of AD pathophysiology. However, similar alternative therapies based on transcranial ultrasound and auditory stimulation proved the therapeutic efficacy of mechanical waves to mitigate AD pathophysiology. Given the potential significance of WBV for neural protection, we hypothesized that WBV may beneficially modulate the glial activation (known to be critically involved in AD pathology; see Rodríguez-Giraldo et al., 2022 for review [14]) and amyloid beta plaque formation in the hippocampus of transgenic human APP-J20 mice, a well-known model of AD [27]. To achieve this aim, we evaluated a five-week long WBV protocol in five months-old transgenic human APP-J20 mice with the main focus on AD molecular pathophysiology. Further, since WBV has been widely acknowledged to improve muscle parameters, minor attention was also paid to the evaluation of motor performance. The age of five months was chosen to start the experiment because at this age J20 mice show an early, but detectable age-related deficit in cognitive and behavioral performance [28–31], as well as in the progression of neuroinflammation and plaque formation [32]. ## Animals Twenty-six transgenic hAPP-J20 male mice (PDGFB-APPSwInd; C57Bl6/J background) and 24 male wild type (WT) littermates (C57Bl6/J) serving as healthy controls were used in this experiment. The age of these animals was 5 months at the start of the experiment. Both J20 and WT mice were randomly allocated to a WBV group [WBV—J20 ($$n = 13$$) and WBV – WT ($$n = 12$$)] or a control group [pseudo WBV—J20 ($$n = 13$$) and pseudo WT ($$n = 12$$)]. Pseudo control mice were subjected to the same environmental stimuli, including placement on the vibration plate and sound of the vibration plate, but were not exposed to vibration. Animals were individually housed during the intervention period. Individual housing started one week before the start of the intervention. Food and water were available ad libitum. Animals were housed under standard laboratory conditions ($\frac{12}{12}$ dark—light cycle (lights on at 9:00 a.m.), temperature (22 ± 1 °C) and humidity control (50 ± $10\%$)). Health status of the animals was checked daily and their body weight was registered each week by the researchers. All experimental procedures were evaluated and approved by the national Central Authority for Scientific Procedures on Animals (CCD) and by the local Institutional Animal Welfare Body of University of Groningen (IvD). ## Whole-body vibration procedure We adhered to the new reporting guidelines for WBV studies in animals [33]. Animals were exposed to a vibration session of 10 min twice per day (i.e.: at 10 a.m. and 16 p.m.), five times per week during 5 consecutive weeks (Fig. 1A). Furthermore, animals also received two WBV days on week 6, but only one session on the second day (i.e.: WBV session at 10 a.m.; 24 h before sacrifice). The WBV device has been described before [34, 35]. In short, this device consists of an oscillator (LEVELL R.C. Oscillator Type TG200DMP), a power amplifier (V406 Shaker Power Amplifier) and a cage (44.5 × 28 × 16 cm) separated by 12 removable compartments (6.5 × 7.5 × 20) attached to the oscillator. Throughout a WBV/pseudo WBV session, mice were randomly placed into these individual compartments and were exposed to constant vertical vibrations with a frequency of 30 Hz, an amplitude of 50 micron (100 micron peak-to-peak displacement) and of a sinusoidal nature. These parameters of vibration were verified by additional measurements using a 3D-accelerometer [34]. The individual compartments and the cage were cleaned with $70\%$ ethanol and dry paper tissue between each training session. Fig. 1Experimental design (A): 5 months old male hAPP-J20 mice and their wild (WT) littermates underwent 5 weeks of whole body vibration intervention (WBV) with twice daily session of 10 min exposure, five times per week (gray color marks the days of treatment). After 5 weeks, balance beam test was performed to assess motor coordination. Animals were terminated on week 6 at the age of ~ 6.5 months and brain tissue was collected for immunohistological analyses. Effects of intervention (pseudo vs. WBV) and genotype (J20 vs. WT) on body weight (B), motor performance (C) and plaque load (D). WT animals showed significatly higher body weight during the intervention compared to the J20 animals (B). Walking distance in the balance beam test was only significantly improved in the J20 animals compared to the WT (C). Amyloid plaque deposition in the hippocampus was not significantly affected by WBV intervention (D). Images of 6e10 were taken about the whole hippocampus at 50 × magnifications to visualze total amyloid plaque distribution (E), representative images of areas marked by + are depicted in D. Data are depicted as mean ± SEM. ** indicates: $P \leq .01.$ Scale bars in D are 50 um and in E are 500 um Animals did not receive prior habituation to the experimental settings. Both J20 and WT mice showed slightly exited behavior during the first week of intervention. This general unprompted activity was reduced from the second week onwards. Some escape attempts (i.e.: jumping) were recognized in the J20 mice during the entire intervention. Similarly, a significantly higher degree of defecation (i.e.: number of pellets) was observed in both J20 and WT mice after the WBV sessions in the first week of intervention, which seemed to be normalized only in the WT mice from the second week onwards. In contrast, J20 mice showed this higher degree of defecation after each training session throughout the entire intervention. All vibration sessions were performed in the housing room of the animals. Finally, the animals did not show any kind of acute, short-term and/or long-term side effects by vibration. ## Balance beam The balance beam test was conducted after 5 weeks of WBV treatment to assess sensorimotor coordination with main focus on the functionality of hind limbs [34]. A 1 m long wooden beam (with diameter of 4.5 mm) was placed horizontally 60 cm above the floor. The home cage of the tested animal was positioned at the end of the beam, serving as motivation and target factor for the animal. Mice were familiarized to the experimental setup by two progressive trials (placed on the beam 10 cm and 40 cm away from the target) and subsequently performed four test trials (at 100 cm distance), with 30 s break between all trials to ensure enough recovery to avoid potential injuries induced by muscle fatigue. Video records were taken during the procedures, which were independently analyzed by two researchers. Time needed to cross the beam served as measure of performance and the mean of the three best trials was used as final outcome variable. If an animal was unable or unwilling to cross over the beam it was excluded from the final statistical analysis (1 animal from the pseudo WBV/J20 group; 1 animal from the WBV/WT group; 3 animals from the pseudo WBV/WT group). The beam was cleaned by $70\%$ ethanol and dry paper tissue after each animal. ## Immunohistochemistry Mice were anesthetized with pentobarbital and transracially perfused with saline and $4\%$ paraformaldehyde (PFA) twenty-four hours after the last WBV session. Brain tissue was harvested and postfixed by $4\%$ PFA for 24 h before being transferred to phosphate buffer (PB) for 3 days. After 3 days of washing, brains were dehydrated by $30\%$ sucrose solution and frozen by liquid nitrogen. Brains were stored at − 80 °C until coronal sectioning (20 μm) on a cryostat. Immunohistochemistry was performed to determine microglia and astrocyte features, as well as plaque deposition in the dorsal hippocampus. Free floating sections were used for all staining procedures. Microglia detection. Ionized calcium binding adaptor molecule 1 (IBA1) staining was performed to visualize the morphological state of microglia cells; and cluster of differentiation factor 68 (CD68) was done to determine the level of microglia activation biochemically. Sections were incubated for 3 days at 4 °C by the following primary antibodies: 1) Rabbit anti IBA1 (Wako, SKN4887, 1:2500) in 0.01 M phosphate buffer saline (PBS) containing $1\%$ bovine serum albumin (BSA) and $0.1\%$ Triton-X (TX) or 2) Rat anti CD68 (BioRad, MCA1957GA; 1:1000) in 0.01 M tris buffered saline (TBS) with $5\%$ BSA, $5\%$ normal donkey serum (NDS). Astrocyte detection. Glial fibrillary acidic protein (GFAP) immunohistochemistry was performed to detect astrocyte volume. Sections were pre—incubated in TBS containing $3\%$ BSA and $0.1\%$ TX followed by incubation of primary antibody (Cell Signaling, E4L7M, 1:10000). Plaque detection. Beta amyloid 1–16 (6E10) was stained to detect plaque deposition in the hippocampus. Sections were pre-incubated in 0.01 M TBS with $0.1\%$ TX and $3\%$ normal goat serum before the overnight incubation of primary antibody (BioLegend, SIG-39320; 1:2000). In addition, prior to all stainings, endogenous peroxidase activity was blocked by hydrogen peroxidase (H2O2) ($0.3\%$ for IBA1, GFAP and 6e10; $1\%$ for CD68). For detection, we used biotinylated anti-mouse secondary antibodies (IBA1 and GFAP: Goat Anti Rabbit 1: 500; CD68: Mouse Anti Rat 1:500; 6E10: Goat Anti Mouse 1:400) followed by processing with ABC kit (Vectastain ABC kit, Vector Laboratories) and developed our signal with diaminobenzidine (Sigma Fast, Cat: D4418) and $0.1\%$ H2O2. All sections were intensively washed during the staining processes in PBS or TBS. Sections were mounted on gelation-coated slides, and placed overnight in a drying cabinet. Finally, sections were dehydrated in graded solutions of ethanol and xylol and cover slipped. ## Microscopy Number of microglia, cell body size, dendrites size and total coverage were determined in the Cornu Ammonis 1 (CA1), Cornu Ammonis 3 (CA3), Dentate Gyrus (DG) and Hilus regions of the hippocampus based on the IBA1 staining (200 × magnification). Since microglia activation based on morphology is defined as shortened dendritic processes and increased cell body size, the ratio of the cell body to total cell size was calculated (see [36] for details) as the outcome measure of microglia activation. The coverage (% of the area of interest covered by the immunostaining) of CD68 and GFAP positive cells were determined in the CA1, CA3, DG and Hilus regions (20 × magnification). Similarly, coverage of 6E10 was measured in the dorsal hippocampus (40 × magnification). All analyses were performed by Image J software. ## Statistical analysis Statistical analysis was performed by Statistica 13.2 software. 2 × 2 factorial ANOVAs were performed with intervention (vibration/pseudo vibration) and genotype (J20/WT) as factors and, in case of significant intervention x genotype interaction, followed by Tukey’s post hoc test to reveal statistical differences between the four groups in balance beam performance, IBA1, GFAP and CD68 stainings. In addition, amyloid plaque deposition between the two J20 groups was compared using independent T-tests. Mixed design repeated measurements ANOVA was performed with intervention (WBV, pseudo WBV) and genotype (J20, WT) as between-subjects factors and time course (week 1–5) as within-subjects factor to analyze differences in body weight. Statistical significance was set at $p \leq 0.05.$ Graphs were created by GraphPad Prism 8 software. Descriptive data were expressed as mean ± SEM. ## Body weight and motor performance Mixed design repeated measures ANOVA showed a significantly higher body weight in the WT groups compared to the J20 animals (main effect genotype: F(1.46) = 15.060; $p \leq 0.001$) (Fig. 1B). In addition, a strong tendency of lower body weight was observed in the WBV treated groups compared to the pseudo-WBV groups (main effect intervention: F(1.46) = 3.978; $$p \leq 0.052$$). A significant effect of time course was also revealed (main effect time course: F(5.230) = 4.871; $p \leq 0.001$). Further, the interaction of genotypetime course showed significant difference (intervention genotype x time interaction: F(5.230) = 2.788; $$p \leq 0.018$$) and additional post-hoc analysis revealed decreased body weight in the WT animals for week 1 vs. week 3–6 (post hoc $p \leq 0.05$), but not in the J20s. The balance beam test was used in week 5 to assess motor coordination and functionality of primarily the hind limbs. Two-way ANOVA revealed that the J20 animals showed significantly better walking performance compared to the WT animals (Main effect genotype: F(1.41) = 9.410; $$p \leq 0.003$$) (Fig. 1C). In addition, a trend of improved motor coordination was found in the vibration treated groups compared with pseudo vibration treated animals (vibration treated animals showed decreased walking time on the beam), but it did not reach statistical significance (Fig. 1C). No significant interaction effect of intervention and genotype was observed in the balance beam test. ## Plaque formation To determine whether WBV affects the level of amyloid deposition in J20 mouse, the total coverage of 6e10 staining was measured in the hippocampus (Hilus, DG, CA3, and CA1 subregions pooled). An early plaque load was found in the hippocampus of J20 mice at the age of six months. In contrast, plaques were not present in the hippocampus of their WT littermates. Furthermore, plaque load was not significantly different between WBV and pseudo-WBV treated J20 groups (Fig. 1D). Representative images of amyloid plaque load are depicted in Fig. 1E. ## Microglia To determine whether differences in microglia activation in the subregions of hippocampus (CA1, CA2, DG and Hilus) existed across genotype and WBV intervention, activated microglia were identified by the expression of IBA1 and CD68 positive cells; two frequently used markers for microglia. Morphological parameters of microglia based on the IBA1 immunostaining were determined including number, total coverage, cell body and dendrites area in the CA1, CA3, DG and Hilus hippocampal subregions. Representative images of IBA1 expression in the Hilus subregion are visualized in Fig. 2A. Significantly decreased total coverage and increased cell body size were detected in the J20 animals compared to the WT controls in all hippocampal subregions. It was also observed that WBV significantly ameliorated the size of dendritic processes in the CA1 (Main effect intervention: F(1.40) = 5.636; $$p \leq 0.022$$) and DG subregions (Main effect intervention: F(1.40) = 15.15; $p \leq 0.001$) (Fig. 2C). Microglia number was not significantly altered. These morphological outcomes are summarized in Table 1. Furthermore, microglia activation was calculated based on the IBA1 expression as the ratio of the cell body to total cell size. Microglia activation of the 4 investigated subregions are summarized in Fig. 2B. Significantly higher degree of microglia activation was detected in the J20 groups compared to their WT littermates in the CA1, CA3, DG and Hilus subregions (Fig. 2B). In addition, decreased microglia activation in the DG subregion was observed in the WBV vs. pseudo-WBV groups (Main effect intervention: F(1.40) = 5.738; $$p \leq 0.021$$) (Fig. 2B).Fig. 2Microglia visualized by IBA1 staining in the Hilus subregions is depicted in (A). Effects of intervention (pseudo vs. WBV) and genotype (WT vs. J20) on microglia activation in the CA1, CA3, DG and Hilus subregions are depicted in (B). Significant increase of microglia activation was observed in the J20 animals compared to the WT controls in all subregions (B, CA1, CA3, Dentate Gyrus and Hilus). Microglia activation was only significantly decreased by WBV in the DG (B, Dentate Gyrus). Furthermore, WBV treatment significantly increased the size of microglia dendritic processes in the CA1 and DG subregions (C, CA1 and Dentate Gyrus). Expression of CD68 coverage in the CA1 subregion is visualized in D. Significant decrease of CD68 expression in the CA1 area was detected in the J20 animals compared to the WT controls (E, CA1). In addition, a tendency ($$p \leq 0.06$$) of interaction effect (intervention vs. genotype) on CD68 expression was also observed in the CA1 subregion; additional post-hoc analysis showed a significant decrease for the J20—WBV group compared to the WT—WBV group, as well as the same tendency ($$p \leq 0.07$$) compared to the WT—pseudo WBV group (E, CA1). A significant effect of interaction (intervention vs. genotype) in CD68 expression was observed in the DG area; additional post hoc revealed a strong tendency ($$p \leq 0.08$$) of decrease CD68 expression in the J20—WBV treated animals compared to the J20—pseudo WBV controls (E, Dentate Gyrus). CD68 expression was not significantly altered in the CA3 and Hilus areas (E, CA3 and Hilus). Representative Images of IBA1 and CD68 were taken about the Hilus and CA1 subregions at 200 × magnifications to visualze microglia (A and C). Relevant statistical differences are marked in B, C and D. Data are depicted as mean ± SEM. indicates: $$P \leq .05.$$ Scale bars in A and D are 100 umTable 1Effects of intervention (pseudo WBV vs. WBV) and genotype (J20 vs. WT) on microglia number (n), microglia coverage (in %), cell body size and dendrites size measured in pixels, and the number of cells in the CA1, CA3, DG and Hilus subregions of the hippocampusGroupRegionMicroglia number (n)Total coverage (in %)Cell body area (px)Dendrites area (px)Pseudo WBV—WTCA133 ± 1.916 ± 0.9237 ± 155647 ± 342CA333 ± 2.016 ± 1.1244 ± 126094 ± 531DG31 ± 1.916 ± 0.9270 ± 164993 ± 393Hilus25 ± 2.015 ± 1.1262 ± 214201 ± 344Pseudo WBV—J20CA133 ± 1.113 ± 0.6 + 353 ± 39 + 5728 ± 332CA332 ± 2.013 ± 0.5 + 328 ± 42 + 5734 ± 510DG33 ± 1.012 ± 0.4 + 365 ± 41 + 4978 ± 277Hilus21 ± 1.511 ± 0.6 + 398 ± 41 + 3464 ± 214WBV—WTCA136 ± 2.716 ± 0.8257 ± 175952 ± 404CA332 ± 1.816 ± 0.7252 ± 145948 ± 379DG31 ± 1.916 ± 0.7267 ± 146129 ± 261Hilus21 ± 2.014 ± 0.8271 ± 124460 ± 308WBV—J20CA134 ± 1.915 ± 0.7 + 348 ± 45 + 6950 ± 163CA331 ± 1.615 ± 0.9 + 349 ± 35 + 6764 ± 355DG34 ± 1.614 ± 0.7 + 368 ± 43 + 6335 ± 299Hilus22 ± 2.711 ± 0.7 + 361 ± 42 + 3829 ± 448Microglia number was not significantly altered by genotype or intervention. Significantly higher coverage was detected in the J20 mice compared to the WT mice in all subregions. In contrast, cell body size was significantly increased in the J20 animals compared to the WT animals. WBV treatment significantly increased the size of dendritic processes in both the J20 and WT mice in the CA1 and DG areas. Data are depicted as mean ± SEM. + indicates a significant difference between the main factors “wild type vs. J20”indicates a significant difference between main factors “WBV vs. pseudo WBV”. Additional figures related to these parameters can be found in the Additional file 1 Total coverage of CD68 staining, a protein highly expressed by activated microglia, was measured in the same subregions of hippocampus. Representative images of CD68 in the CA1 subregion are visualized in Fig. 2D. Data of the 4 investigated subregions are depicted in Fig. 2E. CD68 coverage in the CA1 region was significantly decreased by the J20 genotype (main effect genotype: F(1.43) = 6.767; $$p \leq 0.012$$). In addition, a tendency of intervention x genotype interaction was also observed ($$p \leq 0.06$$) with a lower coverage in the CA1 region for J20-WBV vs. WT pseudo-WBV (post-hoc: 0.07) and WT-WBV animals (post-hoc: 0.01). Furthermore, a significant intervention x genotype effect was found in the DG subregions (F(1.42) = 6.999; $$p \leq 0.011$$). Further post-hoc analysis showed a tendency of lower CD68 coverage in the J20-WBV group compared to the J20 pseudo-WBV controls (post-hoc $$p \leq 0.08$$). ## Astrocytes To determine whether hippocampal astrocyte volume would be affected by WBV and/or genotype, covered area of GFAP positive cells were quantified in CA1, CA3, DG and Hilus hippocampal subregions. Representative images of GFAP in the CA3 subregion are visualized in Fig. 3A. Data of the 4 investigated subregions are also summarized in Fig. 3B. Two-way factorial ANOVA revealed a significantly lower volume of GFAP positive cells in the CA1 and Hilus subregions of J20 animals compared to their WT littermates (Fig. 3B. Further, WBV significantly increased astrocyte coverage in the Hilus subregions (F(1.41) = 20.96; $p \leq 0.001$) (Fig. 3B. A genotype x intervention was found in the CA3 subregion (F(1.39) = 10.58; $$p \leq 0.002$$). It was found that WT animals showed lower coverage in WBV vs. pseudo group, whereas the J20 animals had higher coverage in WBV vs. pseudo group. Additional post-hoc analysis revealed that the coverage in CA3 region was significantly higher for WBV vs. pseudo-WBV in J20 mice (post hoc = 0.023) and it was lower for J20 – pseudo-WBV vs. WT pseudo-WBV mice (post-hoc = 0.009) (Fig. 3B). Consistent with these observations, a similar trend was detected in the CA1 region, but this interaction effect did not reach statistical significance. Fig. 3GFAP + astrocytes in the CA3 hipocampal subregion are visualized in (A). Effects of intervention (pseudo vs. WBV) and genotype (J20 vs. WT) on expression of GFAP positive cells in the CA1, CA3, DG and Hilus regions are depicted in B. Significantly decreased coverage of GFAP positive cells in the J20 animals was detected in the CA1 and Hilus regions (B, CA1 and Hilus). The same trend was observed in the CA3 and Hilus subregions (B, CA3 and Hilus). Significant effect of interaction (intervention vs. genotype) was detected in the CA1 and CA3 subregions (B, CA1 and CA3). Additional posthoc analysis revealed that WT—WBV and pseudo WBV groups showed significantly higher coverage compared to the J20 – pseudo WBV group in the CA1 region (B, CA1). Significantly increased GFAP coverage was detected in the J20—WBV and WT—pseudo WBV groups compared to the J20—pseudo WBV group in the CA3 region (B, CA3). In addition, significant increase of GFAP coverage was detected by WBV in the Hilus region (Panel B, Hilus). Representative images of GFAP + astrocytes were taken about the CA3 hippocampal subregion at 200 × magnifications (A). Data are depicted as mean ± SEM. indicates: $$P \leq .05.$$ $P \leq .01.$ * $P \leq .001.$ Scale bars in A are 100 um ## Discussion The aim of this experiment was to investigate the therapeutic impact of long-term (five weeks) WBV intervention with low intensity of a sinusoidal nature on amyloid deposition, neuroinflammation and motor performance during the early stage of AD in hAPP-J20 transgenic mice. Our results demonstrated that exposure of J20 mice to WBV for five weeks ameliorated the early progression of astroglial pathology. In contrast, WBV did not influence microglia activation or the amyloid plaque load in the human APP-J20 mouse model. A reduction in volume (coverage) of GFAP positive astrocytes in the hippocampus was found in J20 vs. WT mice including the CA1, DG and Hilus. This is in line with other reports showing a reduction in volume of GFAP positive astrocytes in J20 mice of 5–6 months of age, however, it was not visible at 8 months indicating that astrocyte population may only decrease during the early stage of AD pathogenesis [37, 38]. In contrast to these results, a significantly increased coverage of GFAP positive astrocytes has been reported in J20 mice at 6 and 9 months of age [32, 39]. These previously reported findings suggest the presence of an early AD-related deficit; however, certain recovery mechanisms may become activated later on in life. These findings are most likely due to the complex balance between neurotoxic and neuroprotective effects depending on the disease stage and microenvironmental factors (Rodríguez-Giraldo et al., 2022 and references therein [14]). Although WBV did ameliorate the astroglial pathological changes in J20 mice, it did not reduce plaque load. This is in line with the findings from Wang et al. that early activation of astrocytes at the age of 3–5 months does not influence the deposition of amyloid plaques in the brains of J20 mice [40]. As far as we know, neither WBV as a form of passive exercise nor active exercise have been investigated regarding hippocampal functioning in the human APP-J20 mouse model. However, another form of cognitive stimulation, known as enrichment environment, has been reported to prevent astroglial volume and morphological changes in the early stage of AD in the hippocampus of J20 mice [37]. Long-term environmental enrichment restored the astrocyte parameters similar to the age-matched non-transgenic control animals. In their design, no exercise devices (for instance: running wheel or disc) were included to ensure mainly cognitive stimulation by the enrichment environment. In our current study, we demonstrated similar effects that altered astrocyte morphology regarding AD progression can be reversed by exposure to a long-term WBV intervention. Hence, WBV seems to be able to mimic the effects of enrichment environment in the early stage of AD in J20 animals. One of the mostly emphasized effects of WBV is stimulating and improving the musculoskeletal system. This is based on ample clinical and pre-clinical studies [41–43]. In rodents, low-intensity WBV improves neuromuscular dynamics, muscle strength and motor coordination [17, 22, 23, 34]. Although this neuromuscular response to WBV appears to be a pivotal adaptation, only a trend of improved motor coordination was found in the present study for both J20 and WT mice. J20 animals outperformed the WT littermates in the balance beam test. This is most likely due to the known hyperactivity of the J20 mice. Altered locomotor activity such as hyperactivity and disturbed home cage activity has been commonly reported in AD mouse models including J20 mice, which are often associated with increased amyloid levels and disease progression [44]. The onset of these disturbances varies between different models, however, the J20 model is one of them that seems to develop disruptions the earliest (around 1 month of age) [44]. We hypothesize, that the J20 animals approached the upper limit of their performance due to their disturbed locomotor behavior and thereby contributed to the mitigation of WBV’s effects on motor coordination. In addition, this trend of improved motor performance appears to be more pronounced in the WT animals. This observation seems to be in line with our previous studies reported in young mice [34] and old rats [22, 23]. We found in earlier studies from our research group that long-term (5 weeks) WBV has broad effects in young mice including improvements in motor performance, memory functions and levels of neurotransmitters [34, 35, 45]. In contrast, WBV did not influence the body weight of the animals in these studies [34, 35, 45]. A possible explanation is that here we used WBV twice per day, instead of once. The extra physical activity that comes with the handling procedure (for both the WBV and the pseudo-WBV groups) could account for this finding. However, it should be noted that the observed reduction of body weight over the time course of the intervention is very small and probably biologically irrelevant for the mice. In accordance with the literature, we found an early plaque load in the hippocampus of J20 mice at the age of 6.5 months. Further, the number and volume of amyloid beta plaques seem to be comparable to those that have been reported in 5–6 months old J20 mice [37, 39]. Although fewer animals with a relatively high plaque load were observed after WBV, no significant overall reduction was found. Apparently WBV did not affect plaque load, although studies using ultrasound-based vibrational alternatives seem to reduce plaque load in various AD models [46–50]. However, it is important to emphasize that these experiments do not show unanimity regarding outcome measures and methodical approaches, and direct comparison with WBV is limited. Our results may though suggest that the vibratory aspect of ultrasound-based therapies is not the main causal factor for the observed findings. There is evidence for the toxicity of amyloid beta regarding glial activation including both microglia and astrocytes [6–8]. We found significantly increased microglia activation (based on morphology and predominantly an increase in cell body size) in all hippocampal subregions of J20 animals compared to their WT littermates. In contrast, CD68 positive cells only showed an increase in the CA1 region. These observed discrepancies between both microglia markers could be related to the complex early events in the progression of neuroinflammation. Available data from literature also indicate that a significantly increased number of activated CD68 and IBA1 positive microglia cells was detectable in the hippocampus of J20 mice at 6–9 months of age compared to age-matched WT controls [37, 39]. Our findings seem to be in line with these previously reported observations. It was also found that WBV was able to ameliorate microglia activation in both J20 and WT animals in the DG subregions. This finding seems to be associated with increased dendritic processes of microglia in the DG and CA1 regions. However, despite the same tendency microglia activation in the CA1 regions was not significantly altered by WBV treatment. Furthermore, WBV did not affect cell body size in all investigated subregions. This finding indicates that WBV might be a more beneficial stimulus in the DG to shift microglia towards a more ramified morphology due to amelioration of dendritic processes, such as sensing the surrounding tissue. These discrepancies in the observed parameters might also be related to the relative early stage of the disease. Taken together, the beneficial effects of WBV on hippocampal microglia activation have been reported recently. Our research group found that WBV is able to reduce aging-related neuroinflammation associated with higher degree of microglia ramification in 18 months old rats [23]. Our current findings seem to be in line with this observation. Furthermore, it was reported by others, that long-term WBV intervention alleviates increased level of microglia immunostaining and reversed the decreased level of GFAP positive astrocytes in the CA1 hippocampal subregions of Sprague–Dawley rats induced by restrain stress test [21]. These findings suggest that WBV intervention in older J20 mice may yield different results as seen here in young J20 mice due to increased responsiveness of the microglia. The positive impact of WBV on the functioning of compromised astrocytes is a novel finding. Traditionally, astrocytic pathology is characterized by an increase in the volume of GFAP-positive astrocytes, most notably seen in astrogliosis (see, for review, Kim et al., 2018 and references therein [51]). However, a decrease in the volume of hippocampal GFAP-immunoreactive astrocytes has been found in relation to depression and mood-disorders in human tissue and after posttraumatic stress disorder in rats [52, 53]. A decrease in the volume or coverage is most likely caused by a retraction of the astrocytic processes which will cause a decreased participation of astrocytic end feet in the tripartite synapse [54]. Hence, it could be that this type of astrocytic pathology reduces the uptake of excess of synaptic glutamate, leading to increased risk of excitotoxicity as also suggested by others [53]. Reduced astrocytic functioning due to shrinkage also negatively affects the release of neurotrophic factors [51]. We therefore interpret the decrease of GFAP in our data as a sign of pathology, and the recovery by WBV to the levels as seen in control animals as the prevention of pathology or its reversal if it was already present before we started the WBV intervention. Restoring the volume (coverage) of GFAP-positive astrocytes could promote cellular signaling and synaptic plasticity, functions know to be sensitive to WBV [20; 45]. Both astrocytes and microglia can modify their morphology in response to their direct cellular vinicity [6–8] and are endowed with receptors for different types of neurotrophic factors [55–57] and neurotransmitters [58]. For instance: astrocytes posess TrkB1 receptors, the binding site of the neurotrophic factor BDNF, as well as cholinergic, serotonergic and dopaminergic receptors. These findings suggest that these morphological alterations could be mediated through multiple pathways and might be crucial for neural activity, synaptic plasticity and maintenance [58, 59]. It is also known that WBV exposure can stimulate the release of various neurotransmitters in different brain regions including the hippocampus [45, 60, 61], potentially supporting neuronal activity and health. Similarly, an increased level of BDNF was also detected after long-term vibration interventions [21, 25]. Populations of reactive astrocytes localized around the senile plaques could be able to produce, together with microglia, a wide range of pro-inflammatory molecules and contribute to the inflammatory state [62]. WBV may have the therapeutic potential to mitigate the level of pro-inflammatory factors after brain damage [25]. This preventive effect of WBV on astroglia volume might also be associated with the mitigation of pro-inflammatory responses. ## Limitations Some limitations need to be addressed. Although the design of our WBV protocol was chosen and planned carefully, it required the use of pseudo-control groups to determine the sole effect of the vibrations. The control animals underwent pseudo-treatment and may also experience improvements in reducing the progression of amyloid pathology, as a result of the exposure to a new environment (i.e. the compartments of the vibration plate). A significant decline of body weight was detected during the first two weeks of the intervention in the WT animals, but not in the J20 animals (time course * genotype). Further, a strong tendency of lower body weight was explored in the WBV treated group (main factor: intervention), however, the interaction of time course and intervention was not significantly altered (time course * intervention). Although the animals underwent prior habituation to the experimental room and its conditions before the start of the intervention, this indicates that the WT animals might have experienced some discomfort or stress or have been more sensitive during the first two weeks of the intervention. Over the last years, we have not experienced this kind of fluctuation in body weight in mice or rats in our former experiments [22, 23, 34, 35, 45]. Also, neither WBV nor pseudo WBV did influence the body weight of mice in our previous works with the same WBV device and settings [34, 35, 45]. We made the same observations in rats [22, 23]. This minor decline in body weight only in WT mice indicates that modifications in habituation and handling procedures may be considered in future projects. Finally, the WT animals had significantly higher body weights compared to the J20 (main effect genotype). Whether this difference between WT and J20 animals in body weight influenced the efficiency of the WBV treatment is unknown. ## Recommandations for future research The initially stated aim of this study was to determine the contextual underlying mechanisms that might contribute to the beneficial effects of WBV in the J20 mouse model, a well-known model for AD. While recognizing the limitations of this research, we think, that we were able, at least in part, to achieve our objectives. We have identified the involvement of astrocytes in WBV-mediated effects. These findings can lead to specific outcomes in determining the research questions and strategies of future studies. Based on the available body of literature regarding WBV and AD (as well the J20 mouse model), we concluded that future studies should examine the influence of WBV on cognition and behavior including depression, anxiety and memory functions, as well as on further molecular markers related to neurogenesis, synaptic plasticity, growth factors, inflammatory and other neural markers. All of these research objectives could be relevant future perspectives in relation to AD and WBV. ## Conclusion The results of this study suggest that glial changes in the early phase of amyloid pathology could be prevented by chronic (five weeks) exposure to low-intensity WBV. Our results contribute to the understanding of glial plasticity in response to WBV, which can be considered a new, potential therapeutic approach for neurodegenerative diseases. The clinical relevance has yet to be determined, and may be restricted to the early phase of AD as we found that WBV in late phase AD patients could not improve their cognitive performance, despite the demonstrated feasibility of WBV for (fragile) AD patients [63]. Our results seem to be consistent with the existing literature and indicate that glial cells can respond to vibrational stimuli adopting their volume to the condition as was found in the control mice. The underlying mechanism(s) could implicate parallel molecular and cellular responses such as altered energy metabolism, recycling and/or release of neurotransmitters and neurotrophic factors, especially in the plastic brain areas like the hippocampus. Whether these mechanisms indeed play a key role has to be determined in future studies. Further, the understanding and unraveling of these underlying mechanisms by translational scientific investigations can contribute to more advanced and effective study procedures and WBV protocols as indicated earlier by our research group [64]. In conclusion, the possibility that the progression of derailing of microglial and astroglial activation such as seen in neuroinflammation in the early stage of AD (and most likely other types of neurodegenerative diseases) can be slowed by application of WBV, as a passive alternative for active exercise, puts forward WBV as a treatment strategy worthwhile to pursuit. Notably for those unable to participate in active exercise protocols. ## Supplementary Information Additional file 1: Figure S1. Effects of genotype x time course (panel A) and time course (panel B) were observed on body weight. Body weight was significantly decreased from week 1 - 2 to week 3 – 6 in the wild type animals. In contrast, this effect was not observed in the J20 animals. Significant decrease of body weight (effect of tiem course) was also revealed on week 1 vs. week 3, 5 and 6. ## References 1. Breijyeh Z, Karaman R. Comprehensive review on Alzheimer’s disease: causes and treatment. Molecules (2020) 25 5789. DOI: 10.3390/molecules25245789 2. Fakhoury M. Microglia and astrocytes in Alzheimer’s disease: implications for therapy. Curr Neuropharmacol (2018) 16 508-518. DOI: 10.2174/1570159X15666170720095240 3. Hansen DV, Hanson JE, Sheng M. Microglia in Alzheimer’s disease. J Cell Biol (2018) 217 459-472. DOI: 10.1083/jcb.201709069 4. Matsuoka Y, Picciano M, Malester B, LaFrancois J, Zehr C, Daeschner JM. Inflammatory responses to amyloidosis in a transgenic mouse model of Alzheimer’s disease. Am J Pathol (2001) 158 1345-1354. DOI: 10.1016/S0002-9440(10)64085-0 5. Ries M, Sastre M. Mechanisms of Aβ clearance and degradation by glial cells. Front Aging Neurosci (2016). DOI: 10.3389/fnagi.2016.00160 6. Solito E, Sastre M. Microglia function in Alzheimer’s disease. Front Pharmacol (2012). DOI: 10.3389/fphar.2012.00014 7. Lee CYD, Landreth GE. The role of microglia in amyloid clearance from the AD brain. J Neural Transm (2010) 117 949-960. DOI: 10.1007/s00702-010-0433-4 8. Rodríguez JJ, Olabarria M, Chvatal A, Verkhratsky A. Astroglia in dementia and Alzheimer’s disease. Cell Death Differ (2009) 16 378-385. DOI: 10.1038/cdd.2008.172 9. Mendiola-Precoma J, Berumen LC, Padilla K, Garcia-Alcocer G. Therapies for prevention and treatment of Alzheimer’s disease. Biomed Res Int (2016) 2016 1-17. DOI: 10.1155/2016/2589276 10. Vaz M, Silvestre S. Alzheimer’s disease: recent treatment strategies. Eur J Pharmacol (2020) 887 173554. DOI: 10.1016/j.ejphar.2020.173554 11. Sanders LMJ, Hortobágyi T, la Bastide-van GS, van der Zee EA, van Heuvelen MJG. Dose-response relationship between exercise and cognitive function in older adults with and without cognitive impairment: a systematic review and meta-analysis. PLoS ONE (2019) 14 e0210036. DOI: 10.1371/journal.pone.0210036 12. Jahangiri Z, Gholamnezhad Z, Hosseini M. Neuroprotective effects of exercise in rodent models of memory deficit and Alzheimer’s. Metab Brain Dis (2019) 34 21-37. DOI: 10.1007/s11011-018-0343-y 13. De la Rosa A, Olaso-Gonzalez G, Arc-Chagnaud C, Millan F, Salvador-Pascual A, García-Lucerga C. Physical exercise in the prevention and treatment of Alzheimer’s disease. J Sport Heal Sci (2020) 9 394-404. DOI: 10.1016/j.jshs.2020.01.004 14. Rodríguez-Giraldo M, González-Reyes RE, Ramírez-Guerrero S, Bonilla-Trilleras CE, Guardo-Maya S, Nava-Mesa MO. Astrocytes as a therapeutic target in alzheimer’s disease–comprehensive review and recent developments. Int J Mol Sci (2022) 23 13630. DOI: 10.3390/ijms232113630 15. Runge M, Rehfeld G, Resnicek E. Balance training and exercise in geriatric patients. J Musculoskelet Neuronal Interact (2000) 1 61-65. PMID: 15758528 16. Zhang L, Weng C, Liu M, Wang Q, Liu L, He Y. Effect of whole-body vibration exercise on mobility, balance ability and general health status in frail elderly patients: a pilot randomized controlled trial. Clin Rehabil (2014) 28 59-68. DOI: 10.1177/0269215513492162 17. Mettlach G, Polo-Parada L, Peca L, Rubin CT, Plattner F, Bibb JA. Enhancement of neuromuscular dynamics and strength behavior using extremely low magnitude mechanical signals in mice. J Biomech (2014) 47 162-167. DOI: 10.1016/j.jbiomech.2013.09.024 18. Patel VS, Chan ME, Pagnotti GM, Frechette DM, Rubin J, Rubin CT. Incorporating refractory period in mechanical stimulation mitigates obesity-induced adipose tissue dysfunction in adult mice. Obesity (2017) 25 1745-1753. DOI: 10.1002/oby.21958 19. Corbiere T, Weinheimer-Haus E, Judex S, Koh T. Low-intensity vibration improves muscle healing in a mouse model of laceration injury. J Funct Morphol Kinesiol (2017) 3 1. DOI: 10.3390/jfmk3010001 20. Cariati I, Bonanni R, Pallone G, Annino G, Tancredi V, D’Arcangelo G. Modulation of synaptic plasticity by vibratory training in young and old mice. Brain Sci (2021) 11 82. DOI: 10.3390/brainsci11010082 21. Peng G, Yang L, Wu CY, Zhang LL, Wu CY, Li F. Whole body vibration training improves depression-like behaviors in a rat chronic restraint stress model. Neurochem Int (2021) 142 104926. DOI: 10.1016/j.neuint.2020.104926 22. Oroszi T, Geerts E, de Boer SF, Schoemaker RG, van der Zee EA, Nyakas C. Whole body vibration improves spatial memory, anxiety-like behavior, and motor performance in aged male and female rats. Front Aging Neurosci (2022). DOI: 10.3389/fnagi.2021.801828 23. Oroszi T, de Boer SF, Nyakas C, Schoemaker RG, van der Zee EA. Chronic whole body vibration ameliorates hippocampal neuroinflammation, anxiety-like behavior, memory functions and motor performance in aged male rats dose dependently. Sci Rep (2022) 12 9020. DOI: 10.1038/s41598-022-13178-1 24. Huang D, Yang Z, Wang Z, Wang P, Qu Y. The macroscopic and microscopic effect of low-frequency whole-body vibration after cerebral ischemia in rats. Metab Brain Dis (2018) 33 15-25. DOI: 10.1007/s11011-017-0113-2 25. Raval A, Schatz M, Bhattacharya P, d’Adesky N, Rundek T, Dietrich W. Whole body vibration therapy after ischemia reduces brain damage in reproductively senescent female rats. Int J Mol Sci (2018) 19 2749. DOI: 10.3390/ijms19092749 26. Monteiro F, Sotiropoulos I, Carvalho Ó, Sousa N, Silva FS. Multi-mechanical waves against Alzheimer’s disease pathology: a systematic review. Transl Neurodegener (2021) 10 36. DOI: 10.1186/s40035-021-00256-z 27. Mucke L, Masliah E, Yu G-Q, Mallory M, Rockenstein EM, Tatsuno G. High-level neuronal expression of Aβ. J Neurosci (2000) 20 4050-4058. DOI: 10.1523/JNEUROSCI.20-11-04050.2000 28. Harris JA, Devidze N, Halabisky B, Lo I, Thwin MT, Yu G-Q. Many neuronal and behavioral impairments in transgenic mouse models of Alzheimer’s disease are independent of caspase cleavage of the amyloid precursor protein. J Neurosci (2010) 30 372-381. DOI: 10.1523/JNEUROSCI.5341-09.2010 29. Lustbader JW, Cirilli M, Lin C, Xu HW, Takuma K, Wang N. ABAD Directly links Aß to mitochondrial toxicity in Alzheimer’s disease. Science (2004) 304 448-52. DOI: 10.1126/science.1091230 30. Cheng IH, Scearce-Levie K, Legleiter J, Palop JJ, Gerstein H, Bien-Ly N. Accelerating amyloid-β fibrillization reduces oligomer levels and functional deficits in Alzheimer disease mouse models. J Biol Chem (2007) 282 23818-23828. DOI: 10.1074/jbc.M701078200 31. Lalonde R, Kim HD, Fukuchi K. Exploratory activity, anxiety, and motor coordination in bigenic APPswe + PS1/ΔE9 mice. Neurosci Lett (2004) 369 156-161. DOI: 10.1016/j.neulet.2004.07.069 32. Wright AL, Zinn R, Hohensinn B, Konen LM, Beynon SB, Tan RP. Neuroinflammation and neuronal loss precede Aβ plaque deposition in the hAPP-J20 mouse model of Alzheimer’s disease. PLoS ONE (2013) 8 e59586. DOI: 10.1371/journal.pone.0059586 33. van Heuvelen MJG, Rittweger J, Judex S, Sañudo B, Seixas A, Fuermaier ABM. Reporting guidelines for whole-body vibration studies in humans, animals and cell cultures: a consensus statement from an international group of experts. Biology (2021) 10 965. DOI: 10.3390/biology10100965 34. Keijser JN, Marieke JG, van Heuvelen MJG, Nyakas C, Toth K, Schoemaker RG. Whole body vibration improves attention and motor performance in mice depending on the duration of the whole body vibration session. African J Tradit Complement Altern Med (2017) 14 128-34. DOI: 10.21010/ajtcam.v14i4.15 35. Boerema AS, Heesterbeek M, Boersma SA, Schoemaker R, de Vries EFJ, van Heuvelen MJG. Beneficial effects of whole body vibration on brain functions in mice and humans. Dose Response (2018) 16 155932581881175. DOI: 10.1177/1559325818811756 36. Hovens I, Nyakas C, Schoemaker R. A novel method for evaluating microglial activation using ionized calcium-binding adaptor protein-1 staining: cell body to cell size ratio. Neuroimmunol Neuroinflamm (2014) 1 82. DOI: 10.4103/2347-8659.139719 37. Beauquis J, Pavía P, Pomilio C, Vinuesa A, Podlutskaya N, Galvan V. Environmental enrichment prevents astroglial pathological changes in the hippocampus of APP transgenic mice, model of Alzheimer’s disease. Exp Neurol (2013) 239 28-37. DOI: 10.1016/j.expneurol.2012.09.009 38. Fu Y, Rusznák Z, Kwok JBJ, Kim WS, Paxinos G. Age-dependent alterations of the hippocampal cell composition and proliferative potential in the hAβPPSwInd-J20 mouse. J Alzheimer Dis (2014) 41 1177-1192. DOI: 10.3233/JAD-132717 39. Ameen-Ali KE, Simpson JE, Wharton SB, Heath PR, Sharp PS, Brezzo G. The time course of recognition memory impairment and glial pathology in the hAPP-J20 mouse model of Alzheimer’s disease. J Alzheimer Dis (2019) 68 609-624. DOI: 10.3233/JAD-181238 40. Wang D, Zhang X, Wang M, Zhou D, Pan H, Shu Q. Early activation of astrocytes does not affect amyloid plaque load in an animal model of Alzheimer’s disease. Neurosci Bull (2018) 34 912-920. DOI: 10.1007/s12264-018-0262-2 41. Alam MM, Khan AA, Farooq M. Effect of whole-body vibration on neuromuscular performance: a literature review. Work (2018) 59 571-583. DOI: 10.3233/WOR-182699 42. Savage R, Billing D, Furnell A, Netto K, Aisbett B. Whole-body vibration and occupational physical performance: a review. Int Arch Occup Environ Health (2016) 89 181-197. DOI: 10.1007/s00420-015-1062-x 43. Prisby RD, Lafage-Proust M-H, Malaval L, Belli A, Vico L. Effects of whole body vibration on the skeleton and other organ systems in man and animal models: What we know and what we need to know. Ageing Res Rev (2008) 7 319-329. DOI: 10.1016/j.arr.2008.07.004 44. Webster SJ, Bachstetter AD, Nelson PT, Schmitt FA, Van Eldik LJ. Using mice to model Alzheimer’s dementia: an overview of the clinical disease and the preclinical behavioral changes in 10 mouse models. Front Genet (2014). DOI: 10.3389/fgene.2014.00088 45. Heesterbeek M, Jentsch M, Roemers P, Keijser JN, Toth K, Nyakas C. Whole body vibration enhances choline acetyltransferase-immunoreactivity in cortex and amygdala. J Neurol Transl Neurosci (2017) 5 1079 46. Bobola MS, Chen L, Ezeokeke CK, Olmstead TA, Nguyen C, Sahota A. Transcranial focused ultrasound, pulsed at 40 Hz, activates microglia acutely and reduces Aβ load chronically, as demonstrated in vivo. Brain Stimul (2020) 13 1014-1023. DOI: 10.1016/j.brs.2020.03.016 47. Leinenga G, Götz J. Safety and efficacy of scanning ultrasound treatment of aged APP23 mice. Front Neurosci (2018) 12 55. DOI: 10.3389/fnins.2018.00055 48. D’Haese P-F, Ranjan M, Song A, Haut MW, Carpenter J, Dieb G. β-Amyloid plaque reduction in the hippocampus after focused ultrasound-induced blood-brain barrier opening in Alzheimer’s disease. Front Hum Neurosci (2020) 14 593672. DOI: 10.3389/fnhum.2020.593672 49. Jordão JF, Thévenot E, Markham-Coultes K, Scarcelli T, Weng Y-Q, Xhima K. Amyloid-β plaque reduction, endogenous antibody delivery and glial activation by brain-targeted, transcranial focused ultrasound. Exp Neurol (2013) 248 16-29. DOI: 10.1016/j.expneurol.2013.05.008 50. Poon CT, Shah K, Lin C, Tse R, Kim KK, Mooney S. Time course of focused ultrasound effects on β-amyloid plaque pathology in the TgCRND8 mouse model of Alzheimer’s disease. Sci Rep (2018) 8 14061. DOI: 10.1038/s41598-018-32250-3 51. Kim R, Healey KL, Sepulveda-Orengo MT, Reissner KJ. Astroglial correlates of neuropsychiatric disease: from astrocytopathy to astrogliosis. Prog Neuropsychopharmacol Biol Psychiatry (2018) 87 126-146. DOI: 10.1016/j.pnpbp.2017.10.002 52. Saur L, Baptista PP, Bagatini PB, Neves LT, de Oliveira RM, Vaz SP. Experimental post-traumatic stress disorder decreases astrocyte density and changes astrocytic polarity in the CA1 hippocampus of male rats. Neurochem Res (2015) 41 892-904. DOI: 10.1007/s11064-015-1770-3 53. Cobb JA, O’Neill K, Milner J, Mahajan GJ, Lawrence TJ, May WL. Density of GFAP-immunoreactive astrocytes is decreased in left hippocampi in major depressive disorder. Neuroscience (2016) 316 209-220. DOI: 10.1016/j.neuroscience.2015.12.044 54. González-Reyes RE, Nava-Mesa MO, Vargas-Sánchez K, Ariza-Salamanca D, Mora-Muñoz L. Involvement of astrocytes in alzheimer’s disease from a neuroinflammatory and oxidative stress perspective. Front Mol Neurosci (2017). DOI: 10.3389/fnmol.2017.00427 55. Ohira K, Funatsu N, Homma KJ, Sahara Y, Hayashi M, Kaneko T. Truncated TrkB-T1 regulates the morphology of neocortical layer I astrocytes in adult rat brain slices. Eur J Neurosci (2007) 25 406-416. DOI: 10.1111/j.1460-9568.2007.05282.x 56. Condorelli DF, Salin T, Dell’Albani P, Mudò G, Corsaro M, Timmusk T. Neurotrophins and theirtrk receptors in cultured cells of the glial lineage and in white matter of the central nervous system. J Mol Neurosci (1995) 6 237-248. DOI: 10.1007/BF02736783 57. Soontornniyomkij V, Wang G, Pittman CA, Hamilton RL, Wiley CA, Achim CL. Absence of brain-derived neurotrophic factor and trkB receptor immunoreactivity in glia of Alzheimer’s disease. Acta Neuropathol (1999) 98 345-348. DOI: 10.1007/s004010051092 58. Murphy S, Pearce B. Functional receptors for neurotransmitters on astroglial cells. Neuroscience (1987) 22 381-394. DOI: 10.1016/0306-4522(87)90342-3 59. Benediktsson AM, Schachtele SJ, Green SH, Dailey ME. Ballistic labeling and dynamic imaging of astrocytes in organotypic hippocampal slice cultures. J Neurosci Methods (2005) 141 41-53. DOI: 10.1016/j.jneumeth.2004.05.013 60. Zhao L, He LX, Huang SN, Gong LJ, Li L, Lv YY. Protection of dopamine neurons by vibration training and up-regulation of brain-derived neurotrophic factor in a MPTP mouse model of Parkinson’s disease. Physiol Res (2014). DOI: 10.33549/physiolres.932743 61. Ariizumi M, Okada A. Effects of whole body vibration on biogenic amines in rat brain. Occup Environ Med (1985) 42 133-136. DOI: 10.1136/oem.42.2.133 62. Krause DL, Müller N. Neuroinflammation, microglia and implications for anti-inflammatory treatment in Alzheimer’s disease. Int J Alzheimer Dis (2010) 2010 1-9. DOI: 10.4061/2010/732806 63. Heesterbeek M, van der Zee EA, van Heuvelen MJG. Feasibility of three novel forms of passive exercise in a multisensory environment in vulnerable institutionalized older adults with dementia. J Alzheimer Dis (2019) 70 681-690. DOI: 10.3233/JAD-190309 64. Oroszi T, van Heuvelen MJG, Nyakas C, van der Zee EA. Vibration detection: its function and recent advances in medical applications. F1000 Research (2020) 9 619. DOI: 10.12688/f1000research.22649.1
--- title: Factors contributing to exercise tolerance in patients with coronary artery disease undergoing percutaneous coronary intervention authors: - Husheng Li - Minqian Wei - Lili Zhang - Lan Huang - Yiyan Wang - Jiaqi Wang - Shaowei Zhuang - Xubo Wu - Jing Wu journal: BMC Sports Science, Medicine and Rehabilitation year: 2023 pmcid: PMC10026462 doi: 10.1186/s13102-023-00640-4 license: CC BY 4.0 --- # Factors contributing to exercise tolerance in patients with coronary artery disease undergoing percutaneous coronary intervention ## Abstract ### Background Exercise tolerance plays a vital role in the process of cardiac rehabilitation in patients undergoing percutaneous coronary intervention (PCI). The study sought to determine the characteristics, risks and correlates of post-PCI exercise tolerance in patients with coronary artery disease (CAD). ### Methods We analyzed clinical data of 299 CAD patients undergoing elective PCI and completing cardiopulmonary exercise testing (CPET). According to the Weber classification, post-PCI exercise tolerance was evaluated by peak oxygen uptake (VO2 peak). We assessed the impact of 34 predefined clinical features, cardiac functional parameters, and blood biochemistry data on exercise tolerance by univariate analysis and logistics regression analysis. ### Results Of 299 patients, $74.92\%$ were men and average age was 60.90 ± 10.68 years. VO2 peak in the entire population was 17.54 ± 3.38 ml/kg/min, and $24.41\%$ ($$n = 73$$) were less than 16 ml/kg/min, who were considered to have exercise intolerance. Multivariate logistics regression results showed that sex, diabetes mellitus, number of stents, left atrial diameter (LAD), end-diastolic volume (EDV), and hemoglobin influenced the peak oxygen uptake of CAD patients undergoing elective PCI. ( All $p \leq 0.05$). ### Conclusions Nearly one quarter of CAD patients have exercise intolerance in the early post-PCI period. Female, diabetes mellitus, number of stents, LAD, EDV might negatively impacted post-PCI exercise tolerance, which need further warrant by large scale cohort study. ### Supplementary Information The online version contains supplementary material available at 10.1186/s13102-023-00640-4. ## Introduction Coronary artery disease (CAD) is the leading cause of mortality and loss of disability worldwide [1]. Patients with CAD suffers significant symptoms of ischemia and hypoxia, due to the insufficient coronary flow and pressure that rise to meet the demands of physical activity. Percutaneous coronary intervention (PCI) is the priority treatment for CAD patients because it can rapidly unblock the infarct-related artery, restore myocardial perfusion, and reduce the infarct size [2, 3]. Even though, CAD patients undergoing PCI might experience symptomatic complications overtime, such as dyspnea, palpitations, dizziness on exertion. Evidence from epidemiological survey demonstrated that over half of post-PCI patients with the confession of exercise intolerance or muscular fatigue often get overlooked in clinical setting [4, 5]. Exercise tolerance [6, 7] referred to the maximum aerobic exercise capacity that can be tolerated without morbid symptoms and/or medical signs, and represents the body’s ability to absorb oxygen. Peak oxygen uptake (VO2 peak), measured during the cardiopulmonary exercise test (CPET), is the body’s maximum capacity to deliver and utilize oxygen, and is also the gold standard for assessment of exercise tolerance [8]. It can predict reinfarction and all-cause death in CAD patients and was used for prognosis assessment [9–11]. Previous studies [12–14] have reported that factors contributing to exercise tolerance include age, sex, body mass index (BMI), fasting blood glucose, ejection fraction, as well as nephropathy and peripheral arterial disease, etc. However, few existing researches have focused on the exercise tolerance status among PCI patients. We hypothesized that, in addition to demographic and disease factors, measures of cardiac function and blood biochemistry would predict exercise tolerance. The identification of predictors of exercise tolerance may help improve future study designs by revealing confounding variables, as well as providing a theoretical basis for future cardiac rehabilitation schemes for patients. Therefore, this study aimed to describe the current status of exercise tolerance as well as to identify its predictors in CAD patients undergoing PCI. ## Participants The study population consisted of CAD patients undergoing elective PCI for stable angina from January 2019 to December 2020. All subjects complete a maximal symptom limited, incremental CPET 1 month after PCI. Patients without contraindications were routinely treated with secondary prevention drugs such as dual antiplatelet agents, statins, angiotensin-converting enzyme inhibitors (ACEI)/angiotensin receptor blocker (ARB) and β-blockers postoperatively. Patients were excluded from this study if they met any of the following: [1] combination of other major systemic diseases, such as mid- to late-stage tumors, liver disease, renal disease and pulmonary impairment; [2] incomplete medical records; [3] presence of contraindications to CPET, including unstable angina, acute myocardial infarction 3–5 days, arrhythmias associated with unstable hemodynamic disturbances, active myocarditis or pericarditis, aortic stenosis, heart failure or pulmonary embolism, unstable lower extremity venous thrombosis, moderate to severe asthma, in the acute phase of infection, suffering from abnormal psychiatric symptoms or physical disability. ## Interventional procedure All patients provided explicit written informed consent prior to undergoing cardiac catheterization. Antiplatelet therapy was given to all patients before PCI with a specific regimen of Aspirin 100 mg once daily and Clopidogrel 75 mg once daily (or Tegretol 90 mg twice daily). Glycoprotein IIb/IIIa inhibitors were administered during the procedure and immediately after PCI, at the surgeon's discretion. The choice of coronary stent type and other adjuvant therapy is at the discretion of the primary surgeon, with a complete shift to drug-eluting stents in recent years. All stents were implanted at moderate to high deployment pressures (12–16 atm). Routine anticoagulation with low molecular heparin was continued postoperatively. ## Measurements Sociodemographic characteristics, medical and medication history, CPET parameters, echocardiographic parameters, and laboratory data were collected from participants’ medical records and interviews. The investigation conforms to the principles outlined in the Declaration of Helsinki [15]. The study was reviewed and approved by the Human Study Committee of Shanghai Seventh People’s Hospital (Registration No. 2021-7th-HIRB-012), and informed consent was formally obtained from each participant. ## Cardiopulmonary exercise testing Maximal symptom limited, incremental CPET was performed using cycle ergometers (Quark PFT Ergo, COSMED, Rome, Germany) with a ramp protocol [16, 17]. Before the test begins, the clinician will conduct a comprehensive evaluation and formulate an appropriate exercise increment plan for the patient. The exercise protocol started with a 3-min resting phase on the cycle ergometer, followed by a 3-min warm-up phase at 20-Watt initial workloads. Then, workload was set at 35 Watt followed by an increase of 10–30 Watt increments per min at pedaling speed > 60 rpm until the patient has exhaustion or restrictive symptoms or signs [18], e.g., reaching the submaximal heart rate; respiratory exchange rate (RER) ≥ 1.0; electrocardiogram ST segment changes, etc. The following recovery phase consisted of 2-min active recovery at 20 Watt at pedalling speed between 50 and 60 rpm, followed by 3-min passive recovery. During the whole test process, clinicians pay attention to monitoring the patient’s real-time ECG, blood pressure, gas exchange parameters, etc. The test was terminated when the subject showed one of the following conditions: [1] chest pain, dyspnea, pallor, weakness, dizziness, lower extremity pain, or unsteadiness in standing and requested to terminate; [2] ECG suggestive of myocardial ischemia; [3] II- or III-degree atrioventricular block; [4] systolic blood pressure decreased > 20 mmHg; [5] hypertension: systolic blood pressure > 250 mmHg; diastolic blood pressure > 120 mmHg; [6] rating of perceived exertion (RPE) up to Borg 19–20. CPET core indicators such as VO2 peak, oxygen uptake efficiency slope (OUES), ventilatory efficiency (VE/VCO2) slope, etc. were measured. According to the Weber classification [19], VO2 peak < 16 ml/kg/min was considered to have objective exercise intolerance. ## Echocardiographic examination A Vivid E9 Color Doppler Ultrasound System with a 3.4 MHz transducer (GE Ultrasound, Horten, Norway) was used to conduct standard transthoracic 2D echocardiography. Exploring the parasternal long-axis view of the left ventricle can measure the thickness of the interventricular septum (IVST), the end-diastolic diameter of the left ventricle (LVDd), the end-systolic diameter of the left ventricle (LVDs), and the left atrial diameter (LAD). In the apical four-chamber view and the two-chamber view, the left ventricular ejection fraction (LVEF) was calculated using the Simpson formula. Pulse-Doppler can detect the blood flow spectrum of the mitral valve in the apical four chambers, and measure the double peaks of the mitral valve during diastole. The images generated by the echocardiography are all gathered and kept in the instrument's hard disk by two qualified cardiology fellows. ## Laboratory testing All measurements were performed in a central laboratory. Blood hemoglobin and platelet were automatically assessed using high-volume hematology analyzer Siemens Advia 2120 (Siemens Healthcare Diagnostics, Deerfield, IL, USA). Homocysteine, serum creatinine (Scr), uric acid (UA), and lipid profile [blood total cholesterol (TC), triglyceride concentrations (TG), high-density lipoprotein cholesterol (HDL-C), and low-density lipoprotein cholesterol (LDL-C)] were automatically assessed on Roche Cobas 8000 (Roche Diagnostics International Ltd, Rotkreuz, Switzerland). ## Statistical analysis Data were analyzed using the Statistical Package for the Social Sciences ver. 24.0 (SPSS Inc., Chicago, IL, USA). One-sample K–S normality tests were performed on the measurement data. Normally distributed data were expressed as Mean ± SD and compared using the independent samples t-test; skewed data were expressed as median (interquartile range) and compared using the Mann–Whitney U test. The categorical variables were described as number of cases, and comparisons were performed by Chi-squared or Fisher analysis. The level of statistical significance was set at a p-value less than 0.05. One-dimensional linear regression was used to assess the correlation between VO2 peak/Kg and selected variables. Logistic regression was performed to investigate significant predictors to identify exercise intolerance. The independent variables relevant to the logistic regression model were selected from the univariate analysis of demographic and procedural characteristics and clinical indicators, based on a threshold p-value of 0.05. ## Results We finally recruited 299 consecutive post-PCI patients. Figure 1 presents the flow diagram for the recruitment and analysis. Demographic and procedural characteristics were summarized in Table 1. Of the 299 patients, $74.92\%$ were male and the average age was 60.90 ± 10.69 years. The most frequent comorbidity was hypertension ($88.63\%$), followed by dyslipidemia ($73.24\%$) and diabetes mellitus ($24.75\%$). The comparison between the two groups indicated that sex, BMI, number of stents, prevalence of dyslipidemia, and diabetes mellitus were statistically significant. No significant difference in drug categories and lesion vessel were observed between groups. Fig. 1CONSORT diagram of study recruitment. PCI percutaneous coronary intervention, LVEF Left ventricular ejection fraction, CPET Cardiopulmonary exercise testingTable 1Comparison of demographic and procedural characteristicsVariablesAll patients ($$n = 299$$)Exercise tolerance ($$n = 226$$)Exercise intolerance ($$n = 73$$)p-valueMean age (year)60.90 ± 10.6860.47 ± 10.4962.45 ± 11.240.217Male, n (%)224 (74.92)180 (79.65)44 (60.27)0.001BMI (kg/m2)25.68 ± 3.3925.33 ± 3.1226.73 ± 3.940.002Comorbidity, n (%) Hypertension265 (88.63)198 (87.61)67 (91.78)0.329 Dyslipidemia219 (73.24)159 (70.35)60 (82.19)0.047 Diabetes mellitus74 (24.75)46 (20.35)28 (38.36)0.002No. of stents, n (%) < 0.001 179 (26.42)65 (28.76)14 (19.18) 2160 (53.51)131 (57.96)29 (39.73) 342 (14.05)25 (11.06)17 (23.29) ≥ 418 (6.02)5 (2.21)13 (17.81)Lesion vessel, n (%) LAD191 (63.88)139 (61.50)52 (71.23)0.132 LCX107 (35.79)77 (34.07)30 (41.10)0.276 LM37 (12.37)27 (11.95)10 (13.70)0.693 RCA146 (48.83)110 (48.67)36 (49.32)0.924Medication, n (%) Aspirin289 (96.66)218 (96.46)71 (97.26)0.741 Clopidogrel251 (83.95)190 (84.07)61 (83.56)0.918 Statins277 (92.64)209 (92.48)68 (93.15)0.848 ACEI/ARB195 (65.22)146 (64.60)49 (67.12)0.694 Calcium antagonist99 (33.11)72 (31.86)27 (36.99)0.418 β-blockers237 (79.26)174 (76.99)63 (86.30)0.088 Nitrates238 (79.60)178 (78.76)60 (82.19)0.527LAD Left anterior descending artery, LCX Left circumflex branch artery, LM Left coronary artery main stem, RCA Right coronary artery, BMI Body mass index, ACEI Angiotensin-converting enzyme inhibitors, ARB Angiotensin receptor blocker Six patients were limited by musculoskeletal pain and failed to complete the CPET protocol and achieve a maximal effort. All remaining 299 subjects who completed CPET had no major cardiac events, ischemic ECG changes or sustained ventricular arrhythmias during the testing period. VO2 peak in the entire population was 17.54 ± 3.38 ml/kg/min, and $24.41\%$ ($$n = 73$$) were less than 16 ml/kg/min, considered to have objective exercise intolerance. The distribution of VO2 peak is shown in Fig. 2, and the remaining core indicators of CPET were listed in Additional file 1: Table S1. In contrast, patients in exercise intolerance group were more likely to have decreased peak heart rate, VO2 peak, METs, and OUES, whereas VE/VCO2 slope and HRR Max showed increasing trend. Fig. 2Distribution of peak oxygen consumption (VO2 peak/Kg) among entire cohort Table 2 shows the differences in clinical features. In terms of transthoracic echocardiographic variables, LAD and EDV were higher in exercise intolerance group. As for blood biochemistry data, hemoglobin was 138.10 ± 14.13 g/l in the overall population and higher in normal exercise tolerance group. ( All $p \leq 0.001$).Table 2Comparison of clinical indicatorsVariablesAll patients ($$n = 299$$)Exercise tolerance ($$n = 226$$)Exercise intolerance ($$n = 73$$)p-valueTransthoracic echocardiographyLAD (mm)34.46 ± 4.3133.79 ± 4.0536.55 ± 4.46 < 0.001LVDd (mm)45.12 ± 6.1144.96 ± 5.7145.60 ± 7.240.436LVDs (mm)29.73 ± 4.6829.63 ± 4.0730.04 ± 6.210.518IVST (mm)10.75 ± 3.3810.77 ± 3.8110.67 ± 1.340.821LVEF (%)63.77 ± 6.5263.70 ± 6.2763.98 ± 7.260.754EDV (ml)98.16 ± 26.5395.04 ± 23.14107.80 ± 33.35 < 0.001FS (%)34.94 ± 4.8435.07 ± 4.6334.54 ± 5.440.414Blood biochemistryHemoglobin (g/l)138.10 ± 14.13139.87 ± 13.77132.64 ± 13.94 < 0.001Platelet (× 109/l)209.07 ± 60.23206.27 ± 53.12217.81 ± 77.870.155Homocysteine (μmol/l)13.82 ± 4.4813.88 ± 4.1513.62 ± 5.410.666Scr (umol/l)65.11 ± 16.2265.34 ± 14.6964.41 ± 20.350.673UA (umol/l)351.42 ± 93.44356.51 ± 91.53335.67 ± 98.110.098TC (mmol/l)3.91 ± 1.063.95 ± 1.043.77 ± 1.130.192TG (mmol/l)1.92 ± 1.641.95 ± 1.761.84 ± 1.190.615HDL-C (mmol/l)1.07 ± 0.261.08 ± 0.261.03 ± 0.260.160LDL-C (mmol/l)2.31 ± 0.932.33 ± 0.922.25 ± 0.960.545LAD Left atrial diameter, LVDd Left ventricular diastolic diameter, LVDs Left ventricular systolic diameter, IVST Thickness of the interventricular septum, LVEF Left ventricular ejection fraction, EDV End-diastolic volume, FS Fractional shortening, Scr Serum creatinine, UA Uric acid, TC Total cholesterol, TG Triglyceride concentrations, HDL-C High-density lipoprotein cholesterol, LDL-C Low-density lipoprotein cholesterol The linear regression plots between BMI, LAD, EDV, hemoglobin and VO2 peak/Kg were shown in Fig. 3. VO2 peak/Kg was inversely related to BMI (r = − 0.174, $$p \leq 0.003$$), LAD (r = − 0.206, $p \leq 0.001$) and EDV (r = − 0.135, $$p \leq 0.019$$), whereas it was positively correlated with hemoglobin ($r = 0.268$, $p \leq 0.001$). Furthermore, logistic regression models were performed to identify exercise intolerance, as presented in Table 3. Although dyslipidemia showed a trend towards an adverse effect on exercise capacity, it did not reach statistical significance ($$p \leq 0.054$$). Increased BMI (OR = 1.128, $$p \leq 0.003$$) was significantly associated with exercise intolerance in the univariate regression analysis; however, this association did not persist in the multivariate regression. Stepwise multivariate logistic analyses revealed that the number of stents (OR = 4.078, $p \leq 0.001$), diabetes mellitus (OR = 2.138, $$p \leq 0.027$$), LAD (OR = 1.173, $p \leq 0.001$), and EDV (OR = 1.199, $$p \leq 0.003$$) were linked to a higher risk of exercise intolerance, while being male (OR = 0.328, $$p \leq 0.003$$) and having a higher hemoglobin content (OR = 0.705, $$p \leq 0.006$$) were protective factors against exercise intolerance. Fig. 3Regression plots between VO2 peak/Kg, BMI, LAD, EDV and hemoglobin. BMI Body mass index, LAD Left atrial diameter, EDV End-diastolic volumeTable 3Binary logistic regression analysis to identify exercise intoleranceVariablesUnivariate regression analysisMultivariate regression analysisOR$95\%$ CIp-valueOR$95\%$ CIp-valueSex < 0.0010.003 Male0.3880.219–0.6850.3280.156–0.692 FemaleReferenceReferenceNo. of stents < 0.001 < 0.001 > 24.5582.491–8.3404.0782.060–8.071 ≤ 2ReferenceReferenceDiabetes mellitus0.0020.027 Yes2.4351.374–4.3152.1381.091–4.189 NoReferenceReferenceDyslipidemia0.054 Yes1.9420.097–3.778 NoReferenceBMI (kg/m2)1.1281.043–1.2200.003Not entered–LAD (mm)1.1651.091–1.245 < 0.0011.1731.081–1.272 < 0.001EDV (per 10 ml)1.1881.072–1.3160.0011.1991.066–1.3490.003Hemoglobin (per 10 g/l)0.6910.567–0.842 < 0.0010.7050.549–0.9050.006Factors found significant in univariate regression analysis were included in a forward stepwise multivariate logistics regression model with included criteria of $p \leq 0.05$ and removal criteria of $p \leq 0.1.$ And Nagelkerke R2 for the multivariate model = 0.373OR Odds ratio, CI Confidence interval, Ref Reference, BMI Body mass index, LAD Left atrial diameter, EDV End-diastolic volume ## Discussion The VO2 peak is the most objective and reliable indicator to measure exercise tolerance in CAD patients [20, 21]. In our study, the average VO2 peak was found to be 17.54 ± 3.38 ml/kg/min, with nearly a quarter of the patients (73 cases, $24.41\%$) experiencing post-PCI cardiopulmonary dysfunction. These results suggest that even if coronary revascularization is successfully completed, exercise tolerance in the early postoperative period does not return swiftly to the normal level, which is in line with the findings of Li et al. [ 5]. Although PCI can improve symptoms of myocardial ischemia, myocardial contractile function and cardiac compliance remain abnormal in the early stages, resulting in reduced cardiac output and delayed recovery of exercise tolerance. Additionally, patients may opt not to exercise or exercise less due to concerns about wounds or myocardial infarction, or because they lack professional and scientific exercise planning guidance, all of which contributes to a lack of improvement in exercise tolerance in the early stages. Our study found that men and patients with higher hemoglobin levels had better post-PCI exercise tolerance. As noted by Kodama et al. [ 22], men have higher cardiorespiratory fitness values, about 2 METs higher, compared to women of the same age. This disparity can be attributed to differences in anatomy and physiology, such as: [1] Women having smaller left ventricles and lower ejection volumes[23]; [2] Lower left ventricular diastolic compliance in women [24]; [3] A higher proportion of obesity in women [23]; [4] Women being more prone to iron deficiency and having lower hemoglobin levels compared to men [25, 26]. Hemoglobin is a main marker of anemia and a primary performer of red blood cell function, responsible for transporting and carrying oxygen and carbon dioxide within the red blood cells. Decreased hemoglobin levels can result in a further deterioration of the hemodynamic state. In addition, aging in elderly patients leads to an increase in underlying diseases, weakens the body's immune function, and makes them more prone to recurrent infections, all of which exacerbates the decline in post-PCI exercise tolerance. Diabetes mellitus is a separate risk factor. According to Gürdal et al. [ 27], the VO2 peak and anaerobic threshold of diabetic patients were significantly lower than those in healthy adults. Plausibility of this mechanism is strengthened by several pathological pathways, such as microvascular disease, energy metabolism disorders, and autonomic dysfunction, which are independent of hypertension and coronary artery disease. These pathological changes result in ventricular diastolic dysfunction and impaired heart rate recovery, thereby impacting exercise tolerance [28–31]. With more stents implanted, CAD patients are at a higher risk of postoperative exercise intolerance, which is consistent with the findings of the SYNTAX trial [32]. The number of stents and the total length of the stents implanted are important indicators of the complexity of the coronary lesions and play a crucial role in predicting the clinical outcomes of patients undergoing PCI [33]. Studies have shown that an excessive number of stents can increase the damage to the endothelium during the procedure, exacerbating the local inflammatory response of the endothelium undergoing PCI [34, 35], which can trigger symptoms of exercise limitation. LAD and EDV have been proposed as a morphophysiological marker of ventricular dysfunction. This study found that increased LAD and EDV are also risk factors for postoperative exercise intolerance in CAD patients. Previous research has shown that when ventricular dysfunction occurs, diastolic filling pressure increases, ventricular compliance decreases, left atrial pressure increases, pulmonary vein and capillary wedge pressure increases, and pulmonary ventilation perfusion is impaired. These changes result in elevated VE/VCO2, insufficient left ventricular filling, reduced cardiac output, shortness of breath during exertion, and decreased exercise tolerance [4, 36–38]. Interestingly, LVEF, a commonly used indicator, does not seem to predict exercise tolerance. One reason is that post-PCI patients with LVEF less than $50\%$ are usually considered temporarily unfit for CPET and therefore do not have data for analysis in this study. On the other hand, the relationship between exercise capacity and LVEF may be impacted by diverse comorbidities, and various compensatory mechanisms can help preserve exercise ability [39]. Smart et al. [ 40] found that rest LVEF was weakly correlated with peak oxygen uptake, which was more closely related to a composite model filling pressure, systolic and diastolic function. Patients with CAD tend to be sedentary undergoing PCI [41], which may lead to cardiorespiratory deconditioning as well as muscle atrophy and weakness that in turn leads to deterioration in metabolic, cardiorespiratory, and functional health. However, increasing physical activity through comprehensive cardiac rehabilitation can improve exercise tolerance and quality of life in these patients [42–44]. Meta-analyses reported that endurance and resistance training together increased peak oxygen uptake and 6-min walk test distance by 2.2 ml/kg/min and 33 m, respectively [45, 46]. It has been noted that an increase in cardiorespiratory fitness is associated with a reduction in the risk of all-cause mortality and cardiovascular mortality [22]. Thus, it is imperative that medical staff provide early exercise guidance and health education to CAD patients undergoing PCI, and offer personalized exercise rehabilitation programs to enhance their post-PCI exercise tolerance. ## Limitations As an observational study, it has a limited number of included cases and may have problems such as selection bias. It is necessary to further expand the sample size for prospective study design. Enrollment was limited by the prescription of CPET, patients who could not tolerate CPET were not included in the study, and patients with more severe disease were excluded. None of the included patients discontinued β-blockers during CPET, which had a certain impact on the study results. Compared with the treadmill exercise program, the peak oxygen uptake for cycle ergometer exercise program was reduced by about 10–$20\%$, so the result of the CPET index was low. ## Conclusions Our main finding revealed that nearly a quarter (73 cases, $24.41\%$) of CAD patients have exercise intolerance in the early post-PCI period. Sex, the number of stents, diabetes mellitus, LAD, EDV and hemoglobin were identified as independent factors contributing to exercise intolerance. It is recommended to further explore a comprehensive cardiac rehabilitation model including exercise rehabilitation, symptom management and weight management, in order to improve the post-PCI exercise tolerance and relieve postoperative discomfort of CAD patients. ## Supplementary Information Additional file 1. Table S1: Comparison of CPET core indicators. ## References 1. Global, regional, and national age-sex specific all-cause and cause-specific mortality for 240 causes of death, 1990–2013: a systematic analysis for the Global Burden of Disease Study 2013. Lancet (2015) 385 117-171. DOI: 10.1016/S0140-6736(14)61682-2 2. Neumann F-J, 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 3. Cook CM, Ahmad Y, Howard JP, Shun-Shin MJ, Sethi A, Clesham GJ. Impact of percutaneous revascularization on exercise hemodynamics in patients with stable coronary disease. J Am Coll Cardiol (2018) 72 970-983. DOI: 10.1016/j.jacc.2018.06.033 4. Tashiro H, Tanaka A, Ishii H, Motomura N, Arai K, Adachi T. Reduced exercise capacity and clinical outcomes following acute myocardial infarction. Heart Vessels (2020) 35 1044-1050. DOI: 10.1007/s00380-020-01576-2 5. Li Y, Feng X, Chen B, Liu H. Retrospective analysis of exercise capacity in patients with coronary artery disease after percutaneous coronary intervention or coronary artery bypass graft. Int J Nurs Sci (2021) 8 257-263. PMID: 34307773 6. Ross R, Blair SN, Arena R, Church TS, Després J-P, Franklin BA. Importance of assessing cardiorespiratory fitness in clinical practice: a case for fitness as a clinical vital sign: a scientific statement from the American heart association. Circulation (2016) 134 e653-e699. DOI: 10.1161/CIR.0000000000000461 7. Valeur N. The prognostic value of pre-discharge exercise testing after myocardial infarction treated with either primary PCI or fibrinolysis: a DANAMI-2 sub-study. Eur Heart J (2004) 26 119-127. DOI: 10.1093/eurheartj/ehi057 8. Arnett DK, Blumenthal RS, Albert MA, Buroker AB, Goldberger ZD, Hahn EJ. ACC/AHA guideline on the primary prevention of cardiovascular disease: executive summary: a report of the american college of cardiology/American heart association task force on clinical practice guidelines. Circulation (2019) 2019 140 9. Forman DE, Arena R, Boxer R, Dolansky MA, Eng JJ, Fleg JL. Prioritizing functional capacity as a principal end point for therapies oriented to older adults with cardiovascular disease: a scientific statement for healthcare professionals from the American heart association. Circulation (2017) 135 e894-e918. DOI: 10.1161/CIR.0000000000000483 10. Hung RK, Al-Mallah MH, McEvoy JW, Whelton SP, Blumenthal RS, Nasir K. Prognostic value of exercise capacity in patients with coronary artery disease: the FIT (Henry ford exercise testing) project. Mayo Clin Proc (2014) 89 1644-1654. DOI: 10.1016/j.mayocp.2014.07.011 11. Cai H, Zheng Y, Liu Z, Zhang X, Li R, Shao W. Effect of pre-discharge cardiopulmonary fitness on outcomes in patients with ST-elevation myocardial infarction after percutaneous coronary intervention. BMC Cardiovasc Disord (2019) 19 210. DOI: 10.1186/s12872-019-1189-x 12. Fuller A, Okwose N, Scragg J, Eggett C, Luke P, Bandali A. The effect of age on mechanisms of exercise tolerance: Reduced arteriovenous oxygen difference causes lower oxygen consumption in older people. Exp Gerontol (2021) 149 111340. DOI: 10.1016/j.exger.2021.111340 13. Marchionni N, Fattirolli F, Fumagalli S, Oldridge NB, Del Lungo F, Bonechi F. Determinants of exercise tolerance after acute myocardial infarction in older persons. J Am Geriatr Soc (2000) 48 146-153. DOI: 10.1111/j.1532-5415.2000.tb03905.x 14. Haykowsky MJ, Tomczak CR, Scott JM, Paterson DI, Kitzman DW. Determinants of exercise intolerance in patients with heart failure and reduced or preserved ejection fraction. J Appl Physiol (2015) 119 739-744. DOI: 10.1152/japplphysiol.00049.2015 15. Ethical principles for medical research involving human subjects. JAMA (2013) 310 2191. DOI: 10.1001/jama.2013.281053 16. Fletcher GF, Ades PA, Kligfield P, Arena R, Balady GJ, Bittner VA. Exercise standards for testing and training: a scientific statement from the American heart association. Circulation (2013) 128 873-934. DOI: 10.1161/CIR.0b013e31829b5b44 17. Guiraud T, Juneau M, Nigam A, Gayda M, Meyer P, Mekary S. Optimization of high intensity interval exercise in coronary heart disease. Eur J Appl Physiol (2010) 108 733-740. DOI: 10.1007/s00421-009-1287-z 18. ATS/ACCP statement on cardiopulmonary exercise testing. Am J Respir Crit Care Med (2003) 2003 211-277 19. Guazzi M, Adams V, Conraads V, Halle M. Clinical recommendations for cardiopulmonary exercise testing data assessment in specific patient populations. Eur Heart J (2012) 33 2917-2927. DOI: 10.1093/eurheartj/ehs221 20. Al-Mallah MH, Sakr S, Al-Qunaibet A. Cardiorespiratory fitness and cardiovascular disease prevention: an update. Curr Atheroscler Rep (2018) 20 1. DOI: 10.1007/s11883-018-0711-4 21. Kubozono T, Itoh H, Oikawa K, Tajima A, Maeda T, Aizawa T. Peak VO. Circ J (2007) 72 575-581. DOI: 10.1253/circj.72.575 22. Kodama S. Cardiorespiratory fitness as a quantitative predictor of all-cause mortality and cardiovascular events in healthy men and women: a meta-analysis. JAMA (2009) 301 2024. DOI: 10.1001/jama.2009.681 23. Beale AL, Meyer P, Marwick TH, Lam CSP, Kaye DM. Sex differences in cardiovascular pathophysiology: why women are overrepresented in heart failure with preserved ejection fraction. Circulation (2018) 138 198-205. DOI: 10.1161/CIRCULATIONAHA.118.034271 24. Redfield MM, Jacobsen SJ, Borlaug BA, Rodeheffer RJ, Kass DA. Age- and gender-related ventricular-vascular stiffening: a community-based study. Circulation (2005) 112 2254-2262. DOI: 10.1161/CIRCULATIONAHA.105.541078 25. van der Wal HH, Grote Beverborg N, Dickstein K, Anker SD, Lang CC, Ng LL. Iron deficiency in worsening heart failure is associated with reduced estimated protein intake, fluid retention, inflammation, and antiplatelet use. Eur Heart J (2019) 40 3616-3625. DOI: 10.1093/eurheartj/ehz680 26. Martens P, Nijst P, Verbrugge FH, Smeets K, Dupont M, Mullens W. Impact of iron deficiency on exercise capacity and outcome in heart failure with reduced, mid-range and preserved ejection fraction. Acta Cardiol (2018) 73 115-123. DOI: 10.1080/00015385.2017.1351239 27. Gürdal A, Kasikcioglu E, Yakal S, Bugra Z. Impact of diabetes and diastolic dysfunction on exercise capacity in normotensive patients without coronary artery disease. Diab Vasc Dis Res (2015) 12 181-188. DOI: 10.1177/1479164114565631 28. Leenders M, Verdijk LB, van der Hoeven L, Adam JJ, van Kranenburg J, Nilwik R. Patients with type 2 diabetes show a greater decline in muscle mass, muscle strength, and functional capacity with aging. J Am Med Dir Assoc (2013) 14 585-592. DOI: 10.1016/j.jamda.2013.02.006 29. Fang ZY, Sharman J, Prins JB, Marwick TH. Determinants of exercise capacity in patients with type 2 diabetes. Diabetes Care (2005) 28 1643-1648. DOI: 10.2337/diacare.28.7.1643 30. Poirier P, Bogaty P, Garneau C, Marois L, Dumesnil J-G. Diastolic dysfunction in normotensive men with well-controlled Type 2 diabetes. Diabetes Care (2001) 24 5-10. DOI: 10.2337/diacare.24.1.5 31. Carnethon MR, Jacobs DR, Sidney S, Liu K. Influence of autonomic nervous system dysfunction on the development of type 2 diabetes. Diabetes Care (2003) 26 3035-3041. DOI: 10.2337/diacare.26.11.3035 32. Serruys PW, Morice M-C, Kappetein AP, Colombo A, Holmes DR, Mack MJ. Percutaneous coronary intervention versus coronary-artery bypass grafting for severe coronary artery disease. N Engl J Med (2009) 360 961-972. DOI: 10.1056/NEJMoa0804626 33. Généreux P, Giustino G, Redfors B, Palmerini T, Witzenbichler B, Weisz G. Impact of percutaneous coronary intervention extent, complexity and platelet reactivity on outcomes after drug-eluting stent implantation. Int J Cardiol (2018) 268 61-67. DOI: 10.1016/j.ijcard.2018.03.103 34. Koskinas KC, Chatzizisis YS, Antoniadis AP, Giannoglou GD. Role of endothelial shear stress in stent restenosis and thrombosis. J Am Coll Cardiol (2012) 59 1337-1349. DOI: 10.1016/j.jacc.2011.10.903 35. Ellis SG. Refining the art and science of coronary stenting. N Engl J Med (2009) 360 292-294. DOI: 10.1056/NEJMe0809422 36. Left GJ, Function V, Capacity E. Left ventricular function and exercise capacity. JAMA (2009) 301 286. DOI: 10.1001/jama.2008.1022 37. Edelmann F, Gelbrich G, Düngen H-D, Fröhling S, Wachter R, Stahrenberg R. Exercise training improves exercise capacity and diastolic function in patients with heart failure with preserved ejection fraction. J Am Coll Cardiol (2011) 58 1780-1791. DOI: 10.1016/j.jacc.2011.06.054 38. Hsiao C-S, Hsiao S-H. Relationship of left atrial expansion index to exercise tolerance, pretest probability of restenosis, and positive predictive value of treadmill test in coronary artery disease. Echocardiography (2020) 37 388-398. DOI: 10.1111/echo.14620 39. Tsougos E, Angelidis G, Gialafos E, Tzavara C, Tzifos V, Tsougos I. Myocardial strain may predict exercise tolerance in patients with reduced and mid-range ejection fraction. Hellenic J Cardiol (2018) 59 331-335. DOI: 10.1016/j.hjc.2017.11.016 40. Smart N, Haluska B, Leano R, Case C, Mottram PM, Marwick TH. Determinants of functional capacity in patients with chronic heart failure: role of filling pressure and systolic and diastolic function. Am Heart J (2005) 149 152-158. DOI: 10.1016/j.ahj.2004.06.017 41. Charman SJ, van Hees VT, Quinn L, Dunford JR, Bawamia B, Veerasamy M. The effect of percutaneous coronary intervention on habitual physical activity in older patients. BMC Cardiovasc Disord (2016) 16 248. DOI: 10.1186/s12872-016-0428-7 42. Swift DL, Lavie CJ, Johannsen NM, Arena R, Earnest CP, O’Keefe JH. Physical activity, cardiorespiratory fitness, and exercise training in primary and secondary coronary prevention. Circ J (2013) 77 281-292. DOI: 10.1253/circj.CJ-13-0007 43. Franklin BA, Lavie CJ, Squires RW, Milani RV. Exercise-based cardiac rehabilitation and improvements in cardiorespiratory fitness: implications regarding patient benefit. Mayo Clin Proc (2013) 88 431-437. DOI: 10.1016/j.mayocp.2013.03.009 44. Garber CE, Blissmer B, Deschenes MR, Franklin BA, Lamonte MJ, Lee I-M. Quantity and quality of exercise for developing and maintaining cardiorespiratory, musculoskeletal, and neuromotor fitness in apparently healthy adults: guidance for prescribing exercise. Med Sci Sports Exerc (2011) 43 1334-1359. DOI: 10.1249/MSS.0b013e318213fefb 45. Fukuta H, Goto T, Wakami K, Kamiya T, Ohte N. Effects of exercise training on cardiac function, exercise capacity, and quality of life in heart failure with preserved ejection fraction: a meta-analysis of randomized controlled trials. Heart Fail Rev (2019) 24 535-547. DOI: 10.1007/s10741-019-09774-5 46. Ismail H, McFarlane JR, Nojoumian AH, Dieberg G, Smart NA. Clinical outcomes and cardiovascular responses to different exercise training intensities in patients with heart failure. JACC: Heart Fail (2013) 1 514-522. PMID: 24622004
--- title: 'Adherence to oral anticancer hormonal therapy in breast cancer patients and its relationship with treatment satisfaction: an important insight from a developing country' authors: - Amer A. Koni - Bushra A. Suwan - Maisa A. Nazzal - Alaa Sleem - Aiman Daifallah - Majd Hamed allah - Razan Y. Odeh - Sa’ed H. Zyoud journal: BMC Women's Health year: 2023 pmcid: PMC10026465 doi: 10.1186/s12905-023-02276-5 license: CC BY 4.0 --- # Adherence to oral anticancer hormonal therapy in breast cancer patients and its relationship with treatment satisfaction: an important insight from a developing country ## Abstract ### Background Hormone-positive breast cancer is the most common type and represents a burden in all countries. Treatment satisfaction might be a predictor for adherence, as higher satisfaction with medication encourages patients to adhere appropriately to the medication and, consequently, successfully achieve the treatment goals. The present study evaluated the adherence of women with hormone-positive breast cancer to oral hormonal drugs and correlated it with treatment satisfaction and other sociodemographic and clinical factors. ### Methods A cross-sectional design was applied. This study included two cancer centers. Data were collected from patients through face-to-face interviews and medical record reviews. The Medication Adherence Scale was adapted to assess medication adherence, and the Treatment Satisfaction Questionnaire for Medication (TSQM) version 1.4 was adopted to measure treatment satisfaction. ### Results The final analysis included 106 patients, with a mean age ± SD of 51.9 ± 1.2. Approximately $35\%$ were hospitalized in the past year. Current hormonal therapy among cancer patients included letrozole ($38.7\%$), tamoxifen ($31.1\%$), exemestane ($17\%$), and anastrozole ($13.2\%$). The median adherence score was 5.0 [4.8–6.0], and $62.3\%$ adhered fully to their oral hormonal drugs in the past week. The median scores of effectiveness, side effects, convenience, and global satisfaction were 66.67 [61.11.0–72.22], 75.00 [48.44–100.00], 66.67 [66.67–72.22], and 71.43 [57.14–78.57], respectively. A significantly lower adherence score was identified in patients living in camps ($$p \leq 0.020$$). Patients with comorbidities and those who continued on the same hormonal therapy had higher adherence scores, although they were not statistically significant. Multiple linear regression analysis showed that two domains of treatment satisfaction, side effects ($$p \leq 0.013$$) and global satisfaction ($$p \leq 0.018$$), were predictors of adherence to oral hormonal drugs. ### Conclusions The current study revealed a significant association between treatment satisfaction and adherence to oral hormonal therapy. We recommend creating a specialized scale to measure adherence, considering the psychosocial factors that affect hormonal anticancer medication adherence. ## Background In modern times, breast cancer is one of the most common health conditions faced by women worldwide [1]. It represents approximately $24.5\%$ of all types of cancer in females and affects 1 in 8 women during their lifetime [1]. In Palestine, breast cancer is the most common cancer. In 2021, the number of new breast cancer cases in Palestine was 876 [2]. Based on statistics in the United States, Hormonal positive/human epidermal growth factor receptor 2 (HER2) negative breast cancer is the most common subtype and represents approximately $68\%$ [3]. Oral hormonal anticancer drugs (i.e., tamoxifen and aromatase inhibitors) are prescribed for women with estrogen-positive and progesterone-positive breast cancer, with a highly satisfactory result after using these treatments [4]. It is often started as an adjuvant treatment following surgery/radiotherapy/chemotherapy or a combination of these therapies and given for 5 to 10 years [5]. It can also be given as a neo-adjuvant [5]. In a systematic review, the adherence rate to adjuvant hormonal therapy was approximately $66\%$ [6]. Furthermore, it was found that more than half of breast cancer patients had nonadherent behavior to their treatment [7]. This percentage is close to that in Arabic nations, where Saudi Arabia reported a $69\%$ adherence rate to antihormonal therapy [8]. Depression, older age, comorbidities, younger age, and side effects were associated with lower adherence. However, therapy with aromatase inhibitors, received chemotherapy, and prior medication use were associated with improved adherence [6]. In addition, it was found that adherence to hormone therapy increases disease-free survival [9]. Nevertheless, adherence to endocrine treatment decreased with years of therapy [10]. A previous study proved that satisfaction with oral anticancer drugs substantially affects adherence [11]. Breast cancer is a notable burden in all countries, and its incidence is high [12]. Suppose a cancer patient follows the treatment plan and adheres to the medications directed by his physician. In that case, it improves the survival rate and decreases the likelihood of recurrence [13]. Patient satisfaction with treatment is the key that encourages him to adhere to medications and successfully achieve short- and long-term results [11]. However, treatment satisfaction assessment helps healthcare professionals know the exact level of their patient's satisfaction with a specific drug and subsequently modify the treatment plan or find other solutions. This study will be the foundation for other projects that aim to evaluate adherence and treatment satisfaction in different cancer populations or with other therapies. There are limited reports on endocrine therapy adherence and treatment satisfaction in Palestine. Therefore, this study aims to determine the adherence rate and study factors associated with adherence. ## Study design and sampling technique We conducted the current multicenter cross-sectional study to assess breast cancer patients' adherence and satisfaction with oral hormonal medications using two main sets of data: medical records (both on paper and electronic) and women breast cancer patients’ interviews. This research was conducted using convenience sampling between November 2021 and January 2022. All patients who came to the hospital for treatment or follow-up care and met the inclusion criteria were asked to complete the questionnaire. ## Study setting Our research was carried out in the oncology centers of the Al-Watani Hospital and the An-Najah National University Hospital in Nablus, Palestine. These hospitals are the largest and most important referral sites for cancer patients from all locations in Palestine. ## Sample size According to medical records, the number of women with breast cancer visiting the two hospitals during the study period was approximately 175. Therefore, the recommended sample size was 121 patients using an online calculation, Raosoft, with a response of $50\%$, a $5\%$ margin of error, and a $95\%$ confidence interval. ## Exclusion and inclusion criteria This study included women who had survived breast cancer over 18 years of age and had been prescribed and initiated oral hormonal drugs (neoadjuvant or adjuvant) at least four weeks prior to enrollment. Patients with comorbid delirium, dementia, bipolar, substance dependence disorders, untreated psychotic disorders, hospitalized patients, or those who were unable to participate or refused were excluded because of their inability to consent. We also excluded patients with missing findings in their medical records. ## Data collection instrument and procedure Two clinical pharmacists collected the data through face-to-face interviews with patients. Before beginning the data analysis, regular checks were performed for data integrity, proper sequences of information, and an evaluation of missing or incomplete variables. Questionnaires were completed by explaining the questions to the patients, filling in the information on papers using specific scales to assess cancer patients' adherence and satisfaction, and recording the patients’ sociodemographic information (Table 1). In addition, medical records were used to record information related to disease and treatment characteristics. Table 1Sociodemographic and clinical characteristics ($$n = 106$$)FactorFrequency (%)Age (years) < 4532 (30.2) 45–6557 (53.8) > 6517 (16.0)Body mass index < 18.52 (1.9) 18.5–24.925 (23.6) 25–29.937 (34.9) ≥ 3042 (39.6)Residency City53 (50.0) Village45 (42.5) Camp8 (7.5)Smoking No89 (84.0) Yes17 (16.0)University qualification No71 (67.0) Yes35 (33.0)Work Unemployed89 (84.0) Employed17 (16.0)Material status Single23 (21.7) Married83 (78.3)Family history No46 (43.4) Yes60 (56.6)Comorbidities No51 (48.1) Yes55 (51.9)Chronic medication No33 (31.1) Yes73 (68.9)Number of clinic visits per year < 1229 (27.4) ≥ 1277 (72.6)Hospitalized in the last year No69 (65.1) Yes37 (34.9)Pathology type Lobular8 (7.6) Ductal93 (87.7) None4 (3.8) Both1 (0.9)Breast surgery No20 (18.9) Yes86 (81.1)Radiotherapy No37 (34.9) Yes69 (65.1)Chemotherapy No21 (19.8) Yes85 (80.2)Biological therapy No86 (81.1) Yes20 (18.9)Targeted therapy No86 (81.1) Yes20 (18.9)Initial hormonal therapy Tamoxifen50 (47.2) Exemestane11 (10.4) Letrozole33 (31.1) Anastrozole12 (11.3)Current hormonal therapy Tamoxifen33 (31.1) Exemestane18 (17.0) Letrozole41 (38.7) Anastrozole14 (13.2)Hormonal drug switching No76 (71.7) Yes30 (28.3)Duration of starting current hormonal therapy < 1 year36 (34.0) ≥ 1 year70 (66.0)HER2 status Negative82 (77.4) Positive24 (22.6)Disease recurrence No95 (89.6) Yes11 (10.4)Menopausal status Premenopause56 (52.8) Postmenopause50 (47.2) ## Medication adherence The Medication Adherence Rating Scale (MARS) is used to assess adherence to medication [14]. It is a 10-item self-report instrument with yes/no responses to the questions asked, with a summation yielding a maximum of 10 points. MARS scores can range from 0 (low likelihood of adherence) to 10 (high likelihood of adherence). It also has three groups of items: "medication adherence behavior" (questions 1–4), "attitude toward taking medication" (questions 5–8) and "negative side effects and attitudes toward oral hormonal medication" (questions 9, 10). However, three theoretically irrelevant items (questions 5, 7, and 9) were removed due to poor item-total correlation. These excluded items are “I take my medication when I am sick”, “My thoughts are clearer on medication”, and “I feel weird, like a 'zombie', on medication”. Therefore, the maximum point became 7 (high likelihood of adherence). ## Treatment satisfaction The Treatment Satisfaction Questionnaire for Medication (TSQM) version 1.4 assesses patients' perceptions of treatment [15–18]. It evaluates effectiveness (items 1–3), side effects (items 4–8), convenience (items 9–11), and global satisfaction (items 12–14). The TSQM is a validated scale ranging from 0 to 100, with a higher score denoting better satisfaction [19]. The TSQM scale uses 14 questions to evaluate patient satisfaction; questions 1 through 3 inquire about the patient's satisfaction with the drug's efficacy in preventing and treating his disease, as well as the drug's capacity to relieve the patient's symptoms and the length of time it takes to begin working. Questions 4–8 inquire about the drug's adverse effects, the degree to which the patient finds them bothersome, how they affect his bodily and emotional well-being, and how much of an impact they have on the patient's satisfaction with the medication. The ninth and tenth questions concern the ease or difficulty of using the medication and scheduling a time for it to be used, while the eleventh question concerns whether it is proper to take the medication as directed. The confidence of the patient that this medication is helpful to him, that its benefits outweigh its drawbacks, and the degree of his general satisfaction with the medication are evaluated in questions 12 through 14. The Arabic version of the TSQM 1.4 is a valid and reliable instrument for assessing the perceptions of patients about treatment [19]. It has been used in several publications in Palestine [16, 17, 20–25]. In addition, IQVIA™ has given An-Najah National University permission to utilize this questionnaire in their research. ## Pilot study The pilot study sample consisted of 10 breast cancer patients chosen at the same criteria as the study population. The questionnaire was also completed in the same manner as it was for the study's population. Both scales, TSQM and MARS, were tested in the sample to evaluate the simplicity, understandability, and time to fill out all questions of the questionnaire. The Cronbach's alpha was 0.673 for the effectiveness domain of TSQM, 0.899 for side effects, 0.747 for convenience, and 0.878 for global satisfaction. ## Ethical approval The Institutional Review Boards (IRB) of An-Najah National University and the Palestinian Health Authority approved every aspect of the study protocol, including the use of and access to the patients' data. Furthermore, before initiating data collection, we properly explained all parts of the questionnaire to patients and received their verbal consent. ## Statistical analysis The Statistical Package for Social Sciences (IBM-SPSS) version 21 was used to enter and analyze the data. The results were explained using frequencies and percentages. The sociodemographic and clinical characteristics were described using descriptive and comparative statistics. We expressed the continuous variables using the median and interquartile ranges because the data were not normally distributed, as tested by the Kolmogorov–Smirnov test. Therefore, the Mann‒Whitney U and Kruskal‒Wallis tests were applied to examine the differences between variables. The Spearman test (TSQM and MARS scores) determined the association between treatment satisfaction and adherence. After that, all documented significant variables (sociodemographics and treatment satisfaction domains) in univariate analysis were entered in multiple linear regression analysis to determine the predictors for adherence. It was determined that there was a significant association with the outcome variables if the p value was less than 0.05. ## Sociodemographic and clinical characteristics Table 1 describes the sociodemographic and clinical characteristics of the 106 women with breast cancer. Of all 121 recruited patients, 15 refused to participate due to lack of time, privacy, and psychological problems. Approximately $53\%$ of the participants were aged between 45–65 years, $39.6\%$ were obese, $67\%$ had no university education, and $84\%$ were unemployed. According to clinical characteristics, most patients had comorbidities and took other chronic drugs ($51.9\%$ and $68.9\%$, respectively). Furthermore, $81.1\%$ of the patients underwent breast surgery, while $80.2\%$ received chemotherapy. The current hormonal therapy among cancer patients is as follows: letrozole $38.7\%$, tamoxifen $31.1\%$, exemestane $17\%$, and anastrozole $13.2\%$ (Table 1). ## Description of associations between patient characteristics and adherence score Among 106 women with breast cancer, the median adherence score was 5.0 [4.8–6.0] (range: 1.0–7.0). Approximately $62.3\%$ of the patients reported a high likelihood of adherence to oral hormonal drugs in the past week. Regarding the associations between patient characteristics and adherence score, a significantly lower adherence score was identified in patients living in camps ($$p \leq 0.020$$). Patients with comorbidities and those who continued on the same hormonal therapy had higher adherence scores, although they were not statistically significant. In this study, patients with comorbidities had a mean rank of 58.18, with a median of 5.0 [5.0–6.0], while patients without comorbidities had a mean rank of 48.45, with a median of 5.0 [4.0–6.0]. In terms of hormonal drug switching, the mean rank of patients who switched to another hormone therapy was 45.18, with a median of 5.0 [4.5–6.0], while the mean rank of patients who continued with the same hormonal prescription was 56.78, with a median of 5.0 [4.5–6.0] (Table 2).Table 2Associations between patient characteristics and adherence scoreFactorFrequency (%)Median [Q1-Q3]Mean rankP valueAge (years)0.277 < 4532 (30.2)5.0 [4.0–6.0]46. 67 45–6557 (53.8)5.0 [5.0–6.0]56.81 > 6517 (16.0)5.0 [5.0–6.0]55.26Body mass index0.949 < 18.52 (1.9)5.0 [5.0–6.0]64.5 18.5–24.925 (23.6)4.0 [5.0–6.0]52.02 25–29.937 (34.9)5.0 [5.0–6.0]53.73 ≥ 3042 (39.6)4.0 [5.0–6.0]53.65Residency0.020 City53 (50.0)5.0 [5.0–6.0]54.88 Village45 (42.5)5.0 [4.5–6.0]56.77 Camp8 (7.5)4.0 [3.25–4.75]26.00Smoking0.360 No89 (84.0)5.0 [5.0–6.0]59.41 Yes17 (16.0)5.0 [4.0–6.0]52.37University qualification0.619 No71 (67.0)5.0 [4.0–6.0]52.51 Yes35 (33.0)5.0 [5.0–6.0]55.50Work0.809 Unemployed89 (84.0)5.0 [4.5–6.0]53.20 Employed17 (16.0)5.0 [4.5–6.0]55.06Material status0.468 Single23 (21.7)5.0 [5.0–6.0]57.39 Married83 (78.3)5.0 [4.0–6.0]52.42Family history0.399 No46 (43.4)5.0 [4.0–6.0]50.78 Yes60 (56.6)5.0 [5.0–6.0]55.58Comorbidities0.085 No51 (48.1)5.0 [4.0–6.0]48.45 Yes55 (51.9)5.0 [5.0–6.0]58.18Chronic medication0.060 No33 (31.1)5.0 [4.0–6.0]45.61 Yes73 (68.9)5.0 [5.0–6.0]57.07Number of clinic visits per year0.264 < 1229 (27.4)5.0 [4.0–6.0]48.36 ≥ 1277 (72.6)5.0 [5.0–6.0]55.44Hospitalized in the last year0.972 No69 (65.1)5.0 [4.0–6.0]53.43 Yes37 (34.9)5.0 [5.0–6.0]53.64Breast surgery0.986 No20 (18.9)5.0 [4.25–6.0]53.60 Yes86 (81.1)5.0 [4.75–6.0]53.48Radiotherapy0.894 No37 (34.9)5.0 [4.5–6.0]52.99 Yes69 (65.1)5.0 [4.5–6.0]53.78Chemotherapy0.246 No21 (18.9)6.0 [5.0–6.0]60.10 Yes85 (80.2)5.0 [4.0–6.0]51.87Biological therapy0.915 No86 (81.1)5.0 [4.0–6.0]53.35 Yes20 (19.8)5.0 [5.0–6.0]54.13Targeted therapy0.676 No86 (81.1)5.0 [5.0–6.0]54.07 Yes20 (18.9)5.0 [4.0–6.0]51.05Current hormonal therapy0.825 Tamoxifen33 (31.1)5.0 [4.0–6.0]50.08 Exemestane18 (17.0)5.0 [4.0–6.0]54.86 Letrozole41 (38.7)5.0 [5.0–6.0]56.22 Anastrozole14 (13.2)5.0 [4.0–6.0]51.86Hormonal drug switching0.064 No76 (71.7)5.0 [4.5–6.0]56.78 Yes30 (28.3)5.0 [4.5–6.0]45.18Duration of starting current hormonal therapy0.303 < 1 year36 (34.0)5.0 [4.0–6.0]49.44 ≥ 1 year70 (66.0)5.0 [5.0–6.0]55.59HER2 status0.690 Negative82 (77.4)5.0 [4.75–6.0]54.11 Positive24 (22.6)5.0 [4.25–6.0]51.42Disease recurrence0.861 No95 (89.6)5.0 [5.0–6.0]53.33 Yes11 (10.4)5.0 [4.0–6.0]54.95Menopausal status0.478 Premenopause56 (52.8)5.0 [4.0–6.0]51.61 Postmenopause50 (47.2)5.0 [5.0–6.0]55.62 ## Description of the association between treatment satisfaction and adherence As shown in Table 3, there were significant correlations between MARS score and treatment satisfaction, including side effects ($$p \leq 0.024$$) and global satisfaction ($$p \leq 0.008$$). Women with a high adherence rate had higher satisfaction scores than women with a low adherence rate. Spearman’s rank order correlation coefficient between MARS adherence score and side effects and global satisfaction TSQM scores indicated significant positive correlations ($r = 0.220$ and 0.258, respectively).Table 3Spearman’s correlations between treatment satisfaction and adherenceEffectivenessSide effectsConvenienceGlobal satisfactionMARS ScoreCorrelation Coefficient0.1280.2200.1640.258P value0.1920.0240.0920.008The correlation is significant at the 0.01 level (2-tailed)The correlation is significant at the 0.05 level (2-tailed) ## Description of associations between patient characteristics and treatment satisfaction As shown in Table 4, the TSQM score assesses perceived effectiveness, side effects, convenience, and global satisfaction. The median score of each domain was 66.67 [61.11.0–72.22], 75.00 [48.44–100.00], 66.67 [66.67–72.22], and 71.43 [57.14–78.57], respectively. Postmenopausal patients had significantly higher satisfaction towards side effects ($$p \leq 0.049$$). In addition, patients with comorbidities had a higher global satisfaction score ($$p \leq 0.010$$). Furthermore, the satisfaction score toward side effects was significantly lower in patients with experienced side effects ($$p \leq 0.001$$) and those hospitalized in the last year ($$p \leq 0.030$$). Moreover, letrozole therapy was significantly associated with higher satisfaction with perceived effectiveness ($$p \leq 0.002$$) and global satisfaction ($$p \leq 0.004$$).Table 4Associations between patient characteristics and treatment satisfactionFactorEffectiveness Median [Q1-Q3]P valueSide effects Median [Q1-Q3]P valueConvenience Median [Q1-Q3]P valueGlobal satisfaction Median [Q1-Q3]P valueAge (years)0.3010.1850.2300.140 < 4566.67 [55.56–72.22]71.88 [43.75–87.50]66.67 [61.11–75.00]71.43 [42.86–71.43] 45–6566.67 [61.11–77.78]81.25 [43.75–100.00]66.67 [66.67–72.22]71.43 [64.29–78.57] > 6566.67 [61.11–76.39]90.63 [68.75–100.00]66.67 [66.67–76.39]71.43 [57.14–76.79]Body mass index < 18.566.6756.2575.0085.72 18.5–24.966.67 [61.11–75.00]87.5 [56.25–100.00]66.67 [63.89–69.45]71.43 [67.86–75.00] 25–29.966.67 [61.11–72.22]81.25 [53.13–100.00]66.67 [66.67–75.00]71.43 [57.14–71.43] ≥ 3066.67 [61.11–75.00]75.00 [37.50–100.00]66.67 [66.67–72.22]71.43 [53.57–78.57]Residency0.4150.7660.8960.559 City66.67 [61.11–72.22]84.38 [45.31–100.00]66.67 [66.67–77.78]71.43 [57.14–71.43] Village66.67 [61.11–77.78]75.00 [43.75–100.00]66.67 [66.67–72.22]71.43 [57.14–78.57] Camp66.67 [61.11–70.83]68.75 [64.06–98.44]66.67 [66.67–70.83]64.29 [51.79–71.43]Smoking0.3500.9820.3250.922 No66.67 [61.11–72.22]78.13 [50.00–100.00]66.67 [66.67–72.22]71.43 [57.14–76.79] Yes66.67 [61.11–83.33]75.00 [40.63–100.00]66.67 [66.67–77.78]71.43 [57.14–78.57]University qualification0.6460.2250.3850.699 No66.67 [61.11–72.22]81.25 [56.25–100.00]66.67 [66.67–72.22]71.43 [57.14–78.57] Yes66.67 [55.56–73.61]71.88 [35.94–100.0066.67 [66.67–77.78]71.43 [57.14–71.43]Work0.6260.4140.7510.497 Unemployed66.67 [61.11–72.22]75.00 [45.31–100.00]66.67 [66.67–72.22]71.43 [57.14–71.43] Employed66.67 [55.56–83.33]87.50 [46.88–100.00]66.67 [61.11–77.78]71.43 [60.72–78.57]Material status0.6010.1360.9410.058 Single66.67 [61.11–77.78]90.63 [67.19–100.00]66.67 [61.11–73.61]71.43 [69.65–78.57] Married66.67 [61.11–72.22]75.00 [43.75–100.00]66.67 [66.67–72.22]71.43 [57.14–71.43]Family history0.1970.0960.3240.451 No66.67 [61.11–77.78]75.00 [43.75–93.75]66.67 [61.11–69.45]71.43 [57.14–78.57] Yes66.67 [61.11–72.22]87.50 [50.00–100.00]66.67 [66.67–72.22]71.43 [57.14–71.43]Comorbidities0.5430.7500.6850.010 No66.67 [61.11–72.22]75.00 [43.75–100.00]66.67 [66.67–77.78]71.43 [57.14–71.43] Yes66.67 [61.11–77.78]78.13 [48.44–100.00]66.67 [66.67–72.22]71.43 [64.29–78.57]Chronic medication0.9110.8620.7990.618 No66.67 [61.11–72.22]75.00 [43.75–100.00]66.67 [66.67–77.78]71.43 [57.14–71.43] Yes66.67 [61.11–77.78]78.13 [48.44–100.00]66.67 [66.67–72.22]71.43 [64.29–78.57]Side effects (currently)0.5230.0010.7660.153 No66.67 [61.11–76.39]100.00 [57.81–100.00]66.67 [66.67–72.22]71.43 [57.14–71.43] Yes66.67 [61.11–72.22]68.75 [40.63–87.50]66.67 [63.89–72.22]71.43 [64.29–78.57]Number of clinic visits per year < 1266.67 [61.11–76.39]75.00 [57.81–98.44]66.67 [66.67–77.78]71.43 [64.29–78.57] = > 1266.67 [61.11–72.22]81.25 [43.75–100.00]66.67 [66.67–72.22]71.43 [57.14–75.00]Hospitalized in the last year0.3290.0300.9720.154 No66.67 [61.11–72.22]87.50 [50.00–100.00]66.67 [66.67–72.22]71.43 [57.14–71.43] Yes66.67 [61.11–77.78]65.63 [32.81–93.75]66.67 [61.11–76.39]71.43 [66.08–78.57]Breast surgery0.5960.2490.2390.782 No66.67 [61.11–72.22]90.63 [56.25–100.00]66.67 [66.67–76.39]71.43 [58.93–78.57] Yes66.67 [61.11–77.78]75.00 [43.75–100.00]66.67 [66.67–72.22]71.43 [57.14–75.00]Radiotherapy0.7460.5770.6790.249 No66.67 [61.11–77.78]75.00 [37.50–100.00]66.67 [66.67–72.22]71.43 [57.14–78.57] Yes66.67 [61.11–72.22]81.25 [56.25–100.00]66.67 [66.67–72.22]71.43 [57.14–71.43]Chemotherapy0.2620.6360.1830.607 No66.67 [56.95–70.83]78.13 [56.25–100.00]66.67 [66.67–76.39]71.43 [51.79–71.43] Yes66.67 [61.11–75.00]75.00 [43.75–100.00]66.67 [66.67–72.22]71.43 [57.14–78.57]Biological therapy0.2220.6510.6480.348 No66.67 [61.11–77.78]75.00 [50.00–100.00]66.67 [66.67–76.39]71.43 [57.14–78.57] Yes66.67 [51.39–70.83]71.88 [39.06–100.00]66.67 [66.67–76.39]71.43 [51.79–71.43]Targeted therapy0.6930.6930.8630.453 No66.67 [61.11–72.22]81.25 [50.00–100.00]66.67 [66.67–72.22]71.43 [57.14–71.43] Yes66.67 [61.11–77.78]65.63 [39.06–100.00]66.67 [66.67–76.39]71.43 [57.14–78.57]Current hormonal therapy0.0020.0510.3560.004 Tamoxifen66.67 [61.11–72.22]75.00 [43.75–87.50]66.67 [61.11–69.45]71.43 [42.86–71.43] Exemestane63.89 [54.17–68.06]59.38 [29.69–100.00]66.67 [66.67–77.78]71.43 [55.36–78.57] Letrozole72.22 [66.67–77.78]87.50 [56.25–100.00]66.67 [66.67–72.22]71.43 [71.43–78.57] Anastrozole61.11 [50.00–66.67]100.00 [82.81–100.00]66.67 [66.67–73.61]57.14 [50.00–71.43]Hormonal drug switching0.9660.2880.2440.867 No66.67 [61.11–72.22]81.25 [45.31–100.00]66.67 [66.67–77.78]71.43 [57.14–78.57] Yes66.67 [61.11–75.00]68.75 [46.88–100.00]66.67 [66.67–66.67]71.43 [60.72–71.43]Duration of starting current hormonal therapy0.5410.6200.5330.755 < 1 year66.67 [61.11–72.22]75.00 [43.75–100.00]66.67 [66.67–72.22]71.43 [57.14–78.57] ≥ 1 year66.67 [61.11–77.78]78.13 [48.44–100.00]66.67 [66.67–73.61]71.43 [62.50–73.22]HER2 status Negative66.67 [61.11–72.22]75.00 [50.00–100.00]66.67 [66.67–72.22]71.43 [57.14–78.57] Positive66.67 [61.11–76.39]75.00 [43.75–100.00]66.67 [66.67–72.22]71.43 [51.79–76.79]Disease recurrence0.4600.4160.7990.957 No66.67 [61.11–72.22]75.00 [43.75–100.00]66.67 [66.67–72.22]71.43 [57.14–78.57] Yes66.67 [61.11–77.78]81.25 [62.50–100.00]66.67 [61.11–83.33]71.43 [64.29–71.43]Menopausal status0.6100.0490.3090.159 Premenopause66.67 [61.11–72.22]75.00 [39.06–100.00]66.67 [66.67–72.22]71.43 [57.14–71.43] Postmenopause66.67 [61.11–77.78]87.50 [62.50–100.00]66.67 [66.67–72.22]71.43 [57.14–78.57] ## Multivariate analysis of adherence score From the univariate analysis, residency, side effects, and global satisfaction were found to be statistically significant ($p \leq 0.05$). Multiple linear regression analysis revealed that the side effects domain ($$p \leq 0.013$$) and global satisfaction ($$p \leq 0.018$$) were predictors of oral hormonal drug adherence (Table 5).Table 5Multivariate linear regression analysis of the adherence scoreModelUnstandardized CoefficientsUnstandardized CoefficientstSig$95.0\%$ Confidence Interval for B$95.0\%$ Confidence Interval for BBStd. ErrorBetaLower BoundLower BoundVIF1(Constant)3.8000.5546.8630.0002.7024.899Residency-0.2590.153-0.154-1.6910.094-0.5620.0451.006Side effects0.0080.0030.2342.5240.0130.0020.0151.033Global satisfaction0.0160.0070.2232.4060.0180.0030.0301.036aDependent variable: Adherence Score ## Discussion The current study examined the degree of adherence of Palestinian women with breast cancer to their oral hormonal therapy and described its correlation with treatment satisfaction and other variables. The sample of our study represents the age of the breast cancer population, in which approximately half of the breast cancer cases in Palestine fall within the 45–65 age group [26]. Oral hormonal therapy has improved patients' overall survival in breast cancer and long-term outcomes. An important element of treatment success is adherence to the medication. In the current study, $62.3\%$ adhered fully in the past week, with a median adherence score of 5.0 [4.8–6.0]. In general, the adherence rate to oral hormonal drugs ranged from 45 to $95.7\%$ [27]. In a systematic review, the mean rate of adherence at five years for the implementation phase was $66.2\%$, and the mean persistence was $66.8\%$ [6]. Our results showed that women living in refugee camps were less adherent than those who resided in cities or villages. This could be due to low residential stability and social affluence. Patients with comorbidities had a higher adherence score, similar to a previous study [28]. This may be explained by the fact that patients with multiple comorbidities are aware of their diseases and the consequences of being nonadherent to medications. In addition, patients with other conditions may also use co-medication for these indications, which might stimulate them to take antihormonal therapy since they have a ‘cocktail’ to take and follow a medication scheme. Importantly, women who switched from their hormone drugs to another experienced less adherence to the new medication. Similar findings were reported in previous studies [29–31]. However, this finding should be further highlighted to identify the causes of switching and its effect on adherence. Our study found no significant differences in adherence scores between the hormonal drugs used. Similarly, a study did not show a significant association between the adherence of patients using tamoxifen and those receiving aromatase inhibitors [32]. However, our results contradict those of previous studies related to educational level, radiation therapy, age, and hospitalization, all of which were found to be significantly associated with adherence to hormonal therapy [33, 34]. Concerning treatment satisfaction, we found that Palestinian patients had different scores in the four domains of treatment satisfaction, with lower scores in effectiveness and convenience. Patients on oral hormone therapy may not objectively feel an improvement in their health. Furthermore, the long duration of this therapy (5–10 years) may impact treatment satisfaction. However, the treatment satisfaction domain score of side effects was significantly lower in patients with experienced side effects or hospitalization in the past year. It was evident that side effects substantially decreased the patient’s satisfaction with treatment. In this study, treatment satisfaction (side effect and global satisfaction domains) was a predictor of adherence to oral hormonal drugs. This finding means that a high adherence score is associated with low experienced side effects and high global satisfaction rates. A previous study found that greater satisfaction with treatment led to more adherence to oral cancer drugs, including hormonal medications [11]. However, another study revealed no obvious correlation between adherence and patient satisfaction with medication information. The domain of side effects represented an essential impact on treatment satisfaction and adherence. Adverse effects from hormonal therapy were considered the main barrier to nonadherence [28, 34–36], and it negatively impacts the quality of life [32]. In our study, the highest beta coefficient was for the variable side effects. This suggests that side effects contributed the most to explaining differences in hormonal drug adherence. Our result is close to other studies, which indicated a considerably high percentage of nonadherence [32, 33, 37]. Clinicians should pay great attention to this issue, as nonadherence is correlated with all-cause mortality in Asian women with breast cancer [38]. For example, physician‒patient and pharmacist-patient communications should be enhanced [39] or an app-based new technique, such as a smartphone intervention [40] or using bubble packaging [41], should be adopted. ## Strengths and limitations This is the first study to correlate adherence and treatment satisfaction in patients with breast cancer treated with oral hormonal drugs and to analyze twenty-five sociodemographic and clinical factors. However, a cross-sectional design, a small sample size, the inclusion of only two centers, using self-report questionnaires, and convenient selections are considered limitations of the current study, affecting our findings' generalizability. Additionally, certain factors, such as receiving counseling from an oncologist/clinical pharmacist about medications, time since onset treatment, and stages of the disease, were not analyzed, as these variables may have a notable impact on adherence. Furthermore, the TSQM scale was not validated in the Palestinian population. Finally, MARS was developed for a psychiatric population and was not validated in a cancer population. Although the MARS was set for psychiatric patients [14], it had convergent validity, biologically measured adherence, good internal consistency, and test–retest reliability. It has also been used in a previous study among cancer patients receiving oral anticancer agents [42]. Importantly, the adherence scale used in the current study was adapted by removing three irrelevant items from the original MARS scale. ## Conclusions The current study found that higher treatment satisfaction, especially with regard to side effects, was strongly associated with good adherence to oral hormonal therapy. Adjuvant hormone therapy seems to be an exceptional situation for medication adherence because the relationship between psychosocial factors and adherence to hormonal therapy in breast cancer differs from the relationship in other chronic conditions [43]. Therefore, we recommend creating a specialized scale to measure adherence, considering the psychosocial factors that affect hormonal anticancer medication adherence. In addition, pharmacists should counsel cancer patients about hormonal therapy, addressing the reasons for nonadherence and handling them. Finally, awareness of healthcare professionals regarding oral hormonal drug adherence is the cornerstone to openly discussing risks for nonadherence with cancer patients. ## References 1. 1.Global Cancer Observatory. CANCER TOMORROW: A tool that predicts the future cancer incidence and mortality burden worldwide from the current estimates in 2020 up until 2040. 2020. https://gco.iarc.fr/tomorrow/en. Accessed 5 Nov 2022. 2. 2.Palestinian Ministry of Health. Health Annual Report Palestine 2021. 2022. https://site.moh.ps/Content/Books/Hqgu4D5vfT6bDhDUtl36GHhx9oYlCS9JplXYDfOMKrnDt6YoDPkPdl_I6mhnD3xb5MaPpX1mx6k6J4WowTnGUc1135KRHMmuMwEi1Zh1QUmFY.pdf. Accessed 5 Nov 2022. 3. 3.National Cancer Institute. Cancer Stat Facts: Female Breast Cancer Subtypes. 2022. https://seer.cancer.gov/statfacts/html/breast-subtypes.html#:~:text=The%20breast%20cancer%20subtype%20HR,based%20on%202015%E2%80%932019%20cases. Accessed 28 April 2022. 4. Trayes KP, Cokenakes SEH. Breast cancer treatment. Am Fam Physician (2021.0) 104 171-178. PMID: 34383430 5. Gradishar WJ, Moran MS, Abraham J, Aft R, Agnese D, Allison KH, Anderson B, Burstein HJ, Chew H, Dang C. Breast cancer, version 3.2022, NCCN clinical practice guidelines in oncology. J Natl Compr Canc Netw (2022.0) 20 691-722. DOI: 10.6004/jnccn.2022.0030 6. Yussof I, Mohd Tahir NA, Hatah E, Mohamed Shah N. Factors influencing five-year adherence to adjuvant endocrine therapy in breast cancer patients: a systematic review. Breast (2022.0) 62 22-35. DOI: 10.1016/j.breast.2022.01.012 7. Atkins L, Fallowfield L. Intentional and non-intentional non-adherence to medication amongst breast cancer patients. Eur J Cancer (2006.0) 42 2271-2276. DOI: 10.1016/j.ejca.2006.03.004 8. Khobrani A, Alatawi Y, Bajnaid E, Alemam O, Osman A, Attash LB, Jaffal M, AlGhanmi M, Alharbi A, Alnuhait M. Adherence to hormonal therapy in breast cancer patients in Saudi Arabia: a single-center study. Cureus (2022.0) 14 24780 9. Mohamed KEH, Elamin A. Adherence to endocrine therapy and its relation to disease-free survival among breast cancer patients visiting an out-patient clinic at Khartoum Oncology Hospital. Sudan J Eval Clin Pract (2020.0) 26 1731-1743. DOI: 10.1111/jep.13373 10. Camejo-Martínez N, Castillo-Leska C, Artagaveytia-Cóppola NA, Hernández-Ibero AL, Amarillo-Hernández DL, Apud-Yumati PS, Melo-Ramundo F, Delgado-Pebe LB. Adherence to adjuvant hormonal therapy in patients with breast cancer. Rev Med Inst Mex Seguro Soc (2021.0) 57 357-363 11. Jacobs JM, Pensak NA, Sporn NJ, MacDonald JJ, Lennes IT, Safren SA, Pirl WF, Temel JS, Greer JA. Treatment satisfaction and adherence to oral chemotherapy in patients with cancer. J Oncol Pract (2017.0) 13 e474-e485. DOI: 10.1200/JOP.2016.019729 12. Tao Z, Shi A, Lu C, Song T, Zhang Z, Zhao J. Breast cancer: epidemiology and etiology. Cell Biochem Biophys (2015.0) 72 333-338. DOI: 10.1007/s12013-014-0459-6 13. Inotai A, Agh T, Maris R, Erdosi D, Kovacs S, Kalo Z, Senkus E. Systematic review of real-world studies evaluating the impact of medication non-adherence to endocrine therapies on hard clinical endpoints in patients with non-metastatic breast cancer. Cancer Treat Rev (2021.0) 100 102264. DOI: 10.1016/j.ctrv.2021.102264 14. Thompson K, Kulkarni J, Sergejew AA. Reliability and validity of a new Medication Adherence Rating Scale (MARS) for the psychoses. Schizophr Res (2000.0) 42 241-247. DOI: 10.1016/S0920-9964(99)00130-9 15. Atkinson MJ, Kumar R, Cappelleri JC, Hass SL. Hierarchical construct validity of the treatment satisfaction questionnaire for medication (TSQM version II) among outpatient pharmacy consumers. Value Health (2005.0) 8 S9-s24. DOI: 10.1111/j.1524-4733.2005.00066.x 16. Zyoud SH, Al-Jabi SW, Sweileh WM, Arandi DA, Dabeek SA, Esawi HH, Atyeh RH, Abu-Ali HA, Sleet YI, Abd-Alfatah BM. Relationship of treatment satisfaction to health-related quality of life among Palestinian patients with type 2 diabetes mellitus: Findings from a cross-sectional study. J Clin Transl Endocrinol (2015.0) 2 66-71. PMID: 29159112 17. Zyoud SH, Al-Jabi SW, Sweileh WM, Morisky DE. Relationship of treatment satisfaction to medication adherence: findings from a cross-sectional survey among hypertensive patients in Palestine. Health Qual Life Outcomes (2013.0) 11 191. DOI: 10.1186/1477-7525-11-191 18. Atkinson MJ, Sinha A, Hass SL, Colman SS, Kumar RN, Brod M, Rowland CR. Validation of a general measure of treatment satisfaction, the Treatment Satisfaction Questionnaire for Medication (TSQM), using a national panel study of chronic disease. Health Qual Life Outcomes (2004.0) 2 12. DOI: 10.1186/1477-7525-2-12 19. Shilbayeh SAR, Alyahya SA, Alshammari NH, Almutairi WA, Shaheen E. Treatment Satisfaction Questionnaire for Medication: Validation of the Translated Arabic Version among Patients Undergoing Warfarin Therapy in Saudi Arabia. Value Health Reg Issues (2018.0) 16 14-21. DOI: 10.1016/j.vhri.2018.01.007 20. Al-Jabi SW, Zyoud SH, Sweileh WM, Wildali AH, Saleem HM, Aysa HA, Badwan MA, Awang R. Relationship of treatment satisfaction to health-related quality of life: findings from a cross-sectional survey among hypertensive patients in Palestine. Health Expect (2015.0) 18 3336-3348. DOI: 10.1111/hex.12324 21. Abu Hamdeh H, Al-Jabi SW, Koni A, Zyoud SH. Health-related quality of life and treatment satisfaction in Palestinians with rheumatoid arthritis: a cross-sectional study. BMC Rheumatol (2022.0) 6 19. DOI: 10.1186/s41927-022-00251-5 22. Khdour MR, Awadallah HB, Al-Hamed DH. Treatment satisfaction and quality of life among type 2 diabetes patients: a cross-sectional study in west bank. Palestine J Diabetes Res (2020.0) 2020 1834534. PMID: 32908934 23. Sweileh WM, Ihbesheh MS, Jarar IS, Sawalha AF, Abu Taha AS, Zyoud SH, Morisky DE. Differences in medication adherence, satisfaction and clinical symptoms in schizophrenic outpatients taking different antipsychotic regimens. Curr Drug Saf (2011.0) 6 285-290. DOI: 10.2174/157488611798918755 24. Sweileh WM, Ihbesheh MS, Jarar IS, Sawalha AF, Abu Taha AS, Zyoud SH, Morisky DE. Antipsychotic medication adherence and satisfaction among Palestinian people with schizophrenia. Curr Clin Pharmacol (2012.0) 7 49-55. DOI: 10.2174/157488412799218761 25. Sweileh WM, Ihbesheh MS, Jarar IS, Taha AS, Sawalha AF, Zyoud SH, Jamous RM, Morisky DE. Self-reported medication adherence and treatment satisfaction in patients with epilepsy. Epilepsy Behav (2011.0) 21 301-305. DOI: 10.1016/j.yebeh.2011.04.011 26. 26.Palestinian Ministry of Health. Statistical reports: Health Annual Report. Palestine 2020. 2021. https://site.moh.ps/Content/Books/jY1OqK8mOkhBqkduKGf7bxi8UVWhHmld3gx8PdFh4FQffHoRxsUyan_B34WW1q4sxXNmQBvvJWN6LBNg7N5RL1VxOabi5IfvUzmUy1AzY8Fgy.pdf. Accessed 1 June 2022. 27. Ayres LR, Baldoni Ade O, Borges AP, Pereira LR. Adherence and discontinuation of oral hormonal therapy in patients with hormone receptor positive breast cancer. Int J Clin Pharm (2014.0) 36 45-54. DOI: 10.1007/s11096-013-9833-5 28. Liu Y, Malin JL, Diamant AL, Thind A, Maly RC. Adherence to adjuvant hormone therapy in low-income women with breast cancer: the role of provider-patient communication. Breast Cancer Res Treat (2013.0) 137 829-836. DOI: 10.1007/s10549-012-2387-8 29. Bosco-Lévy P, Jové J, Robinson P, Moore N, Fourrier-Réglat A, Bezin J. Persistence to 5-year hormonal breast cancer therapy: a French national population-based study. Br J Cancer (2016.0) 115 912-919. DOI: 10.1038/bjc.2016.276 30. Lailler G, Memoli V, Le Bihan BC, Ben Diane MK, Lauzier S, Mancini J, Bousquet PJ, Bouhnik AD. Five-year adjuvant endocrine therapy adherence trajectories among women with breast cancer: a nationwide French study using administrative data. Clin Breast Cancer (2021.0) 21 e415-e426. DOI: 10.1016/j.clbc.2021.01.007 31. He W, Fang F, Varnum C, Eriksson M, Hall P, Czene K. Predictors of discontinuation of adjuvant hormone therapy in patients with breast cancer. J Clin Oncol (2015.0) 33 2262-2269. DOI: 10.1200/JCO.2014.59.3673 32. Stahlschmidt R, Ferracini AC, de Souza CM, de Medeiros LM, Juliato CRT, Mazzola PG. Adherence and quality of life in women with breast cancer being treated with oral hormone therapy. Support Care Cancer (2019.0) 27 3799-3804. DOI: 10.1007/s00520-019-04671-x 33. Brito C, Portela MC, de Vasconcellos MT. Adherence to hormone therapy among women with breast cancer. BMC Cancer (2014.0) 14 397. DOI: 10.1186/1471-2407-14-397 34. Iacorossi L, Gambalunga F, Fabi A, Giannarelli D, Facchinetti G, Piredda M, De Marinis MG. Adherence to hormone therapy in women with breast cancer: a quantitative study. Prof Inferm (2016.0) 69 113-121. PMID: 27600553 35. Wells KJ, Pan TM, Vazquez-Otero C, Ung D, Ustjanauskas AE, Munoz D, Laronga C, Roetzheim RG, Goldenstein M, Carrizosa C. Barriers and facilitators to endocrine therapy adherence among underserved hormone-receptor-positive breast cancer survivors: a qualitative study. Support Care Cancer (2016.0) 24 4123-4130. DOI: 10.1007/s00520-016-3229-8 36. Ibrar M, Peddie N, Agnew S, Diserholt A, Fleming L. Breast cancer survivors' lived experience of adjuvant hormone therapy: a thematic analysis of medication side effects and their impact on adherence. Front Psychol (2022.0) 13 861198. DOI: 10.3389/fpsyg.2022.861198 37. Kroenke CH, Hershman DL, Gomez SL, Adams SR, Eldridge EH, Kwan ML, Ergas IJ, Kubo A, Kushi LH. Personal and clinical social support and adherence to adjuvant endocrine therapy among hormone receptor-positive breast cancer patients in an integrated health care system. Breast Cancer Res Treat (2018.0) 170 623-631. DOI: 10.1007/s10549-018-4774-2 38. Hsieh KP, Chen LC, Cheung KL, Chang CS, Yang YH. Interruption and non-adherence to long-term adjuvant hormone therapy is associated with adverse survival outcome of breast cancer women–an Asian population-based study. PLoS ONE (2014.0) 9 e87027. DOI: 10.1371/journal.pone.0087027 39. Hurtado-de-Mendoza A, Cabling ML, Lobo T, Dash C, Sheppard VB. Behavioral interventions to enhance adherence to hormone therapy in breast cancer survivors: a systematic literature review. Clin Breast Cancer (2016.0) 16 247-255.e243. DOI: 10.1016/j.clbc.2016.03.006 40. Krok-Schoen JL, Naughton MJ, Young GS, Moon J, Poi M, Melin SA, Wood ME, Hopkins JO, Paskett ED, Post DM. Increasing adherence to adjuvant hormone therapy among patients with breast cancer: a smart phone app-based pilot study. Cancer Control (2019.0) 26 1073274819883287. DOI: 10.1177/1073274819883287 41. Bhandari S, Ngo P, Kute B, Mandadi M, Pitman WA, Wu X, Lloyd S, Brown MC, Rai SN, Jain D. Adherence to anti-estrogen therapy in women with hormone receptor-positive breast cancer utilizing bubble packaging: a pilot study. Breast Cancer Res Treat (2019.0) 177 395-399. DOI: 10.1007/s10549-019-05308-7 42. Toivonen KI, Carlson LE, Rash JA, Campbell TS. A Survey of potentially modifiable patient-level factors associated with self-report and objectively measured adherence to adjuvant endocrine therapies after breast cancer. Patient Prefer Adherence (2021.0) 15 2039-2050. DOI: 10.2147/PPA.S319087 43. Van Liew JR, Christensen AJ, de Moor JS. Psychosocial factors in adjuvant hormone therapy for breast cancer: an emerging context for adherence research. J Cancer Surviv (2014.0) 8 521-531. DOI: 10.1007/s11764-014-0374-2
--- title: Early-induced diabetic obese rat MACAPOS 2 authors: - Joseph Ngakou Mukam - Clémence Mvongo - Sandrine Nkoubat - Gaëtan Olivier Fankem - Adamou Mfopa - Paul Aimé Noubissi - Michel Archange Fokam Tagne - René Kamgang - Jean-Louis Essame Oyono journal: BMC Endocrine Disorders year: 2023 pmcid: PMC10026472 doi: 10.1186/s12902-022-01252-8 license: CC BY 4.0 --- # Early-induced diabetic obese rat MACAPOS 2 ## Abstract ### Background Diabetes mellitus is a metabolic disease characterized by an abnormally high blood glucose level. Glucose intolerance and insulin resistance are two characteristics that promote the onset and development of type 2 diabetes. The aim of this study was to create a diabetic rat model from obese rat MACAPOS 2. ### Methods A group of rats was subjected to a high-fat diet (HFD) compared to a control group (NC) which received a normal diet. After 16 weeks of HFD, Lee index was calculated, obese rats were subjected to an oral glucose tolerance test (OGTT) and insulin tolerance test (ITT). One group of HFD rats (HFDZ) received streptozotocin 22.5 mg/kg (iv). One week later, weight gain, water and food intakes, urine volume and fasting blood glucose levels were evaluated. Animals were also subjected to glucose tolerance and insulin tolerance tests. ### Results After 16 weeks of HFD, rats became obese, glucose intolerant and resistant to insulin. The body weight of rats was significantly high (+ $26.23\%$) compared to normal rats, glycemia remained significantly high (+ $45.46\%$, $P \leq 0.01$) two hours after administration of glucose in high-fat diet rats, water intake and urine volume were comparable to those of NC. In HFD, the streptozotocin injected after one week (HFDZ), amplified glucose intolerance. During ITT, glycemia remained significantly ($P \leq 0.01$) high from 15 min; and did not vary during the 60 min of ITT. The fasting glycemia one week after streptozotocin injection was significantly high (288 mg/dL) compared to HFD (114 mg/dL), associated whit a significant ($P \leq 0.01$) increase in water intake and 24 h urine volume. ### Conclusion These results showed that MACAPOS 2 associated with a low dose of streptozotocin (22.5 mg/dL) early leads to the diabetes in obese albinos Wistar rats and could be a real model to study the type 2 diabetes mellitus. ## Background Diabetes mellitus is one of the most common metabolic diseases that is caused by an absolute or relative deficiency in insulin secretion and/or insulin action [1, 2]. In other to study, understand and manage this metabolic affection, many diabetic animal models are used by the scientific community to evaluate the efficiency of many potential substances in diabetes management. Among the many existing models, the type 1 model is obtained by injection of chemical substances (Streptozotocin, alloxan) that lead either to a total or partial destruction of the pancreatic beta-cells [3–5]. Streptozotocin (STZ) is used at high doses to induce diabetes in rodents (animal models of insulin-dependent diabetes mellitus) characterized by high fasting blood glucose levels. Type 2 diabetes (T2D) is caused by a combination of genetic factors related to insulin resistance and environmental factors such as obesity, lack of exercise, and stress [6]. Different T2D animal models remain indispensable for discovering, validating, and optimizing novel therapeutics for their safe use in human diabetes management. Many genetic and spontaneous models of type 2 diabetes are developed; however, these models are not available in developing countries and, many of these models do not really reflect the human cases of diabetes. Nongenetic models are more like human T2D disease with regard to the initiation and progression phases. The high-fat diet is known to induce obesity, insulin resistance and glucose intolerance in albinos Wistar rats [7]. The MACAPOS 1 diabetic and MACAPOS 2 obese models’ rats were created using 16 weeks of Cameroon’s local hypercaloric diet [8]. MACAPOS 1 combined with dexamethasone reduced to 10 weeks the induction period and induced more severe hyperglycemia [9]. Unlike MACAPOS 1, MACAPOS 2 leads to common obesity: after 16 weeks of high-fat diet, the animal became obese, glucose intolerant and insulin resistant. MACAPOS 2 rats do not develop real fasting hyperglycemia [7, 8]. Streptozotocin, at high doses, causes beta islet cell death by acute oxidative stress [5, 10]. However, the injection of a low dose in rats previously subjected to a high-fat diet leads to the onset of type 2 diabetes [11, 12]. In the aim of finding a new animal model of diabetes study that is closer to the human model in our local context, the present study was undertaken to create a combined model MACAPOS 2-streptozotocin. ## Animals Wistar rats were bred in the animal house of the Laboratory of Human Metabolism and non-Communicable Diseases. They were maintained under natural light /dark cycle and fed with a standard local diet (3.400 kcal: carbohydrates 50–$55\%$, fats 15–$20\%$, proteins 25–$30\%$), and had access to water ad libitum [8]. In vivo experiments were conducted with the approval from the institutional committee of the Cameroonian Ministry of Scientific Research and Innovation which has adopted the guidelines and regulations of the European Union on Animal Care (CEE Council $\frac{86}{609}$) [13]. The ARRIVE guidelines were also observed. ## MACAPOS 2 obese rats Six weeks old male albinos Wistar rats were selected for obesity induction. They were randomly divided into two groups, normal control (NC, submitted to the standard diet) and high-fat diet (HFD) group (Table 1), with free access to water [8]. After four [4] months under the respective diets, Lee index and body weight gain allowed us to select obese rats [7]. They were then subjected to oral glucose and insulin tolerance tests to select glucose intolerant and insulin resistant ones. Table 1Diet compositions and their caloric intake (energy)GroupsNormal Diet (ND)High Fat Diet (HFD)Maize25080Wheat400110Soya bean150280Steeped cassava–220Fish flour10030Sucrose–50Palm oil–200Bones flour1020Cabbage palm cake80–Vitamins complex1010Energy (Kcal/Kg)34004730The components were obtained from Yaoundé (Cameroon) local market and are expressed in g/kg of diet Lee index (Li) was calculated, using the body weight (bw) and Naso-anal length (Lna) as follow [9–14]:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\textrm{Li}=\frac{\sqrt[3]{bw}}{Lna}$$\end{document}Li=bw3Lna ## Obese diabetic rats Obese rats received a unique dose of streptozotocin (22.5 mg/kg i.v. Sigma Aldrich). To ensure the implication of the diet on the genesis of diabetes, another group consisting only of normal rats was used in comparison to the obese rats’ group; and these normal rats also received streptozotocin. One week after streptozotocin administration, weight gain, water and food intakes, 24 hours urine volume and fasting blood glucose levels were recorded. All animals were subjected to glucose and insulin tolerance tests. ## Oral glucose tolerance test (OGTT) After 12 h fasting the rats received per os 2.5 mg/kg of glucose. Prior to glucose administration, blood glucose was estimated, and then at 30, 60, and 120 min after oral glucose administration [8]. Glycemia was evaluated using test strips ACCU-Chek Active and a glucometer of the same brand. ## Insulin tolerance test (ITT) After a 12 h fasting period, the blood glucose level of each animal was evaluated; each animal received 2 UI/kg bw S.c insulin (Insulin Actrapid Human HM). The glycemia was then estimated at 10, 20, 30, and 60 min after insulin administration [9]. ## Statistical analysis The results were expressed as mean ± standard error of the mean. The statistical analyses were performed by one-way analysis of variance (ANOVA) associated with the Turkey test followed by the Dunnett test, using the computer Graphpad Prism 8.0.1. The difference between and within various groups was significant with $P \leq 0.05.$ ## Body weight, water and food intakes, urine volume After 16 weeks of a high-fat diet, rat body weight significantly increased compared to those submitted to a normal diet (NC): (+ 26,$23\%$, $P \leq 0.01$, Fig. 1A). The Streptozotocin administration caused a non-significant reduction in the weight of rats submitted to high-fat diet (Fig. 1A). In HFD, food intake was significantly low ($P \leq 0.01$) compared to NC. HFDZ rats’ food intake significantly increased compared to HFD, but remained significantly low compared to NC (Fig. 1B). The initial caloric intake was higher in HFD (1.25 kcal/g bw) compared to NC (1.15 kcal/g bw) (Fig. 1B). Water intake and 24 h urine volume of HFD were not significantly different to NC (Fig. 1C) but were significantly ($P \leq 0.01$) increased in HFDZ rats compared to NC and HFD rats (Fig. 1C). In HFD rats, the Lee index was greater than 0.3, and less than 0.3 in NC (Fig. 1D). The relative weights of visceral and subcutaneous fat were remarkably high in animals subjected to a high-fat diet compared to NC. Also, streptozotocin administration brought a significant increase of fat relative weight in HFDZ animals (Fig. 1E).Fig. 1Body weight (A), food intake (B), water intake (WI) and 24 h urine volume (UV) (C), Lee index (D), visceral (VF) and subcutaneous (SF) fats (E) of 16 weeks high-fat diet (HFD), and high-fat diet plus streptozotocin (HFDZ). Ici: initial caloric intake ## Obese rat MACAPOS 2 High-fat diet fed rats presented high blood glucose level (110.0 ± 1.6 mg/dL) compared to normal diet fed control group (84.0 ± 3.0 mg/dL, Fig. 2). Glycemia variation remained significantly high (+ $45.46\%$, $P \leq 0.01$) 2 h after administration of glucose in high-fat diet rats compared to the NC rats (Fig. 3A). After insulin administration, glycemia remained significantly ($P \leq 0.01$) high from 15 min in high fat diet rats compared to those of NC (Fig. 3B).Fig. 2Fasting glycemia of 16 weeks high-fat diet (HFD) and high-fat diet plus streptozotocin (HFDZ). NC: normal control rats; NCZ: normal rats receiving streptozotocin. Significant difference: *$P \leq 0.01$, $P \leq 0.05$ compared to NC; bP < 0.01 compared to HFDFig. 3Glycemia of oral glucose (A) and insulin tolerance tests (B) of 16 weeks high-fat diet (HFD) and high-fat diet plus streptozotocin (HFDZ), expressed in percentage of variation for A, and in percentage of the initial value (iv = $100\%$) for B. NC: normal control rats. Significant difference: $P \leq 0.01$ compared to NC; aP < 0.01, bP < 0.01 compared to HFD. (): Area Under the Curve ## Diabetic rat MACAPOS 2 Streptozotocin administration at 22.5 mg/kg (iv) did not significantly increase blood glucose level in normal rats but increased glycemia in HFD rat from 110 mg/kg to 288 mg/dL, ($P \leq 0.01$, Fig. 2). Streptozotocin amplified glucose intolerance; blood glucose levels remained significantly high at 60 min (+ $20.3\%$, $P \leq 0.01$) and 120 min (+ $8.5\%$, $P \leq 0.05$) in HFDZ rats compared to those of HFD (Fig. 3A). In HFDZ rats as HFD rats, blood glucose levels remained high during the 60 min of ITT (Fig. 3B). ## Discussion This study was undertaken to set up in our local context, a new diabetes model close to the human model. Submitted to a high-fat diet, animals became obese after 16 weeks. A high-fat diet leads to obesity, hyperinsulinemia, and altered glucose homeostasis which may be due to insufficient compensation by the beta cells of the pancreatic islets [15]. The weight gain and increase in Lee index observed in HFD rats are generally due to an increase in fat mass, which could also justify the glucose intolerance and insulin resistance of these rats. In the high-fat diet, the total energy intake was greater compared to the normal diet. This could explain why despite the low food intake, the HFD animals gained more weight. The increase of the body fat tissue may not only result from the fat content of the diet, but also from the energy intake, which might lead to various metabolic alterations such as insulin resistance, reduction of lipolytic activity in fat tissue, and impairment of mitochondrial metabolism [16]. Metabolomic studies on HFD-fed mice have shown incomplete oxidation of fatty acids, accompanied by an increase in whole-body fatty acid oxidation [16, 17]; HFD led to the accumulation of fatty acid oxidation byproducts in skeletal muscle, in turn, this contributes to insulin resistance in muscle [16]. In fact, the energy used by the muscle is produced primarily by free fatty acids oxidation with muscle glycogen stores remaining intact, and repression of glycogen synthase [18]. These mechanisms contribute to a rise in blood glucose levels by reducing the muscle capacity to use and store glucose and an increase in hepatic glucose release [19]. Furthermore, in obesity, oxidative stress, inflammation, and excessive production of certain adipokines such as resistin, increase insulin resistance [20–22]. Dietary nutrients (sucrose and palm oil) found in the administered diet might have a profound influence on insulin action, also, it could be associated with impaired mitochondrial function; with the production of malonyl-CoA, which reduces GLUT4 efficiency [23]. These mechanisms might contribute to the development of the observed glucose intolerance [8]. MACAPOS 2 is known to induce dyslipidemia, glucose intolerance and insulin resistance [7, 8, 24]. Administration of a low dose of streptozotocin (22.5 mg/kg) did not induce diabetes in normal rats, but caused in HFD rats, within a week, a diabetes state characterized by fasting hyperglycemia. Streptozotocin at a high dose (40 mg/kg) induces hyperglycemia by a selective toxicity activity on pancreatic beta cells [5, 25, 26]. Streptozotocin damages the pancreatic beta cells and induces hyperglycemia; this effect is not observed at low doses [5]. In this study, a low dose of streptozotocin (22.5 mg/kg) developed fasting hyperglycemia in obese rat MACAPOS 2. This result could be due to diet-induced glucose intolerance and insulin resistance which are characteristic of a common obesity in human. This common obesity created a favorable stage for the onset of hyperglycemia and type 2 diabetes [27]. Generally, STZ-induced type 1 diabetes is also characterized by body mass loss [28]; in our studied model, the body mass, visceral and subcutaneous fat remained significantly higher after STZ administration. This is also the case in the type 2 diabetes model. The hyperglycemia observed in MACAPOS 2 rat was accompanied by diabetes characteristics such as the increase in water and food intake and 24 h urine volume. Hyperglycemia causes an increase in urine volume due to osmotic diuresis [29], which also explains the increase in water intake. In fact, polyuria leads to dehydration, which causes a strong feeling of thirst [29]. These last symptoms associated with insulin resistance and glucose intolerance, confirm a real type 2 diabetes as observed in human beings. In conclusion, the Cameroon’s local diet associated with a low dose of streptozotocin quickly leads in obese MACAPOS 2 albinos Wistar rats to type 2 diabetes characterized by fasting hyperglycemia, glucose intolerance, insulin resistance, as well as an increase in urine volume and water intake. This model could be used as a model of type 2 diabetes mellitus in scientific studies. ## References 1. Diagnosis and Classification of Diabetes Mellitus*. Diabetes Care* (2014.0) 37 81-90. DOI: 10.2337/dc14-S081 2. 2.Word Hearth Organization (WHO)Classification of diabetes mellitus2019GenevaWorld Health Organization. Classification of diabetes mellitus (2019.0) 3. Ankur R, Shahjad A. Alloxan Induced Diabetes: Mechanisms and Effects. Int J Res Pharm Biomed Sci (2012.0) 3 819-823 4. Kottaisamy Chidhambara PD, Raj Divya S, Prasanth Kumar V, Sankaran U. Experimental animal models for diabetes and its related complications-a review. Lab Anim Res (2021.0) 37 1-14. PMID: 33397516 5. Furman BL. Streptozotocin-induced diabetic models in mice and rats. Curr Protoc (2021.0) 1 1-21 6. Kaku K. Pathophysiology of Type 2 Diabetes and Its Treatment Policy. J Japan Med Assoc (2010.0) 53 41-46 7. 7.Adamou M, Mediesse Francine K, Clémence M, Sandrine NC, Lum Ambe A, Eugène S, et al. Antidyslipidemic Potential of Water-Soluble Polysaccharides of Ganoderma applanatum in MACAPOS-2-Induced Obese Rats. Evid Based Complementary Altern Med. 2021;2021:1–9. 8. Kamgang R, Youmbi R, Mengue GP, Ngogang Y. Cameroon local-diet induced glucose intolerance and dyslipidemia in adult Wistar Rat. Diabetes Rese Clin Pract (2005.0) 69 224-230. DOI: 10.1016/j.diabres.2005.02.005 9. Mvongo C, Kamgang R, Minka Minka CS, Mfopa A, Essame Oyono JL. Effect of ethanol-water extract of Crinum jagus on glycemia reactivity in dexamethasone-induced diabetic rat macapos 1. Int J Pharma Res Bio Sci (2014.0) 3 157-167 10. Li W, Minjie W, Hengguang Z, Pufu L, Baosha T, Junchen C, Yibin L. Hypoglycemic effect of okra aqueous extract on streptozotocin-induced diabetic rats. Food Sci Technol (2020.0) 40 972-978. DOI: 10.1590/fst.28619 11. Correia-Santos AM, Suzuki A, Anjos JS, Rêgo TS, KCl A, Boaventura GT. Induction of Type 2 Diabetes by low dose of streptozotocin and high-fat diet-fed in wistar rats. Medicina (Ribeirão Preto) (2012.0) 45 436-444. DOI: 10.11606/issn.2176-7262.v45i4p436-444 12. Diego De Magalhães A, Kume Willian T, Correia Francyele S, Queiroz Thaís S, Allebrandt Neto Edgar W, Dos Santos MP, Kawashita Nair H, De França SA. High-fat diet and streptozotocin in the induction of type 2 diabetes mellitus: a new proposal. Ana Acad Bras Cien (2019.0) 91 e20180314. DOI: 10.1590/0001-3765201920180314 13. Smith J, van den Broek F, Martorel J, Hackbarth H, Ruksenas O, Zeller W. FELASA Working Grp, and Principles and Practice in Ethical Review of Animal Experiments across Europe: Summary of the Report of a Felasa Working Group on Ethical Evaluation of Animal Experiments. Lab Anim (2007.0) 41 143-160. DOI: 10.1258/002367707780378212 14. Nya JGBS, Kamgang R, Nkomo Soua PR, Essame Oyono JL, Njikam N. Effect of methanol/methylene chloride extract of Terminalia glaucescens leaves on normal mice. Asian J Pharm Clin Res (2008.0) 1 54-59 15. Vatandoust N, Rami F, Salehi AR, Khosravi S, Dashti G, Eslami G, Momenzadeh S, Salehi R. Novel High-Fat Diet Formulation and Streptozotocin Treatment for Induction of Prediabetes and Type 2 Diabetes in Rats. Adv Biomed Res (2018.0) 7 1-5. PMID: 29456972 16. Gary LD. Fatty acid oxidation and its relation with insulin resistance and associated disorders. Ann Nutr Metab (2016.0) 68 15-20. PMID: 27931032 17. Scheen AJ. Le concept d’insulinosensibilité. Diabète Métab (2001.0) 27 193-200. PMID: 11452210 18. Cazzolli R, Craig DL, Biden TJ, Schmitz PC. Inhibition of glycogen synthesis by fatty acid in C2C12muscle cells is independent of PKC-α, -ε, and -θ. Am J Physiol Endocrinol Metab (2002.0) 282 E1204-E1213. DOI: 10.1152/ajpendo.00487.2001 19. Jeevendra Martyn JA, Masao K, Shingo Y. Obesity-Induced Insulin Resistance and Hyperglycemia: Etiological Factors and Molecular Mechanisms. Anesthesiology. (2008.0) 109 137-148. DOI: 10.1097/ALN.0b013e3181799d45 20. Jianping Y. Mechanisms of insulin resistance in obesity. Front Med (2013.0) 7 14-24. DOI: 10.1007/s11684-013-0262-6 21. Leal VO, Mafra D. Adipokines in obesity. Clin Chim Acta (2013.0) 419 87-94. DOI: 10.1016/j.cca.2013.02.003 22. Spinas GA, Lehmann R. Diabète sucré : Diagnostic, classification et pathogenèse. Forum Med Suisse (2001.0) 20 519-525 23. Coelho DF, Pereira-Lancha LO, Chaves DS, Diwan D, Ferraz R, Campos-Ferraz PL, Poortmans JR, Lancha Junior AH. Effect of high-fat diets on body composition, lipid metabolism and insulin sensitivity, and the role of exercise on these parameters. Braz J Med Biol Res (2011.0) 44 966-972. DOI: 10.1590/S0100-879X2011007500107 24. Kamgang R, Youmbi Mboumi R, Foyet Fondjo A, Fokam Tagne MA, Mengue Ndille GPR, Ngogang YJ. Antihyperglycaemic potential of the water-ethanol extract of Kalanchoe crenata (Crassulaceae). J Nat Med (2008.0) 62 34-40. DOI: 10.1007/s11418-007-0179-y 25. Liu RM, Dai R, Luo Y, Xiao JH. Glucose-lowering and hypolipidemic activities of polysaccharides from Cordyceps taii in streptozotocin-induced diabetic mice. BMC Complement Altern Med (2019.0) 19 10. PMID: 30621705 26. Kiss ACI, Lima PHO, Sinzato YK, Takaku M, Takeno MA, Rudge MVC, Damasceno DC. Animal models for clinical and gestational diabetes: maternal and fetal outcomes. Diabetol Metab Syndr (2009.0) 1 1-7. DOI: 10.1186/1758-5996-1-21 27. Silvia MF, Marta GR, Gema V, Astier J, Landrier JF, Marta M. High Fat/High Glucose Diet Induces Metabolic Syndrome in an Experimental Rat Model. Nutrients. (2018.0) 10 1-15 28. Zafar M, et Naeem-ul-hassan NS. Effects of STZ-Induced diabetes on the relative weights of kidney, liver and pancreas in albino rats: a comparative study. Int J Morphol (2010.0) 28 135-142. DOI: 10.4067/S0717-95022010000100019 29. Hyun-Jong S, Jae-Ha K, Joo-Hark Y, Sang-Woong H, Ho-Jung K. Polyuria with the Concurrent manifestation of Central Diabetes Insipidus (CDI) & Type 2 Diabetes Mellitus (DM). Electrolyte Blood Press (2012.0) 10 26-30. DOI: 10.5049/EBP.2012.10.1.26
--- title: 'The state of diabetes care and obstacles to better care in Aceh, Indonesia: a mixed-methods study' authors: - Hizir Sofyan - Farah Diba - Suryane S. Susanti - Marthoenis Marthoenis - Ichsan Ichsan - Novi Reandy Sasmita - Till Seuring - Sebastian Vollmer journal: BMC Health Services Research year: 2023 pmcid: PMC10026477 doi: 10.1186/s12913-023-09288-9 license: CC BY 4.0 --- # The state of diabetes care and obstacles to better care in Aceh, Indonesia: a mixed-methods study ## Abstract ### Background Cardio-metabolic diseases are a major cause of death worldwide, including in Indonesia, where diabetes is one of the most critical diseases for the health system to manage. ### Methods We describe the characteristics, levels of control, health behavior, and diabetes-related complications of diabetes patients in Aceh, Indonesia. We use baseline data and blood testing from a randomized-controlled trial. We conducted semi-structured interviews with eight health providers from Posbindu and Prolanis programs that target diabetes and other non-communicable diseases (NCDs). We also conducted three focus group discussions with 24 diabetes patients about their experiences of living with diabetes and the existing support programs. ### Results The blood tests revealed average HbA1c levels indicative of poor glycemic control in 75.8 percent of patients and only 20.3 percent were free from any symptoms. Our qualitative findings suggest that patients are diagnosed after diabetes-related symptoms manifest, and that they find it hard to comply with treatment recommendations and lifestyle advice. The existing programs related to NCDs are not tailored to their needs. ### Conclusion We identify the need to improve diabetes screening to enable earlier treatment and achieve better control of the disease. Among diagnosed patients, there are widespread beliefs about diabetes medication and alternative forms of treatment that need to be addressed in a respectful dialogue between healthcare professionals and patients. Current diabetes screening, treatment and management programs should be revised to meet the needs of the affected population and to better respond to the increasing burden of this disease. ## Introduction In Indonesia, diabetes prevalence is $8.2\%$, slightly above the average of $7.5\%$ for all low- and middle-income countries (LMICs) [1]. Although these rates are similar to those in Western societies, many LMIC health systems are not well prepared to deal with this increased burden of diabetes. Manne-Goehler et al. [ 2019] found, for a sample of 28 LMICs, that less than half of all people living with diabetes have been diagnosed [2]. While most diagnosed people receive some sort of treatment, only half are adequately controlled. In Indonesia, about $21\%$ of people living with diabetes are diagnosed and just one-third of them ($7\%$ of all people living with diabetes) are adequately controlled. This health burden also has economic implications. Bommer et al. [ 2017] estimate that the global cost of diabetes in 2015 was 1.3 trillion US dollars, which is equivalent to $1.8\%$ of global GDP [3]. In Indonesia, direct costs were around 3.6 billion US dollars, equivalent to $0.4\%$ of Indonesia’s GDP and $13.7\%$ of Indonesia’s total health expenditure. The losses to the labor market were of similar magnitude, bringing the total cost of diabetes to $0.8\%$ of GDP. This highlights the importance of reducing the number of new diabetes cases and improving care for people living with diabetes, from both a quality of life and an economic perspective. This study examines how to improve the treatment of people who already have a diabetes diagnosis. Indonesia has several public programs to reduce the burden of NCDs, including health screenings, integrated management of diabetes and hypertension at public health posts (Puskesmas), screening for breast and cervical cancer, smoking cessation programs, increasing physical activity, and healthy eating [4]. Two widely-implemented programs at the primary care level, Posbindu and Prolanis, focus on the early detection of cardio-metabolic diseases and the prevention of secondary complications. Posbindu is a program that runs at Puskesmas and focuses on early detection, prevention and monitoring of NCD risk factors within their communities. It targets at-risk groups and people with NCDs aged 15 years and more. Prolanis is a proactive health care program implemented in an integrated way involving participants, Puskesmas, and the public health insurance system (BPJS). It aims to achieve an optimal quality of life with cost-effective and efficient health services for patients with chronic diseases. While Posbindu focuses on NCDs of any kind, Prolanis focuses on people with diabetes or hypertension. While these programs are intended to be universally accessible under universal health coverage, little is known about their efficacy [5] (see Table 1 for more information about Prolanis and Posbindu).Table 1Description of existing chronic disease programsPosbinduEvery Puskesmas in Indonesia hosts the Posbindu PTM program, which is implemented by the Indonesian Ministry of Health. Each Puskesmas has a dedicated Posbindu PTM program manager who is responsible for the treatment of patients with a NCD, including type 2 diabetes. Because Puskesmas cover relatively large geographical areas, especially in rural regions, the Posbindu program is provided at village level to increase the accessibility of health services. Usually, one doctor, two nurses, and local community health workers from the villages attend monthly meetings and provide health examinations, counselling, and health education, as well as basic physical examinationsProlanisSimilar to Posbindu PTM, all Puskesmas in Banda Aceh and Aceh Besar offer the Prolanis program. Prolanis was designed by the Indonesian National Health Insurance Organization (BPJS Kesehatan) specifically for patients with diabetes and hypertension. At each Puskesmas, a group of at least 30 patients with diabetes or hypertension is established. Prolanis offers weekly activities at the Puskesmas and this includes physical exercise. Unlike Posbindu, all Prolanis activities are held centrally at Puskesmas. Each Prolanis group is supported by a team consisting of one nurse (who could be a Posbindu PTM program manager) who functions as the Prolanis program manager, one doctor and one additional nurse. Each Puskesmas is free to choose the staff involved in the program We use a mixed-methods approach to better understand the characteristics and experiences of people living with diabetes in Aceh province in Indonesia. We do a descriptive investigation of the characteristics of patients with type 2 diabetes, based on data recently collected in Banda Aceh and Aceh Besar. We then investigate how the Posbindu and Prolanis programs function and issues related to their implementation, using information from qualitative interviews with program managers and participants. ## Methods To identify the current health status of people with diabetes in Aceh province, we explore baseline data from a randomized controlled trial (RCT) conducted in early 2019. The sample was drawn from people who had a diagnosis of Type 2 diabetes at primary health posts in Banda Aceh and Aceh Besar [6]. This RCT investigated the effect of peer education sessions for type 2 diabetes on diabetes-related outcomes in the province of Aceh within local primary health care posts. Criteria for inclusion in the RCT were [1] treated for type 2 diabetes in the Puskesmas of Banda Aceh or Aceh Besar and [2] aged between 20 and 89 years. Puskesmas staff contacted people on their patient lists to recruit them to this study. A total of 533 participants were recruited and baseline data were collected in March and April 2019. We obtained written consent from the patients. Patients were informed that they could drop out of the study at any point. Further details about the data collection can be found in Seuring et al. [ 2019] [6]. 521 participants had non-missing data on age, education, and relevant biomarkers. The statistical analysis is limited to the calculation of averages, standard deviations, and frequencies. Poor glycemic control is defined as a HbA1c level greater than or equal to 64 mmol/l; high cholesterol is defined as a total cholesterol level greater than or equal to 6.2 mmol, and hypertension is defined as systolic blood pressure above 130 and diastolic blood pressure above 80. In addition to providing estimates of lab-based measures of HbA1c, cholesterol and BMI (based on measured waist and height) to assess the metabolic profile of type 2 diabetes patients in Aceh and self-reported information on prevalent diabetes complications, we also use information on diabetes-related distress, based on the two-item diabetes distress scale (DDS2) [7]. The DDS2 has been validated and is considered a good alternative to longer, more extensive scales such as the DDS17 [7]. The DDS2 measures diabetes-distress based on two questions with a 6-item Likert response scale, going from 1 (no problem) to 6 (serious problem). The questions are:Feeling overwhelmed by the demands of living with diabetes. Feeling that I am often failing with my diabetes regimen. An average score of ≥ 3 was considered to be indicative of diabetes-specific distress, following the recommendations of Fisher et al. [ 2008] [7]. For physical activity we calculated a binary measure of whether participants achieved World Health Organisation (WHO)-recommended levels of physical activity. The threshold is > 600 min per week of total physical activity metabolic equivalent of task (MET), using data collected with the Global Physical Activity Questionnaire (GPAQ). We conducted semi structured individual interviews with eight health officers from the Aceh Provincial Health Office. The interviews were conducted in eight Puskemas with a high share of type 2 diabetes patients. Two Puskesmas were located in Banda Aceh and six in Aceh Besar. The interviews were designed to ascertain the level of engagement with the current NCD programs and activities conducted in Puskesmas. Following the interviews, diabetes patients from these Puskesmas were invited to participate in focus group discussions (FGDs). Three FGDs were conducted with 24 patients in two Puskesmas in Aceh Besar and one Puskesmas in Banda Aceh. The FGDs sought information on patients’ understanding of diabetes and their motivation for, and feelings about, participating in the Posbindu PTM and Prolanis programs, and explored their experiences of receiving diabetes treatment in Puskesmas. The sampling strategy for interviews and FGDs was purposive sampling. The inclusion criteria for interviews were: health officer of Puskesmas in Banda Aceh or Aceh Besar, who is responsible for Prolanis and Posbindu, and had at least one year of work experience. The inclusion criteria for FGDs were: patients diagnosed with diabetes from Banda Aceh or Aceh Besar, who actively participated in the Posbindu and Prolanis activities, and who were at least 18 years old. We obtained the contact details of diabetes patients from Prolanis and Posbindu staff. The interviews and FGDs were conducted face-to-face by study authors FD and SSS, and lasted approximately one hour. With participants’ permission, the interviews and FGDs were recorded on an encrypted laptop or audio recording device and then transcribed verbatim by the research team. The interviews and the FGDs were conducted in Bahasa Indonesia and later translated to English after transcripts were anonymized and all names were replaced by identifier numbers. We then conducted line by line analysis and utilized inductive thematic analysis to code the transcripts and cluster them into themes, along with anonymized quotes to support the themes, for presentation in the final report. ## Descriptive Statistics Our descriptive results are stratified by men and women, with men being older and having higher formal education levels (Table 2). The results show very high average HbA1c levels of 84 mmol/l ($9.8\%$), well above common thresholds for uncontrolled diabetes of either 53 mmol/l ($7\%$), which is also the target HbA1c level recommended by the Indonesian Society of Endocrinology (PERKENI) T2DM guidelines, or 75 mmol/l ($9\%$) [8]. The majority of women and around $45\%$ of men were either overweight or obese. About one-quarter of the population had blood pressure levels indicative of hypertension and $25\%$ of men and $34\%$ of women had high total cholesterol levels. Table 2Characteristics of patients diagnosed with diabetes in Aceh, IndonesiaMaleFemale($$n = 95$$)($$n = 426$$)p-valueAge Mean (SD)58.12 (8.92)52.83 (10.17) < 0.001Highest level of education Less than primary1 ($1.1\%$)28 ($6.6\%$) < 0.001 Primary21 ($22.1\%$)135 ($31.7\%$) Secondary51 ($53.7\%$)223 ($52.3\%$) Post-secondary22 ($23.2\%$)40 ($9.4\%$)HbA1c (mmol/l) Mean (SD)85.18 (28.75)84.47 (26.03)0.814Poor glycemic control (HbA1c > = 64 mmol/l) No25 ($26.3\%$)99 ($23.2\%$)0.524 Yes70 ($73.7\%$)327 ($76.8\%$)Total cholesterol (mmol/l) Mean (SD)5.42 (1.45)5.67 (1.26)0.087BMI Mean (SD)24.99 (4.55)25.73 (4.59)0.151Waist circumference in cm Mean (SD)93.89 (9.99)93.41 (10.60)0.686Hypertension (syst. > 130 & diast. > 80) Normal blood pressure69 ($72.6\%$)330 ($77.5\%$)0.314 High blood pressure26 ($27.4\%$)96 ($22.5\%$)Cholesterol > = 6.2 mmol Normal cholesterol71 ($74.7\%$)283 ($66.4\%$)0.117 High cholesterol24 ($25.3\%$)143 ($33.6\%$)BMI categories Normal weight52 ($54.7\%$)200 ($46.9\%$)0.337 Overweight33 ($34.7\%$)164 ($38.5\%$) Obese10 ($10.5\%$)62 ($14.6\%$)NCD program participation No41 ($43.2\%$)94 ($22.1\%$) < 0.001 Prolanis19 ($20.0\%$)90 ($21.1\%$) Posbindu35 ($36.8\%$)242 ($56.8\%$)High diabetes distress No to moderate distress26 ($27.4\%$)115 ($27.0\%$)0.941 High distress69 ($72.6\%$)311 ($73.0\%$)Meets WHO recommended:MIC physical activity levels No36 ($58.1\%$)189 ($65.9\%$)0.245 Yes26 ($41.9\%$)98 ($34.1\%$)Has diabetes complications No complications14 ($16.1\%$)82 ($21.5\%$)0.367 Rethinopathy26 ($29.9\%$)121 ($31.7\%$) Kidney disease3 ($3.4\%$)6 ($1.6\%$) Impaired wound healing7 ($8.0\%$)22 ($5.8\%$) Neurological problems16 ($18.4\%$)77 ($20.2\%$) Cardio-vascular disease18 ($20.7\%$)51 ($13.4\%$) Diabetic foot3 ($3.4\%$)23 ($6.0\%$)Number of observations for WHO physical activity levels was 349, for diabetes complications 469. All other variables had no missing observations Only $20\%$ of men and women in our sample reported being free from diabetes-related complications. The most prevalent complication directly attributable to diabetes was retinopathy, with every third respondent reporting diabetes-related eye complications. Severe complications, such as neurological problems that impair pain sensitivity and slow wound healing, were reported more frequently, but there were fewer reports of very severe diabetes complications, such as diabetic foot and kidney disease. Measures of lifestyle and preventive behavior in this population showed that while men were still quite likely to achieve high levels of physical activity throughout the day, these rates were much lower for women. Women, however, were more likely to participate in NCD prevention and treatment programs. Results from the two-item diabetes distress scale showed that the majority of participants reported high levels of diabetes distress resulting from being overwhelmed by the disease and its management, and a feeling of failure in its management [7]. High diabetes distress has been related to adverse effects, including less successful disease management and depression [7]. ## Qualitative results This section provides the results of the qualitative interviews with Posbindu and Prolanis program managers and FGDs with diabetes patients. Qualitative data were analyzed based on the results of interviews with eight health officers and FGDs with 24 diabetes patients (16 women and 8 men) with a mean age of 52 (range of 28–74 years). The results of the data analysis are grouped into themes and sub-themes that emerged during the analysis. The two main themes were: [1] the difficulty of reaching and engaging patients and [2] experiences of diabetes patients. We grouped the sub-themes, consisting of challenges faced by health officers and patients, within these main themes. ## Theme 1: The difficulty of reaching and engaging patients The health officers responsible for Posbindu and Prolanis often mentioned that it was challenging to motivate patients to participate in the programs beyond basic laboratory tests (e.g. blood sugar, uric acid, and cholesterol blood tests).“I don’t know what to say…. most of the time, there is little interest in attending the Posbindu sessions because the participants find them boring. … They are less motivated to attend sessions with physical activity and health education… unless we come to the village (Posbindu post) with the drugs or blood sugar, cholesterol or uric acid tests… sometimes they do not want to attend if we only offer blood sugar tests… because they want to have another test (cholesterol, uric acid).” ( PAB1) Another challenge was to reach the working-age population, as activities took place during general working hours. “The problem is that we conduct the Posbindu activities during working hours, and therefore cannot reach those of working age but only elderly people… because those of working age are usually at work or attend school during the Posbindu activity in the village… we therefore cannot attend to many patients of younger age, despite our records showing that we have diabetes patients in their thirties….” ( PBA1) The Prolanis program required a minimum of 30 attendees per session. It was often difficult to reach this number, because people were unwilling to come to Puskesmas each time or they had other commitments. Puskesmas would then bring in other patients to meet BPJS Kesehatan’s minimum number of 30 diabetes and hypertension patients. This meant that, although the number of attendees was stable, the composition of the group was different for each session. “We have 30 diabetes and hypertension patients in our Prolanis club, however, the participants are constantly changing… because not all 30 participants attend the sessions every month… hence we keep changing the participants since the BPJS required at least 30 patients in each session.” ( PAB4) ## Theme 2: Experiences of diabetes patients Three focus group discussions (FGD) were conducted with diabetes patients in Banda Aceh and Aceh Besar to understand their experiences of living with diabetes. The respondents were predominantly women who reported homemaking as their occupation, and were mostly between 50 and 55 years of age. Many reported having experienced symptoms such as impaired wound healing or fatigue, frequent urination, and thirst that forced them to visit the hospital, where they were diagnosed with diabetes. “I found out that I had DM when I had a wound that never healed. It was in 2016, I could not walk so I used the walker, and I went to the hospital and my blood sugar level was 350.” P5AB2 Some received a diagnosis during pregnancy. Almost all respondents reported a family history of diabetes. Respondents reported difficulties with adhering to the medication treatment regime post-diagnosis, not taking medication (metformin or insulin) when blood sugars were perceived to be good, and taking the medication when they felt symptoms such as fatigue. Related to that, respondents reported that they were afraid of taking too much medication and believed that it would affect their kidneys, or that they could become addicted to insulin injections. “If I take the medicine regularly, my kidney will hurt, I am afraid of that. But if I feel fatigue, then I take it. Sometimes I boil water with pandan leaves and drink the water, I feel strong. But I have not done it so often.” ( P7BNA) Some reported replacing metformin pills with alternative natural products based on the belief that these would also lower blood sugar levels. “People said, pagar-pagar leaf is good, so if I drink these boiled leaves, then I don't take the pill anymore.” ( P1AB1) Respondents also reported efforts to change their diet by replacing rice or sugar with products perceived to have less effect on blood sugar levels. Some also tried to eat less but had problems maintaining their diet. “I substitute the sugar to Tropicana slim (low glucose sugar), I reduce my meal like I was told in Prolanis. We often feel hungry, so I prepare bread at 10 am. Sometimes we crave for something sweet, once we have it, it is hard to stop.” ( P7BNA) Some of the information provided by Prolanis is difficult for patients to comprehend. Many respondents found presentations and videos more engaging than brochures. In addition, many respondents had difficulties reading the information in the brochures. “Honestly, brochures we don't like, because we are lazy to read, it is better to have a PowerPoint presentation.” ( P5AB2)“Sometimes playing the video is also good, so it will not be so boring.” ( P3BA) ## Discussion Diabetes and other NCDs have become a pressing issue for health systems in LMICs. From our data on individuals with type 2 diabetes in Banda Aceh and Aceh Besar, it is evident that current measures to treat type 2 diabetes are not meeting their needs. We found that the average person with diabetes had HbA1c levels of almost twice the level that would be taken to indicate pre-diabetes. There is clear evidence that elevated HbA1c levels are highly predictive of diabetes-related morbidity and mortality [9–11], which is also reflected in the large number of people who reported an existing diabetes complication in our data. Similarly, we found high rates of obesity, hypertension and dyslipidemia, making it more likely that those without existing diabetes complications will develop complications in the near future. Given the high HbA1c levels indicative of largely unmanaged diabetes, even relatively minor changes to the treatment regime could have important effects. For example, a recent study in the same area evaluated the usability of novel smartphone-based test devices in Puskesmas (Rhode et al.: Smartphone-based point-of-care diagnostics in primary health care to monitor HbA1c levels in patients with diabetes: a validation study, unpublished manuscript). These could be a cheap and effective alternative to laboratory-based tests for HbA1c levels and lipid profiles which are currently not available at all Puskesmas, or are only provided irregularly, preventing their use for continuous monitoring to inform treatment decisions. The qualitative findings indicate that people with diabetes often only engage with the health system after experiencing diabetes symptoms, which is too late. There is evidence from another study carried out in the same region, that relatively simple interventions, such as SMS reminders to attend health screenings, can increase healthcare-seeking behavior [12]. Similar interventions have also been shown to improve diabetes care and might be used to increase healthcare seeking in people who have diabetes, or are at risk of developing diabetes, in our setting [13]. We further found that existing governmental programs suffer from low participation rates, low engagement of participants beyond the use of medical services offered during these meetings, and materials that are difficult for many participants to comprehend or read. One way to increase participation could be to align meeting times with the schedules of working-age adults. Another improvement could be to design meetings and information material to be interesting, accessible and actionable for participants. We find that patients find it difficult to adhere to recommended treatment and lifestyle advice and use alternative treatments to replace or complement their medical therapy. False beliefs about diabetes are also common in other settings [14, 15]. To address false beliefs, an intervention must respect local circumstances and beliefs, but has to overcome myths that interfere with treatment of patients who are in dire need of insulin but do not take it. Our mixed-methods approach reveals that structural problems are impeding the effectiveness of diabetes-management programs in Aceh, Indonesia. Relatively simple interventions based on scientific evidence, derived from other studies in the region, could be used to improve care-seeking behavior and the health information that is available to people with diabetes. ## References 1. Seiglie JA, Marcus M, Ebert C, Prodromidis N, Geldsetzer P, Theilmann M, Agoudavi K, Andall-Brereton G, Aryal KK, Bicaba BW, Bovet P, Brian G, Dorobantu M, Gathecha G, Singh Gurung M, Guwatudde D, Msaidie M, Houe-hanou C, Houinato D, Jorgensen JMA, Kagaruki GB, Karki KB, Labadarios D, Martins JS, Mayige MT, McClure RW, Mwangi JK, Mwalim O, Norov B, Quesnel-Crooks S, Silver BK, Sturua L, Tsabedze L, Wesseh CS, Stokes A, Atun R, Davies JI, Vollmer S, Bärnighausen T, Jaacks L, Meigs J, Wexler DJ, Manne-Goehler J. Diabetes prevalence and its relationship with education, wealth, and BMI in 29 low- and middle-income countries. Diabetes Care (2020.0) 43 767-775. DOI: 10.2337/dc19-1782 2. Manne-Goehler J, Geldsetzer P, Agoudavi K, Andall-Brereton G, Aryal KK, Bicaba BW, Bovet P, Brian G, Dorobantu M, Gathecha G, Gurung MS, Guwatudde D, Msaidie M, Houehanou C, Houinato D, Jorgensen JMA, Kagaruki GB, Karki KB, Labadarios D, Martins JS, Mayige MT, McClure RW, Mwalim O, Mwangi JK, Norov B, Quesnel-Crooks S, Silver BK, Sturua L, Tsabedze L, Wesseh CS, Stokes A, Marcus M, Ebert C, Davies JI, Vollmer S, Atun R, Bärnighausen T, Jaacks L. Health system performance for people with diabetes in 28 low- and middle-income countries: A cross-sectional study of nationally representative surveys. PLoS Med (2019.0) 16 e1002751. DOI: 10.1371/journal.pmed.1002751 3. Bommer C, Heesemann E, Sagalova V, Manne-Goehler J, Atun R, Bärnighausen T, Vollmer S. The Global Economic Burden of Diabetes in Adults aged 20 to 79: A Cost-of-Illness Study. Lancet Diabetes Endocrinol (2017.0) 5 423-430. DOI: 10.1016/S2213-8587(17)30097-9 4. 4.Direktorat P2P. (2019). Program Pencegahan dan Pengendalian Penyakit Tidak Menular (Programs for Non communicable diseases Prevention and Control). 5. 5.Ruby, M. (2021). NON-COMMUNICABLE DISEASE POLICY IN NATIONAL Non-Communicable Disease Policy in National Health Security Program Implementation. Banda Aceh - Indonesia: USK. 6. Seuring T, Marthoenis R. Using peer education to improve diabetes management and outcomes in a low-income setting: a randomized controlled trial. Trials (2019.0) 20 548. DOI: 10.1186/s13063-019-3656-1 7. Fisher L, Glasgow RE, Mullan JT, Skaff MM, Polonsky WH. Development of a brief diabetes distress screening instrument. Ann Fam Med (2008.0) 6 246-252. DOI: 10.1370/afm.842 8. 8.Pantalone, K. M., Misra-Hebert, A. D., Hobbs, T. M., Kong, S. X., Ji, X., Ganguly, R., … Kattan, M. W. (2020). The probability of a1c goal attainment in patients with uncontrolled type 2 diabetes in a large integrated delivery system: A prediction model. Diabetes Care. 43(8): 1910–1919. 9. Forbes A, Murrels T, Mulnier H, Sinclair AJ. Mean HbA1c, HbA1c variability, and mortality in people with diabetes aged 70 years and older: a retrospective cohort study. Lancet Diabetes Endocrinol (2018.0) 6 476-486. DOI: 10.1016/S2213-8587(18)30048-2 10. Luo M, Lim WY, Tan CS, Ning Y, Chia KS, van Dam RM, Tang WE, Tan NC, Chen R, Tai S, Venkataraman K. Longitudinal trends in HbA1c and associations with comorbidity and all-cause mortality in Asian patients with type 2 diabetes: A cohort study. Diabetes Res Clin Pract (2017.0) 133 69-77. DOI: 10.1016/j.diabres.2017.08.013 11. Raghavan S, Vassy JL, Ho Y-L, Song RJ, Gagnon DR, Cho K, Wilson P, Phillips LS. Diabetes Mellitus-Related All-Cause and Cardiovascular Mortality in a National Cohort of Adults. J Am Heart Assoc (2019.0) 8 e011295. DOI: 10.1161/JAHA.118.011295 12. 12.Marcus M, Reuter A, Rogge L, Vollmer S (2021) The Effect of SMS Reminders on Health Screening Uptake: A Randomized Experiment in Indonesia. Courant Research Centre ‘Poverty, Equity and Growth’ Discussion Paper No. 284. 13. 13.Huo X, Krumholz HM, Bai X, Spatz ES, Ding Q, Horak P, Zhao W, Gong Q, Zhang H, Yan X, Sun Y, Liu J, Wu X, Guan W, Wang X, Li J, Li X, Spertus JA, Masoudi FA, Zheng X (2019) Effects of Mobile Text Messaging on Glycemic Control in Patients With Coronary Heart Disease and Diabetes Mellitus: A Randomized Clinical Trial. Circulation: Cardiovascular Quality and Outcomes, 12(9) :e005805. 14. Pujilestari CU. “It is not possible for me to have diabetes”-community perceptions on diabetes and its risk factors in Rural Purworejo District, Central Java, Indonesia. Global J Health Sci (2014.0) 6 204-218. DOI: 10.5539/gjhs.v6n5p204 15. Widayanti AW, Heydon S, Norris P, Green JA. Lay perceptions and illness experiences of people with type 2 diabetes in Indonesia: a qualitative study. Health Psychol Behav Med (2020.0) 8 1-15. DOI: 10.1080/21642850.2019.1699101
--- title: 'Association between gestational weight gain and preterm birth and post-term birth: a longitudinal study from the National Vital Statistics System database' authors: - Yifang Zhu - Jiani Zhang - Qiaoyu Li - Min Lin journal: BMC Pediatrics year: 2023 pmcid: PMC10026488 doi: 10.1186/s12887-023-03951-0 license: CC BY 4.0 --- # Association between gestational weight gain and preterm birth and post-term birth: a longitudinal study from the National Vital Statistics System database ## Abstract ### Background To evaluate the association between gestational weight gain (GWG) and preterm birth and post-term birth. ### Methods This longitudinal-based research studied singleton pregnant women from the National Vital Statistics System (NVSS) [2019]. Total GWG (kg) was converted to gestational age-standardized z scores. The z-scores of GWG were divided into four categories according to the quartile of GWG, and the quantile 2 interval was used as the reference for the analysis. Univariate and multivariate logistic regression analyses were performed to investigate the association between GWG and preterm birth, post-term birth, and total adverse outcome (preterm birth + post-term birth). Subgroup analysis stratified by pre-pregnancy body mass index (BMI) was used to estimate associations between z-scores and outcomes. ### Results Of the 3,100,122 women, preterm birth occurred in $9.45\%$ [292,857] population, with post-term birth accounting for $4.54\%$ [140,851]. The results demonstrated that low GWG z-score [odds ratio (OR): 1.04, $95\%$ confidence interval (CI): 1.03 to 1.05, $P \leq 0.001$], and higher GWG z-scores (quantile 3: OR: 1.42, $95\%$ CI: 1.41 to 1.44, $P \leq 0.001$; quantile 4: OR: 2.79, $95\%$ CI: 2.76 to 2.82, $P \leq 0.001$) were positively associated with preterm birth. Low GWG z-score (OR: 1.18, $95\%$ CI: 1.16 to 1.19, $P \leq 0.001$) was positively associated with an increased risk of post-term birth. However, higher GWG z-scores (quantile 3: OR: 0.84, $95\%$ CI: 0.83 to 0.85, $P \leq 0.001$; quantile 4: 0.59, $95\%$ CI: 0.58 to 0.60, $P \leq 0.001$) was associated with a decreased risk of post-term birth. In addition, low GWG z-score and higher GWG z-scores were related to total adverse outcome. A subgroup analysis demonstrated that pre-pregnancy BMI, low GWG z-score was associated with a decreased risk of preterm birth among BMI-obesity women (OR: 0.96, $95\%$ CI: 0.94 to 0.98, $P \leq 0.001$). ### Conclusion Our result suggests that the management of GWG may be an important strategy to reduce the number of preterm birth and post-term birth. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12887-023-03951-0. ## Background Preterm birth is defined as birth before the completion of 37 weeks gestation, with one in 10 babies being born preterm, and every year, around 15 million babies are born preterm in the world, putting the global preterm birth rate at $11\%$ [1, 2]. Post-term is defined as a pregnancy that has extended to or beyond 42 weeks (294 days) from the first day of the last normal menstrual period or 14 days beyond the best obstetric estimate of the date of delivery [3]. Preterm birth and post-term birth are both the cause of perinatal mortality and severe morbidity [4, 5], and impose a considerable burden on health, education, and social services, as well as on families and caregivers [6]. Given both preterm and post-term births are associated with unfavorable maternal and neonatal outcomes, the identification of modifiable risk factors is of great importance for the prevention of adverse outcomes from preterm birth and post-term birth. Gestational weight gain (GWG) is necessary to ensure fetal health [7]. GWG reflects a variety of characteristics, including the accumulation of maternal fat, fluid swelling, and the growth of the fetus, placenta, and uterus [8]. Nevertheless, studies have found that excessive or insufficient GWG was associated with adverse outcomes [9–12]. Previous studies have identified an association between GWG and the risk of preterm birth [13, 14]. However, there is a limited study reporting the effect of GWG on post-term birth. Moreover, it is of particular importance to understand the relationships of GWG with outcomes of preterm and post-term birth combined, and to develop a reasonable pregnancy weight control plan to help to reduce the likelihood of both preterm birth and post-term birth. The purpose of this study was to examine the associations between GWG and preterm birth, post-term birth, and the combined outcome of preterm and post-term birth; investigate the effect of GWG on preterm birth, post-term birth, and the combined outcome of preterm and post-term birth among different pregnancy body mass index (BMI) women. ## Study design and population This longitudinal study recruited pregnant women from the National Vital Statistics System (NVSS) [2019], which is a U.S. population-based retrospective cohort study from 50 States and the District of Columbia [15]. The NVSS is a major cooperative effort between the U.S. Centers for Disease Control and Prevention (CDC) and all U.S. states, which gathers information on maternal exposures before and during pregnancy and infant outcomes at delivery using two uniform documents: a facility worksheet and a maternal worksheet. Detailed methods, quality control, and vital statistics can be found on the CDC website (https://www.cdc.gov/nchs/nvss/births.htm). We included singleton pregnant women aged 18 years or older. Women with pre-pregnancy hypertension or diabetes, height < 55 inches, gestation at < 22 weeks or > 44 weeks, fetal malformations, chromosomal disorders, and pregnant women who used assisted reproduction were excluded. The de-identified data are publicly available online, so the ethical board review of the corresponding author's institution is exempted. ## Definitions Maternal GWG was classified by GWG z-score of standardized maternal weight gain in gestational age, which was calculated as: (observed total weight gain- mean week-specific weight gain)/standard deviation of week-specific weight gain, with week-specific means and standard deviations [16]. The z-scores of GWG were divided into four categories according to the quartile of GWG, and the quantile 2 interval was used as the reference for analysis. Quantile 1 was 0.284, the Median was 0.439, and Quantile 3 was 0.573 of the z-scores levels of GWG. The maternal GWG ranges for different gestational ages based on the calculation of z-scores are shown in the Supplementary Table 1. Gestational age was determined based on the last menstrual period [17]. We classified maternal pre-pregnancy BMI as underweight (BMI < 18.5 kg/m2), normal weight (18.5–24.9 kg/m2), overweight (25–29.9 kg/m2), and obesity (≥ 30 kg/m2) [18]. ## Potential covariates Potential covariates included the number of prenatal visits, maternal age at pregnancy, smoking before pregnancy, prior other terminations, previous preterm birth, multipara or not, maternal race, the special supplemental nutrition program for women, infants, and children (WIC) food during pregnancy, maternal education, smoking during pregnancy. Maternal race was divided into White, Black, Asian, and other. The maternal education level was divided into < 12 grade, high school or general educational development (GED), some college, bachelor or above. A multipara is a woman who has given birth to more than one child. ## Outcomes Preterm birth referred to < 37 weeks of gestation, term birth was > = 37–42 weeks of gestation, and post-term birth was defined as > 42 weeks of gestation [19]. ## Statistical analysis The normally distributed measurement data were expressed as mean +—standard deviation, and the one-way analysis of variance (ANOVA) was used for comparison between groups; abnormally distributed measurement data were described as median and quartile [M (Q1, Q3)], and the comparison between groups was conducted by Kruskal–Wallis test. The enumeration data were compared using the chi-square test or Fisher’s exact test, manifesting as cases and the constituent ratio (n (%)). For the handling of missing data, the simple deletion method was used to delete the cases with missing values, sensitivity analysis before and after data deletion was performed. Comparison between groups showed the characteristics of the study population with preterm birth, term birth, and post-term birth. With preterm birth and post-term birth as outcomes, the incidence of preterm birth and post-term birth in different GWG z-scores were calculated. We combined the preterm birth and post-term birth as a new outcome (total adverse outcome), and calculated the incidence of total adverse outcome in different GWG z-scores. To determine which confounders required adjustment, directed acyclic graph (DAG) were drawn (Supplementary Fig. 1). Univariate multivariate logistic regression model was used to explore the effect of GWG (weight gain z-score) on different gestational age groups. Model 1 was an unadjusted model, model 2 adjusted for number of prenatal visits, maternal age at pregnancy, smoking before pregnancy, prior other terminations, previous preterm birth, multipara or not, maternal race, WIC food during pregnancy, maternal education, and smoking during pregnancy. Subgroup analysis was stratified by early-pregnancy BMI to investigate the effects of GWG z-scores on preterm birth and post-term birth, and total adverse outcome. All statistical tests were two-sided and a p-value < 0.05 was considered statistically significant. All statistical analyses were completed using Statistical Analysis Software version 9.4 (SAS Institute Inc.). ## Characteristics of the study population Following our exclusions, 3,100,122 women with live singleton births were included in the analysis (Fig. 1). Preterm birth occurred in $9.45\%$ [292,857] population, with post-term birth accounting for $4.54\%$ [140,851]. In our study, $3.08\%$ [95,608] women were underweight, $41.87\%$ [1,297,882] normal weight, $27.23\%$ [844,091] were overweight, and $27.82\%$ [862,541] obese. Over half of our study was White ($74.19\%$); Black, Asian, and other races constituted $15.21\%$, $6.68\%$, and $3.91\%$ respectively. The mean GWG for women with preterm birth was 11.35 (7.26, 15.89) kg, and the mean GWG for women with post-term birth was 13.62 (9.08, 17.71) kg. Table 1 shows the baseline characteristics of the study population. Fig. 1The flow diagram of participants selectionTable 1Characteristics of the study populationVariablesTotal ($$n = 3$$,100,122)Preterm birth ($$n = 292$$,857)Term birth ($$n = 2$$,666,414)Post-term birth ($$n = 140$$,851)Statistics P Maternal age at pregnancy, years, Mean ± SD29.05 ± 5.6029.11 ± 5.9729.09 ± 5.5628.07 ± 5.47F = 2267.709< 0.001Maternal race, n (%)χ2 = 11,577.64< 0.001 White2,300,082 (74.19)199,429 (68.10)1,993,838 (74.78)106,815 (75.84) Black471,532 (15.21)63,396 (21.65)387,968 (14.55)20,168 (14.32) Asian207,156 (6.68)17,141 (5.85)182,473 (6.84)7542 (5.35) Others121,352 (3.91)12,891 (4.40)102,135 (3.83)6326 (4.49)Maternal education, n (%)χ2 = 16,491.96< 0.001 < 12 grade347,915 (11.22)44,010 (15.03)285,328 (10.70)18,577 (13.19) High school or GED819,024 (26.42)89,561 (30.58)687,805 (25.80)41,658 (29.58) Some college884,826 (28.54)84,866 (28.98)758,856 (28.46)41,104 (29.18) Bachelor or above1,048,357 (33.82)74,420 (25.41)934,425 (35.04)39,512 (28.05)Pre-pregnancy BMI, n (%)χ2 = 2585.980< 0.001 Underweight95,608 (3.08)10,951 (3.74)80,502 (3.02)4155 (2.95) Normal1,297,882 (41.87)113,921 (38.90)1,127,107 (42.27)56,854 (40.36) Overweight844,091 (27.23)77,880 (26.59)728,318 (27.31)37,893 (26.90) Obesity862,541 (27.82)90,105 (30.77)730,487 (27.40)41,949 (29.78)GWG, kg, M (Q1, Q3)13.17 (9.08, 17.25)11.35 (7.26, 15.89)13.17 (9.08, 17.25)13.62 (9.08, 17.71)χ2 = 15,994.35#< 0.001GWG z-score, M (Q1, Q3)0.44 (0.28, 0.57)0.52 (0.35, 0.69)0.44 (0.28, 0.57)0.40 (0.23, 0.52)χ2 = 43,068.24#< 0.001Smoking before pregnancy, n (%)χ2 = 3930.086< 0.001 No2,862,009 (92.32)262,842 (89.75)2,471,714 (92.70)127,453 (90.49) Yes238,113 (7.68)30,015 (10.25)194,700 (7.30)13,398 (9.51)Smoking during pregnancy, n (%)χ2 = 4932.560 < 0.001 No2,941,838 (94.89)270,492 (92.36)2,539,377 (95.24)131,969 (93.69) Yes158,284 (5.11)22,365 (7.64)127,037 (4.76)8882 (6.31)Number of prenatal visits, M (Q1, Q3)12.00 (9.00, 13.00)10.00 (7.00, 12.00)12.00 (10.00, 13.00)12.00 (10.00, 14.00)χ2 = 68,358.24#< 0.001Prior other terminations, n (%)χ2 = 1352.827< 0.001 No2,258,968 (72.87)205,929 (70.32)1,947,264 (73.03)105,775 (75.10) Yes841,154 (27.13)86,928 (29.68)719,150 (26.97)35,076 (24.90)History of preterm birth, n (%)χ2 = 31,201.89< 0.001 No2,995,626 (96.63)266,590 (91.03)2,591,520 (97.19)137,516 (97.63) Yes104,496 (3.37)26,267 (8.97)74,894 (2.81)3335 (2.37)WIC food during pregnancy, n (%)χ2 = 5200.329< 0.001 No2,064,986 (66.61)179,731 (61.37)1,796,666 (67.38)88,589 (62.90) Yes1,035,136 (33.39)113,126 (38.63)869,748 (32.62)52,262 (37.10)Gestational age, Mean ± SD38.73 ± 2.0134.38 ± 2.1839.00 ± 1.1142.58 ± 0.74F = 2,563,966< 0.001Gender of newborn, n (%)χ2 = 801.768< 0.001 Female1,513,848 (48.83)136,304 (46.54)1,306,395 (48.99)71,149 (50.51) Male1,586,274 (51.17)156,553 (53.46)1,360,019 (51.01)69,702 (49.49)Multipara, n (%)χ2 = 1769.685< 0.001 No1,178,778 (38.02)104,300 (35.61)1,015,028 (38.07)59,450 (42.21) Yes1,921,344 (61.98)188,557 (64.39)1,651,386 (61.93)81,401 (57.79)Eclampsia, n (%)χ2 = 4606.815< 0.001 No3,092,573 (99.76)290,422 (99.17)2,661,519 (99.82)140,632 (99.84) Yes7549 (0.24)2435 (0.83)4895 (0.18)219 (0.16)Hypertension during pregnancy, n (%)χ2 = 30,026.00< 0.001 No2,865,842 (92.44)247,322 (84.45)2,485,062 (93.20)133,458 (94.75) Yes234,280 (7.56)45,535 (15.55)181,352 (6.80)7393 (5.25) GED general educational development, BMI body mass index, GWG gestational weight gain, WIC special supplemental nutrition program for women, infants, and children, χ2 chi-square test, F analysis of variance, SD standard deviation, M Median, Q1 1st Quartile, Q3 3rd Quartile ## Incidences of preterm birth, post-term birth, and total adverse outcome in different GWG z-score intervals U-shaped relations have been observed between the GWG z-score and preterm birth. The relationship between GWG z-score and incidence of preterm birth is depicted in Fig. 2a and b. The results showed an approximately inverted U-shape between GWG z-score and post-term birth. The incidence of post-term birth in different GWG z-score intervals is depicted in Fig. 3a and b. Similar to the association between GWG z-score and preterm birth, the association of total GWG z-scores with total adverse outcome tended to be U-shaped. The relationship between GWG z-score and incidence of total adverse outcome is shown in Fig. 4a and b.Fig. 2Incidences of preterm birth in different GWG z-score intervals; a total; b subgroup analysis of pre-pregnancy BMIFig. 3Incidences of post-term birth in different GWG z-score intervals; a total; b subgroup analysis of pre-pregnancy BMIFig. 4Incidences of total adverse outcome in different GWG z-score intervals; a total; b subgroup analysis of pre-pregnancy BMI ## Associations of GWG with preterm birth, post-term birth, and total adverse outcome The result demonstrated that the weight-gain z-score in quantile 1 [odds ratio (OR): 1.04, $95\%$ confidence interval (CI): 1.03 to 1.05, $P \leq 0.001$], quantile 3 (OR: 1.42, $95\%$ CI: 1.41 to 1.44, $P \leq 0.001$), quantile 4 (OR: 2.79, $95\%$ CI: 2.76 to 2.82, $P \leq 0.001$), and total weight-gain z-score (OR: 5.46, $95\%$ CI: 5.37 to 5.55, $P \leq 0.001$) was associated with an increased risk of preterm birth. As for the post-term birth, the weight-gain z-score in quantile 1 was related to an increased risk of post-term birth (OR: 1.18, $95\%$ CI: 1.16 to 1.19,$P \leq 0.001$). Nevertheless, weight-gain z-scores in 3 (OR: 0.84, $95\%$ CI: 0.83 to 0.85, $P \leq 0.001$), quantile 4 (0.59, $95\%$ CI: 0.58 to 0.60, $P \leq 0.001$), and total weight-gain z-score (OR: 0.49, $95\%$ CI: 0.48 to 0.50, $P \leq 0.001$) related to a decreased risk of post-term birth. Concerning the total adverse outcome, weight-gain z-scores in quantile 1 (OR: 1.06, $95\%$ CI: 1.05 to 1.07, $P \leq 0.001$), quantile 3 (OR: 1.17, $95\%$ CI: 1.16 to 1.18, $P \leq 0.001$), quantile 4 (OR: 1.77, $95\%$ CI: 1.75 to 1.78, $P \leq 0.001$), and total weight-gain z-score (OR: 2.31, $95\%$ CI: 2.28 to 2.34, $P \leq 0.001$) were all associated with an increased risk of total adverse outcome. Associations of GWG with preterm birth, post-term birth, and total adverse outcome are presented in Table 2.Table 2Associations of GWG with preterm birth, post-term birth, and total adverse outcomePreterm birthPost-term birthTotal adverse outcomeModel 1Model 2Model 1Model 2Model 1Model 2VariablesOR ($95\%$ CI) P OR ($95\%$ CI) P OR ($95\%$ CI) P OR ($95\%$ CI) P OR ($95\%$ CI) P OR ($95\%$ CI) P Levels of weight-gain-for-gestational-age z-scores Quantile 11.24 (1.22–1.25)< 0.0011.04 (1.03–1.05)< 0.0011.19 (1.17–1.21)< 0.0011.18 (1.16–1.19)< 0.0011.19 (1.18–1.20)< 0.0011.06 (1.05–1.07)< 0.001 Quantile 2RefRefRefRefRefRef Quantile 31.34 (1.32–1.35)< 0.0011.42 (1.41–1.44)< 0.0010.85 (0.84–0.86)< 0.0010.84 (0.83–0.85)< 0.0011.13 (1.12–1.14)< 0.0011.17 (1.16–1.18)< 0.001 Quantile 42.68 (2.65–2.71)< 0.0012.79 (2.76–2.82)< 0.0010.62 (0.61–0.63)< 0.0010.59 (0.58–0.60)< 0.0011.77 (1.76–1.79)< 0.0011.77 (1.75–1.78)< 0.001Weight-gain-for-gestational-age z-scores4.26 (4.19–4.33)< 0.0015.46 (5.37–5.55)< 0.0010.51 (0.50–0.52)< 0.0010.49 (0.48–0.50)< 0.0011.93 (1.90–1.95)< 0.0012.31 (2.28–2.34)< 0.001Model 1: unadjusted model; Model 2 adjusted for number of prenatal visits, maternal age at pregnancy, smoking before pregnancy, prior other terminations, previous preterm birth, multipara or not, maternal race, WIC food during pregnancy, maternal education, smoking during pregnancy, pre-pregnancy BMI GWG gestational weight gain, Ref Reference, OR odds ratio, CI confidence interval ## Associations of GWG with preterm birth, post-term birth, and total adverse outcome in different pre-pregnancy BMI populations Associations of GWG with preterm birth, post-term birth, and total adverse outcome in different pre-pregnancy BMI populations are shown in Table 3. Regarding preterm birth, weight-gain z-scores in quantile 1, quantile 3, quantile 4, and total weight-gain z-scores were all associated with an increased risk of preterm birth in women who were underweight, normal, and overweight. However, in women who were obese, the weight-gain z-score in quantile 1 was related to a lower risk of preterm birth (OR: 0.96, $95\%$ CI: 0.94 to 0.98, $P \leq 0.001$) while weight-gain z-scores in quantile 2, in quantile 4, and total weight-gain z-scores were associated with a higher risk of preterm birth. Table 3Associations of GWG with preterm birth, post-term birth, and total adverse outcome in different pre-pregnancy BMI populationsSubgroupsOR ($95\%$ CI) P OR ($95\%$ CI) P OR ($95\%$CI) P Underweight ($$n = 84$$,657) Levels of weight-gain-for-gestational-age z-scores Quantile 11.27 (1.16–1.38)< 0.0011.28 (1.16–1.42)< 0.0011.23 (1.15–1.31)< 0.001 Quantile 2RefRefRef Quantile 31.47 (1.38–1.56)< 0.0010.85 (0.79–0.92)< 0.0011.23 (1.17–1.29)< 0.001 Quantile 42.88 (2.72–3.06)< 0.0010.60 (0.55–0.65)< 0.0011.90 (1.81–1.99)< 0.001 Weight-gain-for-gestational-age z-scores11.77 (10.51–13.17)< 0.0010.31 (0.26–0.36)< 0.0014.30 (3.90–4.73)< 0.001 Normal ($$n = 1$$,183,961) Levels of Weight-gain-for-gestational-age z-scores Quantile 11.17 (1.15–1.20)< 0.0011.19 (1.16–1.22)< 0.0011.13 (1.11–1.15)< 0.001 Quantile 2RefRefRef Quantile 31.46 (1.43–1.49)< 0.0010.80 (0.78–0.82)< 0.0011.14 (1.12–1.16)< 0.001 Quantile 43.07 (3.02–3.13)< 0.0010.56 (0.54–0.57)< 0.0011.78 (1.75–1.80)< 0.001 Weight-gain-for-gestational-age z-scores11.88 (11.49–12.28)< 0.0010.33 (0.32–0.34)< 0.0013.31 (3.22–3.40)< 0.001 Overweight ($$n = 766$$,211) Levels of Weight-gain-for-gestational-age z-scores Quantile 11.00 (0.98–1.03)0.9781.18 (1.15–1.21)< 0.0011.04 (1.02–1.06)< 0.001 Quantile 2RefRefRef Quantile 31.40 (1.37–1.44)< 0.0010.88 (0.85–0.90)< 0.0011.17 (1.15–1.20)< 0.001 Quantile 42.64 (2.59–2.70)< 0.0010.62 (0.60–0.64)< 0.0011.72 (1.69–1.75)< 0.001 Weight-gain-for-gestational-age z-scores5.71 (5.53–5.90)< 0.0010.50 (0.48–0.52)< 0.0012.37 (2.31–2.43)< 0.001 Obesity ($$n = 772$$,436) Levels of Weight-gain-for-gestational-age z-scores Quantile 10.96 (0.94–0.98)< 0.0011.18 (1.16–1.21)< 0.0011.02 (1.01–1.04)0.011 Quantile 2RefRefRef Quantile 31.44 (1.41–1.47)< 0.0010.89 (0.86–0.92)< 0.0011.23 (1.20–1.25)< 0.001 Quantile 42.55 (2.50–2.61)< 0.0010.62 (0.60–0.64)< 0.0011.78 (1.75–1.81)< 0.001 Weight-gain-for-gestational-age z-scores3.30 (3.23–3.38)< 0.0010.59 (0.58–0.61)< 0.0011.81 (1.78–1.85)< 0.001The model adjusted number of prenatal visits, maternal age at pregnancy, smoking before pregnancy, prior other terminations, previous preterm birth, multipara or not, maternal race, WIC food during pregnancy, maternal education, smoking during pregnancy GWG gestational weight gain, BMI body mass index, Ref Reference, OR odds ratio, CI confidence interval As for the post-term birth, weight-gain z-score in quantile 1 was related to an increased risk of post-term birth in women who were underweight (OR: 1.28, $95\%$ CI: 1.16 to 1.42, $P \leq 0.001$), normal (OR: 1.19, $95\%$ CI: 1.16 to 1.22, $P \leq 0.001$), overweight (OR: 1.18, $95\%$ CI: 1.15 to 1.21, $P \leq 0.001$), and obesity (OR: 1.18, $5\%$ CI: 1.16 to 1.21, $P \leq 0.001$), however, weight-gain z-scores in quantile 3, quantile 4, and total weight-gain z-scores were all associated with a decreased risk of post-term birth in women who were underweight, normal, overweight, and obesity. In terms of total adverse outcome, weight-gain z-scores in quantile 1, quantile 3, quantile 4, and total weight-gain z-scores were all associated with total adverse outcome among underweight, normal, overweight, and obese women. ## Discussion In this large population-based study of more than a million women with live singleton births in the U.S., we found that low GWG z-score and higher GWG z-scores were positively associated with preterm birth. A low GWG z-score was positively associated with an increased risk of post-term birth. However, higher GWG z-score (excessive GWG) was associated with a decreased risk of post-term birth. In addition, a low GWG z-score and higher GWG z-scores were related to total adverse outcome. When stratified by pre-pregnancy BMI, low GWG z-score was associated with a decreased t risk of preterm birth among BMI-obesity women. In this study, we found that low GWG z-score and higher GWG z-scores were positively associated with preterm birth. In accordance with our findings, a study by Santos et al. reported that both lower and higher total gestational weight gain z-scores were associated with a higher risk of preterm birth [20]. A study compared GWG z-scores and traditional weight gain measures in relation to perinatal outcomes found that both low and high z-scores were associated with preterm birth at 32 and 37 weeks of gestation, however, the effect of preterm birth observed with low weight gain was significantly weaker than that observed using total weight gain [21]. In our study, we observed similar associations of low and high GWG z-scores with a greater risk of preterm birth in women with underweight BMI, normal BMI, and overweight BMI, while only low GWG z-scores were associated with a higher risk of preterm birth in women with obesity BMI. We speculate that this finding may be due to the small range of o low GWG s-score among obese persons in our study. Using the z-score, Leonard et al. found that low and high weight gain was associated with an increased risk of preterm birth and that the optimal range of weight gain with minimal risk of preterm birth decreased with increasing severity of pre-pregnancy overweight/obesity [22]. The different association also informs healthcare planning and commissioning of services, as the level of GWG required to prevent adverse outcomes associated with preterm birth will differ according to BMI classification. Postdate pregnancy represents a circumstance in which labor does not occur within the physiological term of gestation; this prolongation suggests an alteration in the physiological processes regulating the onset of labor, thus representing a potential risk factor for the fetus [6]. In addition to the relationship between pre-pregnancy BMI and post-term birth [23], we found that a low GWG z-score was associated with an increased risk of post-term birth, nevertheless, higher GWG z-scores were related to a decreased risk of post-term birth. A retrospective cohort study of term, singleton births has demonstrated that GWG increases the risk of a post-term delivery [24]. Denison et al. reported that a greater increase in maternal BMI between the first and third trimesters was also associated with longer gestation [25]. Although our findings indicated that elevated GWG z-scores were associated with a decreased risk of post-term birth, excessive GWG is both associated with an increased risk of complications during pregnancy and childbirth [26]. Thereby, when considering appropriate GWG to reduce the risk of post-term delivery, other adverse outcomes caused by GWG need to be taken into account. Moreover, in this study, low GWG z-score and higher GWG z-scores were associated with both preterm birth and post-term birth. Further studies to estimate the association between the range of perceived ideal GWG with fewer pregnancy outcomes are needed. This study was a longitudinal-based research study and a larger sample frame that can better understand and seek the association between GWG and preterm birth and post-term birth. However, this study has some limitations. The main limitation is the retrospective nature of this study as the collected data depended entirely on the available data. In addition, although multivariable analyses were utilized to minimize the effect of confounders, potentially unknown or unidentified confounders may exist. Thereby, the associations between GWG and preterm birth, post-term birth, and an adverse outcome combining preterm birth and post-term birth need further studies. ## Conclusions GWG was associated with preterm and post-term birth outcomes. In clinical practice, pregnant women should be guided to have a clear understanding of weight gain, regular check-ups during pregnancy, reasonable weight control, and reducing the risk of adverse neonatal outcomes from preterm and post-term birth. ## Supplementary Information Additional file 1.Additional file 2. ## References 1. Blencowe H, Cousens S, Oestergaard MZ, Chou D, Moller AB, Narwal R. National, regional, and worldwide estimates of preterm birth rates in the year 2010 with time trends since 1990 for selected countries: a systematic analysis and implications. Lancet (London, England) (2012.0) 379 2162-2172. DOI: 10.1016/S0140-6736(12)60820-4 2. Li H, Huang Q, Liu Y, Garmire LX. Single cell transcriptome research in human placenta. Reproduction (Cambridge, England) (2020.0) 160 R155-R167. DOI: 10.1530/REP-20-0231 3. 146: Management of late-term and postterm pregnancies. Obstet Gynecol (2014.0) 124 390-396. DOI: 10.1097/01.AOG.0000452744.06088.48 4. Pierrat V, Marchand-Martin L, Arnaud C, Kaminski M, Resche-Rigon M, Lebeaux C. Neurodevelopmental outcome at 2 years for preterm children born at 22 to 34 weeks' gestation in France in 2011: EPIPAGE-2 cohort study. BMJ (Clinical research ed) (2017.0) 358 j3448. DOI: 10.1136/bmj.j3448 5. Norman JE, Norrie J, MacLennan G, Cooper D, Whyte S, Chowdhry S. Evaluation of the Arabin cervical pessary for prevention of preterm birth in women with a twin pregnancy and short cervix (STOPPIT-2): An open-label randomised trial and updated meta-analysis. PLoS Med (2021.0) 18 e1003506. DOI: 10.1371/journal.pmed.1003506 6. Coyle K, Quan AML, Wilson LA, Hawken S, Bota AB, Coyle D. Cost-effectiveness of a gestational age metabolic algorithm for preterm and small-for-gestational-age classification. American journal of obstetrics & gynecology MFM (2021.0) 3 100279. DOI: 10.1016/j.ajogmf.2020.100279 7. Zhou YB, Liu JM. Optimal gestational weight gain. The Lancet regional health Western Pacific (2021.0) 13 100207. DOI: 10.1016/j.lanwpc.2021.100207 8. Gaillard R. Maternal obesity during pregnancy and cardiovascular development and disease in the offspring. Eur J Epidemiol (2015.0) 30 1141-1152. DOI: 10.1007/s10654-015-0085-7 9. Goldstein RF, Abell SK, Ranasinha S, Misso ML, Boyle JA, Harrison CL. Gestational weight gain across continents and ethnicity: systematic review and meta-analysis of maternal and infant outcomes in more than one million women. BMC Med (2018.0) 16 153. DOI: 10.1186/s12916-018-1128-1 10. Champion ML, Harper LM. Gestational Weight Gain: Update on Outcomes and Interventions. Curr DiabRep (2020.0) 20 11 11. Simko M, Totka A, Vondrova D, Samohyl M, Jurkovicova J, Trnka M. Maternal Body Mass Index and Gestational Weight Gain and Their Association with Pregnancy Complications and Perinatal Conditions. Int J Environ Res Public Health (2019.0) 16 1751. DOI: 10.3390/ijerph16101751 12. Foratori-Junior GA, Jesuino BG, Caracho RA, Orenha ES, Groppo FC, Sales-Peres SHC. Association between excessive maternal weight, periodontitis during the third trimester of pregnancy, and infants' health at birth. Journal of applied oral science : revista FOB (2020.0) 28 e20190351. DOI: 10.1590/1678-7757-2019-0351 13. Kominiarek MA, Saade G, Mele L, Bailit J, Reddy UM, Wapner RJ. Association Between Gestational Weight Gain and Perinatal Outcomes. Obstet Gynecol (2018.0) 132 875-881. DOI: 10.1097/AOG.0000000000002854 14. Heusschen L, Krabbendam I, van der Velde JM, Deden LN, Aarts EO, Merién AER. A Matter of Timing-Pregnancy After Bariatric Surgery. Obes Surg (2021.0) 31 2072-2079. DOI: 10.1007/s11695-020-05219-3 15. Liu B, Xu G, Sun Y, Du Y, Gao R, Snetselaar LG. Association between maternal pre-pregnancy obesity and preterm birth according to maternal age and race or ethnicity: a population-based study. Lancet Diabetes Endocrinol (2019.0) 7 707-714. DOI: 10.1016/S2213-8587(19)30193-7 16. Hutcheon JA, Bodnar LM. Good Practices for Observational Studies of Maternal Weight and Weight Gain in Pregnancy. Paediatr Perinat Epidemiol (2018.0) 32 152-160. DOI: 10.1111/ppe.12439 17. Arach AAO, Tumwine JK, Nakasujja N, Ndeezi G, Kiguli J, Mukunya D. Perinatal death in Northern Uganda: incidence and risk factors in a community-based prospective cohort study. Glob Health Action (2021.0) 14 1859823. DOI: 10.1080/16549716.2020.1859823 18. Bastola K, Koponen P, Härkänen T, Luoto R, Gissler M, Kinnunen TI. Pregnancy complications in women of Russian, Somali, and Kurdish origin and women in the general population in Finland. Womens Health (Lond Engl) (2020.0) 16 1745506520910911. DOI: 10.1177/1745506520910911 19. Zielinska MA, Rust P, Masztalerz-Kozubek D, Bichler J, Hamułka J. Factors Influencing the Age of Complementary Feeding-A Cross-Sectional Study from Two European Countries. Int J Environ Res Public Health (2019.0) 16 3799. DOI: 10.3390/ijerph16203799 20. 20.Santos S, Voerman E, Amiano P, Barros H, Beilin LJ, Bergström A, et al. Impact of maternal body mass index and gestational weight gain on pregnancy complications: an individual participant data meta-analysis of European, North American and Australian cohorts. BJOG. 2019; 126(8): 984–995. 21. Bodnar LM, Hutcheon JA, Parisi SM, Pugh SJ, Abrams B. Comparison of gestational weight gain z-scores and traditional weight gain measures in relation to perinatal outcomes. Paediatr Perinat Epidemiol (2015.0) 29 11-21. DOI: 10.1111/ppe.12168 22. 22.Leonard SA, Hutcheon JA, Bodnar LM, Petito LC, Abrams B. Gestational Weight Gain-for-Gestational Age Z-Score Charts Applied across U.S. Populations. Paediatric Perinatal Epidemiol. 2018;32(2):161–171. 23. Qazi Q, Liaqat N, Hussain SS, Syed W. Association of high Body Mass Index and postdates pregnancy. Pakistan journal of medical sciences (2022.0) 38 1371-1375. DOI: 10.12669/pjms.38.5.4815 24. Halloran DR, Cheng YW, Wall TC, Macones GA, Caughey AB. Effect of maternal weight on postterm delivery. J Perinatol: official journal of the California Perinatal Association (2012.0) 32 85-90. DOI: 10.1038/jp.2011.63 25. Denison FC, Price J, Graham C, Wild S, Liston WA. Maternal obesity, length of gestation, risk of postdates pregnancy and spontaneous onset of labour at term. BJOG : an international journal of obstetrics and gynaecology (2008.0) 115 720-725. DOI: 10.1111/j.1471-0528.2008.01694.x 26. Goldstein RF, Abell SK, Ranasinha S, Misso M, Boyle JA, Black MH. 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
--- title: FGF-21 and GDF-15 are increased in migraine and associated with the severity of migraine-related disability authors: - Jiahui He - Mengting Zhou - Fanglin Zhao - Hongrong Cheng - Hao Huang - Xiaopei Xu - Jian Han - Wenwu Hong - Faming Wang - Yujin Xiao - Jinjin Xia - Kaiming Liu journal: The Journal of Headache and Pain year: 2023 pmcid: PMC10026504 doi: 10.1186/s10194-023-01563-8 license: CC BY 4.0 --- # FGF-21 and GDF-15 are increased in migraine and associated with the severity of migraine-related disability ## Abstract ### Background Migraine is a prevalent disorder with significant socioeconomic impact. The impairment of metabolic homeostasis in migraine warrants further investigation. Changes in serum levels of Fibroblast-growth-factor 21 (FGF-21) and Growth-differentiation-factor 15 (GDF-15) are characteristic of some metabolic and mitochondrial diseases. This study aimed to assess whether the presence of migraine affects serum levels of FGF-21 and GDF-15, and taking metabolic disorders into account as potential confounding factors. ### Methods We collected serum samples from 221 migraine patients (153 episodic migraineurs and 68 chronic migraineurs) and 124 healthy controls. The serum concentrations of FGF-21 and GDF-15 were measured using an enzyme-linked immunosorbent assay (ELISA) based approach. Clinical variables, including monthly headache days, peak headache pain intensity, the 6-item Headache Impact Test (HIT-6), and the Migraine Disability Assessment (MIDAS), were also addressed. The associations between the clinical variables of migraine patients and serum levels of FGF-21 and GDF-15 were studied. ### Results In the multiple regression that corrected for age, we found that the serum levels of FGF-21 and GDF-15 were significantly higher in migraine sufferers than in healthy controls. A significant elevation in serum concentration of FGF-21, but not GDF-15, was observed in patients with chronic migraine (CM) compared to those with episodic migraine (EM). Regarding migraine-related disability, higher scores on the HIT-6 and MIDAS were associated with higher levels of FGF-21 and GDF-15. For the receiver operating characteristic (ROC) analysis, the diagnosis of migraine using GDF-15 showed that the area under the ROC curve (AUC) was 0.801 and the AUC of chronic migraine was 0.880. ### Conclusion Serum GDF-15 and FGF-21 levels are increased in patients with migraine and associated with the severity of migraine-related disability. ## Background Migraine is a widespread neurological disorder that affects a substantial portion of the global population, with an estimated prevalence of over $15\%$. As a primary headache disorder that is debilitating in nature, it is ranked second in terms of the number of disability-adjusted life years it causes globally, and holds the top position among the causes of such years among young women [1]. The diagnosis of migraine is established through a comprehensive evaluation of the patient's clinical presentation, which includes an assessment of their medical history, headache duration, accompanying symptoms, and response to treatment. However, the diagnostic criteria have limitations as they do not fully capture the heterogeneity of migraine, including the potential role of metabolic abnormalities as contributing factors. Evidence supports the idea that mitochondrial dysfunction and an imbalance between energy supply and demand may play a role in the pathophysiology and susceptibility to migraine [2]. Certain triggers, such as skipping meals or fasting, excessive exercise [3–5], dehydration, hypoxia [6–8], and lack of sleep [9], have been shown to have a clear link to metabolism. Intense sensory stimuli, including odors [10], phthalate-containing perfumes [11], blue light [12], and loud noises [13], can contribute to an increase in oxidative stress. Neuroimaging studies have revealed impairments in mitochondrial oxidative phosphorylation, reduced levels of ATP, and elevated lactate levels in the brains of patients with migraine [14, 15]. During a migraine attack, it has been observed that the brain's energy homeostasis is restored and harmful oxidative stress levels are reduced [2]. Studies have shown that oxidative mitochondrial metabolism is impaired in migraine patients including platelet mitochondrial dysfunction and production of peripheral markers of oxidative stress [16]. Okada et al. reported higher lactic acid levels in migraine patients [17], suggesting that mitochondrial dysfunction may be involved in the development of migraines. Animal models have also demonstrated that impairments in mitochondrial function can lead to the development of migraines, further supporting the role of mitochondrial dysfunction in migraine pathogenesis [18]. Furthermore, the activity of various mitochondrial enzymes, such as monoamine oxidase, succinate dehydrogenase, NADH dehydrogenase, cycloxygenase, and citrate synthetase, has been found to be decreased in the platelets of migraine patients with or without aura [19, 20]. Mitochondrial dysfunction may cause a decrease in energy production in the brain, leading to an increased susceptibility to migraine attacks [21]. Mitochondrial-targeted therapies have shown promising results in reducing the frequency and severity of migraine attacks, providing additional evidence for the role of mitochondrial impairment in migraines [22, 23]. Mitochondrial stress has been shown to impact the expression and secretion of fibroblast growth factor-21 (FGF-21) and growth differentiation factor-15 (GDF-15) [24]. GDF-15, a member of the transforming growth factor-β family, is known for its ability to regulate systemic energy metabolism by influencing appetite and food intake in the brainstem and hypothalamus. This protein is widely expressed in human tissues and has demonstrated neuroprotective effects in the brain [25–27]. Elevated blood levels of GDF-15 may be a reflection of mitochondrial function in patients [25] and have been shown to protect neuron cells from toxicity by preserving mitochondrial function and reducing apoptosis [28, 29]. FGF-21, which is primarily known as a hepatokine, regulates sugar intake through the central nervous system. Additionally, FGF-21 is involved in various cellular activities, including mitosis and viability, and is generally induced by mitochondrial-dependent mechanisms. Both FGF-21 and GDF-15 are considered to be circulating markers of mitochondrial disorders [30, 31] and have been linked to the severity of such diseases [29]. They are classified as stress-responsive cytokines that can modulate energy balance and play a role in the development of obesity and related comorbidities [32], as well as other diseases such as cancer, cardiovascular disease, diabetes, osteoporosis, and neurodegenerative disorders [30, 33]. It remains elusive for the role of FGF-21 and GDF-15 in migraine. Hence, in this study, we analyzed serum concentrations of FGF-21 and GDF-15 in patients with migraine and healthy controls and the relationship of these cytokines with patients’ clinical parameters. ## Participants We evaluated the serum level of GDF-15 and FGF-21 in migraine patients and healthy controls from September 2019 to January 2023. The diagnosis of migraine is based on the criteria for migraine of International Classification of Headache Disorders 3rd edition criteria. According to attack frequency, patients were split into episodic migraine (EM) and chronic migraine (CM). Exclusion criteria were: (a) under the age of 18; (b) no informed consent could be taken; (c) at least one of the following criteria is met: proved or supposed mitochondrial disorder, cardiac diseases, cancer and chronic inflammatory diseases; (d) other primary or secondary headache disorders. The local ethics committee of the second affiliated hospital of Zhejiang University approved the study. Written informed consent was obtained from all individual participants included in the study. ## Plasma Concentrations of FGF-21 and GDF-15 Plasma FGF-21 and GDF-15 were measures using enzyme-linked immunosorbent assay (ELISA) kits (Abcam, MA, USA). Blood samples were drawn from a peripheral vein into a 9 mL promoting coagulating tubes. Blood samples were centrifuged at 2500 g for 15 min and plasma was then transferred to polypropylene tubes and stored at − 80℃. The scientist who performed the assays was blinded to the study group the sample belonged to. ## Clinical assessment Monthly headache days, peak headache pain intensity, concomitant symptoms, the presence of aura symptoms and the family history of headache were addressed. Furthermore, participants in the migraine group were asked to fill out Migraine Disability Assessment (MIDAS) and Headache Impact Test (HIT-6) to measure the degree of migraine-related functional disability. Patients Health Questionnaire (PHQ-9) and Generalized Anxiety Disorder (GAD-7) questionnaires were performed to measure the degree of migraine-related anxiety and depression. ## Statistical analysis SPSS 23.0 statistics were used for statistics. Kolmogorov–Smirnov test was used to check the normality of the data distribution. Values were expressed as mean ± SE or median (interquartile range). Data were analyzed by one-way analysis of variance (ANOVA) followed by Dunnett’s post hoc test for multiple comparisons. Kruskal–Wallis test was used to evaluate the differences between CM and EM. Logistic regression analyses adjusted for age was used for groups comparison. Receiver operating characteristic (ROC) analysis explored the diagnostic ability of serum FGF-21 and GDF-15. Partial correlation analysis was conducted to detect the potential associations between cytokine levels and clinical parameters as well as questionnaires. The individual statistical tests are indicated within the results section. The $p \leq 0.05$ was considered statistically significant. ## Demographics A total of 221 migraine patients (153 EM and 68 CM) and 124 healthy controls were recruited. In the CM group, 33 patients ($48.53\%$) had comorbid medication-overuse headache (MOH). Patient demographics are summarized in Tables 1 and 2. The age and gender of migraineurs and normal controls were matched [age: 40.50 ± 0.88 (migraineur) vs. 38.98 ± 1.00 (control), $$p \leq 0.815$$; gender (male): $22.17\%$ (migraineur) vs. $19.35\%$ (control), $$p \leq 0.789$$, Table 1]. Patients with CM were older than EM [45.38 ± 1.77 (CM) vs. 38.33 ± 0.95 (EM), $p \leq 0.001$). Patients with CM exhibit not only more headache frequency but also more severe symptoms than patients with EM [monthly headache days, $p \leq 0.001$; peak headache pain intensity, $$p \leq 0.041$$].Table 1Characteristics of migraine patients and healthy controls. Values are presented as mean ± SECharacteristicsControl Group ($$n = 124$$)All Migraine ($$n = 221$$)CM ($$n = 68$$)EM ($$n = 153$$)age, years38.98 ± 1.0040.50 ± 0.8845.38 ± 1.7738.33 ± 0.95BMI, kg/m221.92 ± 0.2922.11 ± 0.2422.48 ± 0.4021.95 ± 0.30males, n (%)24 ($19.35\%$)49 ($22.17\%$)9 ($13.24\%$)40 ($26.14\%$)FGF-21, pg/mL103.49 ± 9.36259.53 ± 18.56313.68 ± 38.33235.30 ± 20.47GDF-15, pg/mL401.63 ± 19.33944.44 ± 35.551022.45 ± 55.06909.76 ± 44.97CM Chronic migraine, EM Episodic migraineTable 2Characteristics of CM and EM patients. Average headache days were calculated in the previous three months. Headache intensity was evaluated using VAS. For female patients, the relationship between headache attack and menstruation was counted. Values are presented as mean ± SE. Kruskal–Wallis test was used to analysis the difference of clinical parameters between CM and EM. The mean values were considered different if $p \leq 0.05$Clinical parametersCM ($$n = 68$$)EM ($$n = 153$$)p-Valueage, years45.38 ± 1.7738.33 ± 0.95 < 0.001BMI, kg/m222.48 ± 0.4021.95 ± 0.300.217males, n (%)9 ($13.24\%$)40 ($26.14\%$)0.168monthly headache days23.12 ± 0.853.73 ± 0.23 < 0.001peak headache pain intensity (VAS)6.53 ± 0.106.31 ± 0.180.041family history41 positive /27 negative113 positive/40 negative0.068migraine with aura, n (%)9 ($13.24\%$)21 ($13.73\%$)0.937nausea or vomit, n (%)44 ($64.71\%$)115 ($75.16\%$)0.070vertigo n (%)26 ($38.24\%$)58 ($37.91\%$)0.988photophobia and phonophobia, n (%)37 ($54.41\%$)109 ($71.24\%$)0.028menstrual migraine, n (%)24 ($40.68\%$)63 ($54.78\%$)0.445MOH, n (%)33 ($48.53\%$)0 < 0.001use of analgesics, n (%)52 ($76.47\%$)47 ($30.72\%$) < 0.001migraine-specific9 ($13.24\%$)6 ($3.92\%$)0.008non-migraine specific48 ($70.59\%$)44 ($28.76\%$%) < 0.001migraine preventive medication use, n (%)61 ($89.71\%$)72 ($47.06\%$) < 0.001antidepressant7 ($10.29\%$)12 ($7.84\%$)0.657antiepileptic14 ($20.59\%$)31 ($20.26\%$)0.130beta-blocker6 ($8.82\%$)9 ($5.88\%$)0.545calcium channel blocker28 ($41.18\%$)48 ($31.37\%$) < 0.001drug acting on the CGRP pathway00/lisinopril or candesartant1 ($1.47\%$)00.657onabotulinumtoxin A4 ($5.88\%$)00.011other (e.g. Chinese traditional medicine)12 ($17.65\%$)39 ($25.49\%$)0.407CM Chronic migraine, EM Episodic migraine, MOH Medication-overuse headache, VAS Visual analogue scale ## FGF-21 and GDF-15 values between groups In order to account for the influence of confounding variables on FGF-21 and GDF-15 concentrations, we took into consideration the factors of age in our analysis. Serum levels of both FGF-21 and GDF-15 in migraine patients were significantly higher than that in healthy controls ($p \leq 0.001$). Meanwhile, the concentration of FGF-21 in patients with CM were significantly higher than EM ($$p \leq 0.026$$) (Fig. 1). No difference between chronic and episodic migraine for GDF-15 was found in this study ($$p \leq 0.290$$). Furthermore, no significant difference in the concentrations of FGF-21 and GDF-15 was observed between the CM group with MOH and the non-MOH group [FGF-21: 272.26 ± 47.63 (MOH) vs. 274.06 ± 52.83 (non-MOH), $$p \leq 0.977$$; GDF-15: 1033.35 ± 89.12 (MOH) vs. 1016.87 ± 115.17 (non-MOH), $$p \leq 0.552$$].Fig. 1FGF-21 and GDF-15 values within the different groups. The $p \leq 0.05$ was considered statistically significant. CM: Chronic migraine; EM: Episodic migraine ## Receiver operating characteristic (ROC) curve analysis We calculated the area under the receiver operating characteristic (ROC) curve to evaluate the effectiveness of FGF-21 and GDF-15 prediction model for patients with migraine (Fig. 2). The area under curve (AUC) in predicting migraine, EM, CM in all subjects for GDF-15 was 0.801 (sensitivity $64.62\%$; specificity $89.66\%$), 0.780 (sensitivity $65.32\%$, specificity $86.29\%$), 0.880 (sensitivity $75.00\%$, specificity $92.00\%$), respectively. For serum FGF-21, the AUC in predicting migraine, EM, CM in all subjects for FGF-21 was 0.729 (sensitivity $59.70\%$; specificity $87.93\%$), 0.755 (sensitivity $59.06\%$, specificity $86.29\%$), 0.755 (sensitivity $65.79\%$; specificity $85.60\%$), respectively. Fig. 2ROC curve analysis of serum GDF-15 for diagnosis of (A) migraine, B CM, (C) EM from control group. ROC curve analysis of serum FGF-21 for diagnosis of (D) migraine, E CM, F EM from control group. CM: Chronic migraine; EM: Episodic migraine; ROC: Receiver operating characteristic ## Relationship between serum factors and clinical variables In migraineurs, the correlation of FGF-21 and GDF-15 with each clinical parameter are summarized in Table 3. We found that FGF-21 and GDF-15 exhibited a positive association with aging. There was no correlation of serum FGF-21 and GDF-15 in migraine with or without aura, nausea or vomit, vertigo and menses (Table 3). We performed a correlation analysis between the cytokines and the burden of migraine, as well as anxiety and depression (Fig. 3). Accelerated disability according to HIT-6 and MIDAS was associated with higher serum level of FGF-21 (HIT-6: $r = 0.266$, $p \leq 0.001$, MIDAS: $r = 0.375$, $p \leq 0.001$) and GDF-15 (HIT-6: $r = 0.297$, $p \leq 0.001$,MIDAS: $r = 0.368$, $p \leq 0.001$). There was no statistical correlation between PHQ-9 and GAD-7 and cytokines. Table 3Clinical parameters of headache and their impact on FGF-21 and GDF-15 levels in migraineurs. Partial correlation analysis was used for analysis. The results are given as r values (p values). The results are given as p valuesClinical parametersFGF-21GDF-15age0.285 ($p \leq 0.001$)0.233 ($p \leq 0.001$)BMI0.075 ($$p \leq 0.254$$)-0.083 ($$p \leq 0.204$$)gender0.088 ($$p \leq 0.179$$)0.060 ($$p \leq 0.356$$)monthly headache days0.061 ($$p \leq 0.357$$)0.017 ($$p \leq 0.799$$)peak headache pain intensity (VAS)-0.065 ($$p \leq 0.328$$)0.096 ($$p \leq 0.146$$)family history-0.148 ($$p \leq 0.035$$)-0.004 ($$p \leq 0.960$$)migraine with aura-0.007 ($$p \leq 0.918$$)0.037 ($$p \leq 0.602$$)nausea or vomit-0.076 ($$p \leq 0.247$$)-0.107 ($$p \leq 0.102$$)vertigo-0.022 ($$p \leq 0.756$$)0.001 ($$p \leq 0.993$$)photophobia and phonophobia-0.005 ($$p \leq 0.938$$)-0.111 ($$p \leq 0.112$$)MOH-0.037 ($$p \leq 0.623$$)0.093 ($$p \leq 0.213$$)menstrual migraine-0.084 ($$p \leq 0.294$$)0.041 ($$p \leq 0.609$$)MOH Medication-overuse headache, VAS Visual analogue scaleFig. 3Correlation between cytokines and clinical variables in migraineurs. A FGF-21 and HIT-6. B FGF-21 and MIDAS. C GDF-15 and HIT-6. D GDF-15 and MIDAS. HIT-6: 6-item Headache Impact Test; MIDAS: Migraine Disability Assessment ## Discussion This study investigated whether the presence of migraine affects serum levels of FGF-21 and GDF-15, both of which have been implicated in metabolic disorders. An increasing number of studies have indicated the impairment in mitochondrial homeostasis has been proposed as a potential contributor to the pathophysiology and susceptibility of migraine [34–37]. FGF-21 and GDF-15 are considered to be markers of mitochondrial disorders [24, 25] and have been linked to the severity of such diseases [23]. Despite all this, the levels of FGF-21 and GDF-15 are likely influenced by age, body max index (BMI) and some certain diseases (e.g. heart failure, myocardial infarction, pulmonary hypertension, diabetes, metabolic syndrome, autoimmune diseases, and cancer) [38, 39]. Since these specific diseases have been listed as exclusion criteria, the remaining confounding variables in this research mainly include gender, BMI, and age. It was confirmed that age, BMI, and gender were matched when comparing the migraine and control groups. The study found significantly heightened serum levels of FGF-21 and GDF-15 in patients with migraine compared to a control group, which suggests that both cytokines may have potential value in distinguishing between migraine sufferers and healthy controls. In this study, the migraine group comprised approximately $30\%$ of individuals with CM, almost half of whom reported experiencing MOH. In comparison of CM and EM, the BMI and gender were matched between the two groups, but age was not, the individuals with CM were older than those with EM. CM is known to develop from EM [40, 41], and the mean age at onset of CM was higher than that of EM [42, 43]. Importantly, age had a significant effect on GDF-15 and FGF-21 levels. As blood indexes closely related to mitochondrial stress, FGF-21 and GDF-15 were established to be related to aging and age-related diseases [44]. Our research found a positive correlation between age and the concentrations of GDF-15 and FGF-21 not only in the migraine group but also in the healthy group. Given that the observed difference in cytokines might be influenced by the older age of CM patients, we corrected for age in the multiple regression when comparing the levels of cytokines in CM and EM patients, however, the significant difference in FGF-21 between the two groups persisted. Additionally, among the migraine group, higher concentrations of FGF-21 and GDF-15 were significantly correlated with increased migraine-related burden and disability, as evidenced by higher scores on the HIT-6 and MIDAS assessments. The higher concentrations of serum FGF-21 and GDF-15 observed in individuals with migraine and increased burden and disability suggest a potential correlation between disease severity. Analogous circumstances are also encountered in other disease. A recent meta-analysis of individual patients showed that GDF-15 demonstrated a strong and consistent independent association with cardiovascular death and heart failure across all presentations of atherosclerotic cardiovascular disease [45]. In a longitudinally sampled cohort of patients with multiple sclerosis, mean GDF-15 concentrations may serve as a biomarker for disease stability [46]. Furthermore, there have been several randomized control trials to evaluate the efficacy of metabolic-targeted nutraceuticals such as coenzyme Q10, magnesium, and riboflavin in the prevention of migraine. These trials have shown promising results in reducing the frequency and severity of migraine attacks, and it is recommended by migraine treatment guidelines in multiple countries [47]. The question of whether the levels of FGF-21 and GDF-15 can serve as a reference for the personalized use of nutraceuticals in the treatment of migraine is a topic that merits further investigation. This study is still subject to limitations. Our results are based on cross-sectional observation, and longitudinal data is necessary to validate our conclusions. Furthermore, a larger study with a more detailed assessment would be necessary to confirm these relationships. Life circumstances, such as food habits or physical activity, were not considered. In addition, serum samples were obtained from patients, regardless of whether they were experiencing a migraine attack or were in a period of remission. There are significant differences in medication use between CM and EM patients, including analgesics and preventive medications. Due to the extremely complex types and dosages of these medications, it is difficult to include all of them as corrective factors in our analysis of the results, which may have introduced bias into the analysis of our results. While as an additional investigation, we conducted a comparative analysis of the concentrations of FGF-21 and GDF-15 in the MOH and non-MOH subgroups of CM patients, and did not find a significant difference of excessive use of analgesics on metabolic levels. This research is deficient in mitochondria-associated parameters as measured in the serum of patients, especially those that are downstream to FGF-21 and GDF-15. The elevated levels of GDF-15 and FGF-21 in patients with migraine are insufficient to support the potential mechanism of mitochondria in migraine. We have demonstrated that serum levels of GDF-15 and FGF-21 are elevated in patients with migraine. The heightened concentrations of GDF-15 and FGF-21 are linked to greater disease burden, indicating their potential as peripheral blood markers for evaluating the severity of migraine-related disability. ## References 1. Feigin V, Vos T, Nichols E, Owolabi M, Carroll W, Dichgans M. The global burden of neurological disorders: translating evidence into policy. Lancet Neurol (2020.0) 19 255-265. DOI: 10.1016/s1474-4422(19)30411-9 2. Gross E, Lisicki M, Fischer D, Sándor P, Schoenen J. The metabolic face of migraine - from pathophysiology to treatment. Nat Rev Neurol (2019.0) 15 627-643. DOI: 10.1038/s41582-019-0255-4 3. Powers S, Radak Z, Ji L. Exercise-induced oxidative stress: past, present and future. J Physiol (2016.0) 594 5081-5092. DOI: 10.1113/jp270646 4. Pingitore A, Lima G, Mastorci F, Quinones A, Iervasi G, Vassalle C. Exercise and oxidative stress: potential effects of antioxidant dietary strategies in sports. Nutrition (2015.0) 31 916-922. DOI: 10.1016/j.nut.2015.02.005 5. Schiavone S, Jaquet V, Trabace L, Krause K. Severe life stress and oxidative stress in the brain: from animal models to human pathology. Antioxid Redox Signal (2013.0) 18 1475-1490. DOI: 10.1089/ars.2012.4720 6. Schoonman G, Evers D, Terwindt G, van Dijk J, Ferrari M. The prevalence of premonitory symptoms in migraine: a questionnaire study in 461 patients. Cephalalgia (2006.0) 26 1209-1213. DOI: 10.1111/j.1468-2982.2006.01195.x 7. Arngrim N, Schytz H, Britze J, Amin F, Vestergaard M, Hougaard A. Migraine induced by hypoxia: an MRI spectroscopy and angiography study. Brain (2016.0) 139 723-737. DOI: 10.1093/brain/awv359 8. Broessner G, Rohregger J, Wille M, Lackner P, Ndayisaba J, Burtscher M. Hypoxia triggers high-altitude headache with migraine features: A prospective trial. Cephalalgia (2016.0) 36 765-771. DOI: 10.1177/0333102415610876 9. Trivedi M, Holger D, Bui A, Craddock T, Tartar J. Short-term sleep deprivation leads to decreased systemic redox metabolites and altered epigenetic status. PloS one. (2017.0) 12 e0181978. DOI: 10.1371/journal.pone.0181978 10. Angelucci F, Silva V, Dal Pizzol C, Spir L, Praes C, Maibach H. Physiological effect of olfactory stimuli inhalation in humans: an overview. Int J Cosmetic Sci (2014.0) 36 117-123. DOI: 10.1111/ics.12096 11. Franken C, Lambrechts N, Govarts E, Koppen G, Den Hond E, Ooms D. Phthalate-induced oxidative stress and association with asthma-related airway inflammation in adolescents. Int J Hyg Envir Heal (2017.0) 220 468-477. DOI: 10.1016/j.ijheh.2017.01.006 12. Kwiatkowska K, Bacalini M, Sala C, Kaziyama H, de Andrade D, Terlizzi R. Analysis of Epigenetic Age Predictors in Pain-Related Conditions. Front (2020.0) 8 172. DOI: 10.3389/fpubh.2020.00172 13. Hahad O, Prochaska J, Daiber A, Muenzel T. Environmental Noise-Induced Effects on Stress Hormones, Oxidative Stress, and Vascular Dysfunction: Key Factors in the Relationship between Cerebrocardiovascular and Psychological Disorders. Oxid Med Cell Longev (2019.0) 2019 4623109. DOI: 10.1155/2019/4623109 14. Reyngoudt H, Achten E, Paemeleire K. Magnetic resonance spectroscopy in migraine: what have we learned so far?. Cephalalgia (2012.0) 32 845-859. DOI: 10.1177/0333102412452048 15. Sándor P, Dydak U, Schoenen J, Kollias S, Hess K, Boesiger P. MR-spectroscopic imaging during visual stimulation in subgroups of migraine with aura. Cephalalgia (2005.0) 25 507-518. DOI: 10.1111/j.1468-2982.2005.00900.x 16. Yorns W, Hardison H. Mitochondrial dysfunction in migraine. Seminars in pediatric neurology (2013.0) 20 188-193. DOI: 10.1016/j.spen.2013.09.002 17. Okada H, Araga S, Takeshima T, Nakashima K. Plasma lactic acid and pyruvic acid levels in migraine and tension-type headache. Headache (1998.0) 38 39-42. DOI: 10.1046/j.1526-4610.1998.3801039.x 18. Eikermann-Haerter K, Dileköz E, Kudo C, Savitz S, Waeber C, Baum M. Genetic and hormonal factors modulate spreading depression and transient hemiparesis in mouse models of familial hemiplegic migraine type 1. J Clin Investig (2009.0) 119 99-109. DOI: 10.1172/jci36059 19. Littlewood J, Glover V, Sandler M, Peatfield R, Petty R, Clifford Rose F. Low platelet monoamine oxidase activity in headache: no correlation with phenolsulphotransferase, succinate dehydrogenase, platelet preparation method or smoking. J Neurol Neurosurg Psychiatry (1984.0) 47 338-343. DOI: 10.1136/jnnp.47.4.338 20. Sangiorgi S, Mochi M, Riva R, Cortelli P, Monari L, Pierangeli G. Abnormal platelet mitochondrial function in patients affected by migraine with and without aura. Cephalalgia (1994.0) 14 21-23. DOI: 10.1046/j.1468-2982.1994.1401021.x 21. Sparaco M, Feleppa M, Lipton R, Rapoport A, Bigal M. Mitochondrial dysfunction and migraine: evidence and hypotheses. Cephalalgia (2006.0) 26 361-372. DOI: 10.1111/j.1468-2982.2005.01059.x 22. Woldeamanuel Y, Rapoport A, Cowan R. The place of corticosteroids in migraine attack management: A 65-year systematic review with pooled analysis and critical appraisal. Cephalalgia (2015.0) 35 996-1024. DOI: 10.1177/0333102414566200 23. 23.Derry C, Derry S, and Moore R (2014) Caffeine as an analgesic adjuvant for acute pain in adults. Cochrane Database Syst. Rev. 2014:CD009281. 10.1002/14651858.CD009281.pub3. 24. Kim K, Lee M. GDF15 as a central mediator for integrated stress response and a promising therapeutic molecule for metabolic disorders and NASH. Biochim. Biophys. Acta. Gen. Subj. (2021.0) 1865 129834. DOI: 10.1016/j.bbagen.2020.129834 25. 25.Fujita Y, Taniguchi Y, Shinkai S, Tanaka M, and Ito M (2016) Secreted growth differentiation factor 15 as a potential biomarker for mitochondrial dysfunctions in aging and age-related disorders. Geriatr. Gerontol. Int.:17–29. 10.1111/ggi.12724. 26. Strelau J, Sullivan A, Böttner M, Lingor P, Falkenstein E, Suter-Crazzolara C. Growth/differentiation factor-15/macrophage inhibitory cytokine-1 is a novel trophic factor for midbrain dopaminergic neurons in vivo. J Neurosci (2000.0) 20 8597-8603. DOI: 10.1523/jneurosci.20-23-08597.2000 27. Subramaniam S, Strelau J, Unsicker K. Growth differentiation factor-15 prevents low potassium-induced cell death of cerebellar granule neurons by differential regulation of Akt and ERK pathways. J Biol Chem (2003.0) 278 8904-8912. DOI: 10.1074/jbc.M210037200 28. Li P, Lv H, Zhang B, Duan R, Zhang X, Lin P. Growth Differentiation Factor 15 Protects SH-SY5Y Cells From Rotenone-Induced Toxicity by Suppressing Mitochondrial Apoptosis. Front. in Aging Neurosci (2022.0) 14 869558. DOI: 10.3389/fnagi.2022.869558 29. Montero R, Yubero D, Villarroya J, Henares D, Jou C, Rodríguez M. GDF-15 Is Elevated in Children with Mitochondrial Diseases and Is Induced by Mitochondrial Dysfunction. PloS one (2016.0) 11 e0148709. DOI: 10.1371/journal.pone.0148709 30. Cardoso A, Fernandes A, Aguilar-Pimentel J, de Angelis M, Guedes J, Brito M. Towards frailty biomarkers: Candidates from genes and pathways regulated in aging and age-related diseases. Ageing Res Rev (2018.0) 47 214-277. DOI: 10.1016/j.arr.2018.07.004 31. Davis R, Liang C, Sue C. A comparison of current serum biomarkers as diagnostic indicators of mitochondrial diseases. Neurology (2016.0) 86 2010-2015. DOI: 10.1212/wnl.0000000000002705 32. Keipert S, Ost M. Stress-induced FGF21 and GDF15 in obesity and obesity resistance. Trends Endocrin Met (2021.0) 32 904-915. DOI: 10.1016/j.tem.2021.08.008 33. Conte M, Martucci M, Chiariello A, Franceschi C, Salvioli S. Mitochondria, immunosenescence and inflammaging: a role for mitokines?. Semin Immunopathol (2020.0) 42 607-617. DOI: 10.1007/s00281-020-00813-0 34. 34.Grech O, Mollan S, Wakerley B, Fulton D, Lavery G, and Sinclair A (2021) The Role of Metabolism in Migraine Pathophysiology and Susceptibility. Life-Basel. 11. http://doi.org/10.3390/life11050415. 35. van Tilburg M, Parisien M, Boles R, Drury G, Smith-Voudouris J, Verma V. A genetic polymorphism that is associated with mitochondrial energy metabolism increases risk of fibromyalgia. Pain (2020.0) 161 2860-2871. DOI: 10.1097/j.pain.0000000000001996 36. Boles R, Zaki E, Lavenbarg T, Hejazi R, Foran P, Freeborn J. Are pediatric and adult-onset cyclic vomiting syndrome (CVS) biologically different conditions? Relationship of adult-onset CVS with the migraine and pediatric CVS-associated common mtDNA polymorphisms 16519T and 3010A. Neurogastroenterol Motil (2009.0) 21 936-e972. DOI: 10.1111/j.1365-2982.2009.01305.x 37. Zaki E, Freilinger T, Klopstock T, Baldwin E, Heisner K, Adams K. Two common mitochondrial DNA polymorphisms are highly associated with migraine headache and cyclic vomiting syndrome. Cephalalgia (2009.0) 29 719-728. DOI: 10.1111/j.1468-2982.2008.01793.x 38. Verhamme F, Freeman C, Brusselle G, Bracke K, Curtis J. GDF-15 in Pulmonary and Critical Care Medicine. Am J Respir Cell Mol Biol (2019.0) 60 621-628. DOI: 10.1165/rcmb.2018-0379TR 39. Wischhusen J, Melero I, Fridman W. Growth/Differentiation Factor-15 (GDF-15): From Biomarker to Novel Targetable Immune Checkpoint. Front Immunol (2020.0) 11 951. DOI: 10.3389/fimmu.2020.00951 40. May A, Schulte L. Chronic migraine: risk factors, mechanisms and treatment. Nat Rev Neurol (2016.0) 12 455-464. DOI: 10.1038/nrneurol.2016.93 41. Scher A, Stewart W, Ricci J, Lipton R. Factors associated with the onset and remission of chronic daily headache in a population-based study. Pain (2003.0) 106 81-89. DOI: 10.1016/s0304-3959(03)00293-8 42. Buse D, Manack A, Fanning K, Serrano D, Reed M, Turkel C. Chronic migraine prevalence, disability, and sociodemographic factors: results from the American Migraine Prevalence and Prevention Study. Headache (2012.0) 52 1456-1470. DOI: 10.1111/j.1526-4610.2012.02223.x 43. Buse D, Manack A, Serrano D, Turkel C, Lipton R. Sociodemographic and comorbidity profiles of chronic migraine and episodic migraine sufferers. J Neurol Neurosurg Psychiatry (2010.0) 81 428-432. DOI: 10.1136/jnnp.2009.192492 44. Amorim J, Coppotelli G, Rolo A, Palmeira C, Ross J, Sinclair D. Mitochondrial and metabolic dysfunction in ageing and age-related diseases. Nat Rev Endocrinol (2022.0) 18 243-258. DOI: 10.1038/s41574-021-00626-7 45. Kato E, Morrow D, Guo J, Berg D, Blazing M, Bohula E. Growth differentiation factor 15 and cardiovascular risk: individual patient meta-analysis. Eur Heart J (2023.0) 44 293-300. DOI: 10.1093/eurheartj/ehac577 46. Amstad A, Coray M, Frick C, Barro C, Oechtering J, Amann M. Growth differentiation factor 15 is increased in stable MS. Neurol Neuroimmunol Neuroinflamm (2020.0) 7 e675. DOI: 10.1212/nxi.0000000000000675 47. Ducros A, de Gaalon S, Roos C, Donnet A, Giraud P, Guégan-Massardier E. Revised guidelines of the French headache society for the diagnosis and management of migraine in adults. Part 2: Pharmacological treatment. Revue neurologique (2021.0) 177 734-752. DOI: 10.1016/j.neurol.2021.07.006
--- title: Body mass index in nursing home residents during the first year after admission authors: - Corinna Vossius - Miguel G. Borda - Bjørn Lichtwarck - Janne Myhre - May Ingvild Volungholen Sollid - Tom Borza - Ingvild Hjorth Feiring - Jūratė Šaltytė Benth - Sverre Bergh journal: BMC Nutrition year: 2023 pmcid: PMC10026511 doi: 10.1186/s40795-023-00710-3 license: CC BY 4.0 --- # Body mass index in nursing home residents during the first year after admission ## Abstract ### Background Malnutrition - comprising both undernutrition and overweight - has to be addressed in the medical follow-up of older adults due to the negative consequences for the functional state and general health. Still, little is known about the nutritional state of nursing home (NH) residents, especially with respect to weight gain or weight loss after NH admission. Therefore, this study aims to evaluate changes in the body mass index (BMI) during the first year following NH admission, and to explore demographic and clinical characteristics related to BMI changes. ### Methods Data from two prospective studies that recruited participants at NH admission were combined. Demographic and clinical characteristics including the BMI were assessed at baseline and after one year. A linear regression model was estimated to explore the impact of demographic and clinical characteristics on the change in BMI. ### Results The study cohort consisted of 1,044 participants with a mean age of 84.3 years (SD7.6) at baseline; $64.2\%$ were female. At baseline, $33\%$ of the NH residents had severe to moderate undernutrition, while $10\%$ were obese. During the first year of their NH stay, residents with severe to moderate undernutrition had an average increase in BMI of 1.3 kg/m2 (SD 2.2; $p \leq 0.001$), while weight changes were either very small or not significant in the other BMI groups. Characteristics related to weight gain were younger age and less agitation. ### Conclusion Malnutrition is a common health challenge at NH admission, with one third of NH residents being moderately to severely underweight and $10\%$ being obese. However, during the first year of NH stay, there was a favourable development for underweight NH residents, as they increased their BMI, and $43.6\%$ changed to a higher weight classification, while we observed no changes in the BMI in residents with obesity. As NH residents are in the last phase of their lives, interventions to prevent malnutrition or overweight should be initiated while still home-dwelling, and then continued in the nursing homes. ## Introduction “Feed me, feed me, Simon, feed me all night long!” sings the bloodthirsty Venus flytrap in the Little Shop of Horrors. In contrast to the man-eating plant in the musical, most humans are content with standard food, and their main inclination is that it satisfies their nutritional needs, taste, and eating habits. However, malnutrition is a global challenge, with undernutrition denoting insufficient intake of energy and nutrients to meet an individual’s needs to maintain good health, and overweight as abnormal or excessive fat accumulation that presents a risk to health, and it has to be addressed in the care and medical follow-up of older adults [1, 2]. The Norwegian Directorate of Health estimates that as many as $40\%$ of residents in Norwegian nursing homes (NHs) and persons receiving home care suffer from, or are at risk of, developing undernutrition [3]. Additionally, obesity represents an increasing challenge in nursing home residents [4]. Norwegian NHs are designed for people in need of continuous care and supervision. Mean age at admission is 84 years, and $84\%$ of the long-term residents suffer from dementia [5, 6]. Many residents are no longer capable of identifying and attaining to their basic needs, including nutritional intake, and thus depend on the health care personnel staff to provide adequate nutrition and hydration. A review published by Alzheimer’s Disease International in 2014 described existing research about nutrition in people with dementia [7]. They reported that weight loss might be part of the normal process of aging, and how it is exacerbated in people with dementia, which might lead to several negative consequences. Undernutrition decreases bodily reserves required for stress response and increases the risk of other complications such as infections, falling tendency, fractures, frailty, and sarcopenia [8]. Furthermore, undernutrition is related to longer convalescence periods, increased care needs, and extensive medical expenses [3]. Previous research has shown that undernutrition in NH residents was related to lower functioning in activities of daily living (ADL), having dementia, problems with food intake, higher age, and higher mortality [5, 9–11]. However, reports show that a good nutritional status is not only associated with less disability and lower mortality, but as well with slower cognitive decline in persons suffering from dementia [12–15]. At the same time, a review on studies from the US describes different medical and functional profiles for residents with obesity as compared to normal or underweight residents, younger age at NH admission and a need for more extensive assistance [4]. Still, little is known about the nutritional state of residents at NH admission, especially with respect to weight gain or loss, and demographic and clinical characteristics related to weight changes. Therefore, this study aims to evaluate changes in the body mass index (BMI) during the first year following the NH admission, and to explore demographic and clinical characteristics related to BMI changes. ## Settings and participants We combined data from two prospective clinical studies that recruited participants from 71 NHs in 45 municipalities in Norway at NH admission. The demographic and clinical characteristics from both cohorts were similar. These studies were: Resource Use and Disease Course in Dementia - Nursing Home (REDIC-NH) including 696 persons. Inclusion took place between 2012 and 2014 [16].Cooperation between The Department of Old Age Psychiatry, Innlandet Hospital Trust, and municipal nursing homes in the Innlandet County (SAM-AKS III). SAM-AKS III is an ongoing study that started in 2014 [17]. Inclusion criteria were: (i) 65 years of age or older in REDIC-NH and 60 years of age or older in SAM-AKS III or (ii) having dementia irrespective of age. ( iii) In addition, expected survival should be six weeks or more for REDIC-NH and four weeks or more for SAM-AKS III. Only residents that completed baseline assessment (BL) were included in the studies. For the present study additional inclusion criteria were: (a) Participants had completed BL within 90 days after NH admission, (b) participants had completed the 12-months follow-up examination (FU12) within one year and 90 days after BL or had died before FU12, and (c) BMI recorded at all assessments. ## Data collection Data collection was performed by trained healthcare workers at the NH, mainly registered nurses, under supervision of research nurses from the Research Centre for Age-Related Functional Decline and Disease. The research nurses completed a five-day training prior to study start, while the data collectors completed a two-day training. Data were collected through structured interviews with participants and a caregiver [16]. All rating scales and inventories were applied using validated, Norwegian versions. The following demographic and clinical data were collected: ## Demographic data Gender, age, and living arrangements before admission to NH, were collected by reviewing patients’ journals. ## Body mass index (BMI) relates a person’s weight to the height (BMI = weight/height2). The height was assessed at BL by either measuring the resident or by asking the residents or their proxies when the resident was unable to stand erect. The weight was established by weighing the residents at BL and at FU12. ## The global leadership conversation: addressing malnutrition (GLIM)-criteria were applied, and the participants were categorized according to their BMI into severe undernutrition (BMI under 18.5 for persons younger than 70 years and under 20 for persons 70 years or older), moderate undernutrition (BMI 18.5 to 20, respectively 20 to 22), normal weight (BMI 20 to 25, respectively 22 to 27) and overweight (BMI over 25, respectively over 27). In addition, participants with a BMI of 30 or higher were classified as obese [2, 18]. ## The clinical dementia rating scale (CDR) was applied to assess the severity of dementia. The rating scale comprises six items, where the total CDR score is obtained based on an algorithm [19]. For statistical analyses we calculated the CDR-sum of boxes (CDR-SoB) that offers an extended range of values compared to the algorithm-based scoring, and it is calculated by adding the item scores (range 0–18), where higher scores indicate more severe dementia [20]. ## The neuropsychiatric inventory (NPI) assesses neuropsychiatric symptoms. The instrument contains 12 items and is conducted as an interview with a caregiver. Severity (scored 0–3) is multiplied by frequency (scored 0–4), giving an item score from 0 to 12, where higher scores indicate more severe symptoms [21, 22]. Based on a previous principal component analysis, we created the following sub-syndromes: NPI-Agitation (agitation/aggression, disinhibition, and irritability), NPI-Psychosis (delusions and hallucinations), and NPI-Affective symptoms (depression and anxiety) [16]. ## Physical self-maintenance scale (PSMS) consists of six items (scored 1–5) and assesses personal activities of daily living (PADL) function. The overall score ranges from 6 to 30, where higher scores indicate higher PADL dependency [23]. ## General medical health rating (GMHR) rates physical health and was assessed by the health care workers performing the examinations. It consists of one item comprising four categories: excellent, good, moderate, or poor [24]. ## The mobilization-observation-behaviour-intensity-dementia pain scale (MOBID-2) was applied to assess pain. For this study, we only included the overall proxy rating ranging from 0 to 10, where a higher score indicates more severe pain. [ 25] ## Ethics The residents’ capacity to consent to participation in the study was considered by the NH staff, including the NH physician. Written informed consent was obtained by the participants with full capacity to consent, or by next of kin on behalf of the participants in case of reduced capacity to consent. The Regional Ethics Committee for Medical research in South-Eastern Norway approval for the two studies ($\frac{2011}{1378}$a and $\frac{2014}{917}$) includes the analysis for the present study. ## Statistics Demographic factors and clinical characteristics were presented as means and standard deviations (SDs), or frequencies and percentages. The group differences were assessed by Student’s t-test for continuous variables and χ2-test for categorical variables. The analyses comparing participants with severe undernutrition, moderate undernutrition, normal weight, overweight, and obesity according to pre-specified BMI cut-offs were considered exploratory. For some of the exploratory analyses the sub-cohorts of participants with severe and moderate undernutrition were combined. The BMI change between BL and FU12 was assessed by paired samples t-test. A linear regression model was estimated to explore the impact of demographic and clinical characteristics at BL on the change in BMI between BL and FU12. The following characteristics were included: age, gender, living alone before NH admission, GMHR, PSMS, CDR-SoB, the NPI sub-syndromes Agitation, Psychosis and Affective, the overall proxy rating of MOBID-2, and BMI at BL. Due to mathematical coupling and regression to the mean, a straightforward regression analysis between change in BMI and BL BMI might provide biased results. To obtain an accurate estimate of the association between the BMI at BL and the change in BMI, Blomqvist’s method adjusting the estimated coefficient and its standard error for measurement error variance, was applied [26, 27]. Measurement error variance was estimated by employing the BL and FU6 data from the participants of the REDIC study cohort. Due to participants being included from different NH, a cluster effect on NH-level was assessed by intra-class correlations coefficient. As no such effect was found, no adjustment was implemented. Only participants with FU12 assessment and no missing values in covariates were included in the regression analysis. Thus, the sub-cohort of participants not included into the analysis comprised participants deceased before FU12 or with incomplete datasets. Results with p-values below 0.05 were considered statistically significant. The analyses were performed in SPSS v27 and STATA v17. ## Results As shown in Fig. 1a total of 5,127 persons were eligible for study inclusion, whereof $57\%$ did not participate because they or their next of kin did not consent ($14\%$), the resident died before BL assessment ($9\%$), or other reasons ($32\%$). Those not included were more often male than those included (39 vs. $35\%$; $p \leq 0.027$). A total of 2,226 participants were included into REDIC or SAM-AKS III. Out of these, 932 performed BL assessment within 90 days after NH admission, and FU12 within one year and 90 days after BL. Further, 356 performed BL assessment within 90 days after NH admission, but they died before FU12. Among these 1,283 included participants, BMI was recorded at BL and FU12 in 769 cases, and respectively at BL in 275 participants who deceased before FU12. The study cohort thus consisted of 1,044 participants. Fig. 1Flow chart of eligible participants and the inclusion process of the studyREDIC-NH = Resource Use and Disease Course in Dementia - Nursing Home; SAM-AKS III = Cooperation between The Department of Old Age Psychiatry, Innlandet Hospital Trust, and municipal nursing homes in the Innlandet County. BL = baseline; FU12 = follow-up at 12 months. Table 1 shows demographic and clinical characteristics at BL for both included and excluded participants. Except for a higher ADL-dependency, there were no differences between excluded and included participants. Table 1Demographic and clinical characteristics at BL for included participants and participants excluded from the study, participants included into the regression analysis and those not included into regression analysis, and a comparison of the respective groupsIncluded participantsExcluded participants from the studyPIncl. vs. excl. from the studyParticipants included into regression analysisParticipants not included into regression analysisPIncl. vs. not incl. into regression analysisN10441182640404Age, mean (SD)84.3 (7.6)84.7 (7.7)0.22184.0 (7.9)84.8 (7.2)0.084Gender, female (%)670 (64.2)754 (63.8)0.871408 (63.7)262 (64.9)0.718Living alone before NH admission (%)716 (68.9)785 (67.8)0.592443 (69.2)273 (68.4)0.787BMI, mean (SD)24.3 (5.2)24.9 (4.6)0.01624.4 (4.5)24.0 (6.0)0.215GMHR, poor or moderate (%)486 (49.8)597 (53.4)0.087289 (45.2)197 (58.8) < 0.001 PSMS, mean (SD)14.5 (4.5)15.3 (4.6) < 0.001 13.9 (4.2)15.5 (4.5) < 0.001 CDR-SoB, mean (SD)10.2 (4.0)10.3 (4.5)0.44810.1 (3.8)10.3 (4.4)0.560NPI, mean (SD)14.1 (16.9)14.3 (17.7)0.74312.9 (16.5)15.9 (17.3) 0.007 - NPI-AGI- NPI-PSY- NPI-AFF5.4 (8.3)1.8 (3.9)4.1 (6.5)5.2 (8.4)1.8 (3.9)4.6 (6.7)0.5200.6720.1373.7 (6.1)1.7 (3.8)5.0 (8.3)4.9 (7.0)1.8 (4.0)6.0 (8.3) 0.008 0.8270.069MOBID-2, mean (SD)2.1 (2.1)2.3 (2.2)0.1191.9 (2.1)2.0 (2.2)0.292BL = baseline; incl.=included; excl.=excluded; SD = standard deviation; GMHR = General medical health rating; BMI = Body mass index; PSMS = Physical self-maintenance scale; NPI = Neuropsychiatric inventory; NPI-AGI = NPI sub-syndromes agitation/aggression, disinhibition and irritability, NPI-PSY = NPI sub-syndromes delusions and hallucinations; NPI-AFF = NPI sub-syndromes depression and anxiety; CDR-SoB = Clinical dementia rating scale – sum of boxes; MOBID-2 = The Mobilization-Observation-Behaviour-Intensity-Dementia Pain Scale, overall proxy rating. Table 2 shows demographic and clinical characteristics at BL stratified according to weight classification. About one third of the participants had severe or moderate undernutrition, and about $10\%$ were obese. Explorative analyses showed that participants with severe to moderate undernutrition at NH admission were older ($$p \leq 0.001$$), had more neuropsychiatric symptoms ($$p \leq 0.019$$), were living more often alone before NH admission, and had a higher risk to die before FU12 ($$p \leq 0.029$$) than participants with normal weight or overweight. Obesity was related to lower age at NH admission ($$p \leq 0.023$$), lower general health state ($p \leq 0.001$), and lower degree of cognitive impairment ($$p \leq 0.012$$) when compared to the other weight groups. Both residents with severe undernutrition and those with obesity suffered from more pain ($$p \leq 0.012$$ and $$p \leq 0.002$$, respectively) when compared to the other weight groups. On average, the BMI increase was 0.6 kg/m2 (SD 2.5, $p \leq 0.001$) between BL and FU12. Persons with severe to moderate undernutrition had the highest increase in BMI, with 1.3 kg/m2 (SD 2.2; $p \leq 0.001$) and $43.6\%$ of residents in these weight classifications changed to a higher weight classification during their first year of NH stay. Table 2Demographic and clinical characteristics at BL, stratified according to weight classification, and BMI changes and changes in weight classification between BL and FU12.AllSevere under-nutritionModerate under-nutritionNormal weightOverweightObesityN1044169 (16.2)178 (17.0)453 (43.2)139 (13.3)107 (10.2)Age, mean (SD)84.3 (7.6)85.6 (7.0)85.1 (7.0)84.3 (7.6)82.7 (8.4)82.3 (7.8)Gender, female (%)670 (64.2)127 (75.1)118 (66.3)266 (59.0)85 (61.2)74 (69.2)Living alone before NH admission (%)716 (68.9)128 (75.7)125 (70.2)301 (67.2)69 (64.0)73 (68.9)BMI, mean (SD)24.3 (5.2)17.9 (1.6)21.0 (0.6)24.4 (1.5)28.2 (1.0)34.2 (6.8)GMHR, poor or moderate486 (49.8)84 (49.7)79 (47.3)193 (45.7)63 (48.5)67 (67.0)PSMS, mean (SD)14.5 (4.5)14.9 (4.7)14.5 (4.6)14.2 (4.2)14.7 (4.5)14.7 (3.9)CDR-SoB, mean (SD)10.2 (4.0)10.7 (4.1)10.0 (4.1)10.2 (3.9)10.4 (3.8)9.2 (4.1)NPI, mean (SD)14.1 (16.9)15.5 (16.5)16.1 (17.9)13.9 (17.6)11.6 (14.6)12.1 (15.0)- NPI-AGI- NPI-PSY- NPI-AFF4.1 (6.5)1.8 (3.9)5.4 (8.3)4.0 (6.0)1.9 (4.4)5.6 (8.6)4.9 (7.1)2.1 (4.2)6.2 (9.5)4.3 (6.8)1.7 (3.8)5.4 (8.5)3.6 (5.3)1.4 (3.1)4.4 (6.4)3.3 (5.6)1.6 (3.6)4.6 (7.1)MOBID-2, mean (SD)2.1 (2.1)2.4 (2.2)2.0 (2.2)1.7 (2.0)1.7 (1.7)2.5 (2.4)Deceased before FU12275 (26.3)58 (34.3)48 (27.0)115 (25.5)33 (23.7)21 (19.6)BMI change between BL and FU12 ($95\%$CI)0.6(0.4–0.8)1.6(1.2–2.1)1.0(0.6–1.4)0.4(0.1–0.7)0.4(0–0.8)-0.3(-1.0–0.4)Changes in weight classification between BL and FU12 (%)- Change to a higher group- Change to a lower group221 (28.7)69 (9.0)44 (39.6)-61 (46.2)17 (12.9)79 (23.5)20 (6.0)37 (31.7)18 (17.0)-14 (16.3)BL = baseline; FU12 = follow-up at 12 months; SD = standard deviation; GMHR = General medical health rating; BMI = Body mass index; PSMS = Physical self-maintenance scale; NPI = Neuropsychiatric inventory; NPI-AGI = NPI sub-syndromes agitation/aggression, disinhibition and irritability, NPI-PSY = NPI sub-syndromes delusions and hallucinations; NPI-AFF = NPI sub-syndromes depression and anxiety; CDR-SoB = Clinical dementia rating scale – sum of boxes; MOBID-2 = The Mobilization-Observation-Behaviour-Intensity-Dementia Pain Scale, overall proxy rating. Table 3 presents the results of the linear regression model assessing the association between the change in BMI between BL and FU12 and demographic and clinical characteristics at BL. In both the bivariate and multiple models, the BMI at BL is significantly negatively correlated to the change in BMI from BL to FU12. However, after Blomqvist’s adjustment, the association between BMI at BL and the change in BMI is no longer significant. According to the Blomqvist-adjusted multiple model, there would be less weight gain and respectively a higher weight loss with higher age and with more symptoms of agitation. Table 3Results of the linear regression model for association between the change in BMI between BL and FU12 and demographic and clinical characteristics at BL.CovariateBivariate modelsMultiple modelRegr.coeff. ( $95\%$ CI)p-valueRegr.coeff. ( $95\%$ CI)p-valueBMI BL – unadjustedBMI BL – adjustedAgeGender, femaleLiving alone, yesGMHR, poor/moderatePSMSCDR-SoBNPI-AGINPI-PSYNPI-AFFMOBID-2-0.12 (-0.16; -0.07)-0.02 (-0.07; 0.04)0.001 (-0.02; 0.03)0.45 (0.05; 0.84)0.57 (0.16; 0.98)-0.38 (-0.77; 0.001)-0.06 (-0.10; -0.01)-0.07 (-0.12; -0.01)-0.04 (-0.07; -0.01)-0.03 (-0.08; 0.02)-0.03 (-0.05; -0.01)-0.05 (-0.14; 0.05) < 0.001 0.4980.949 0.028 0.007 0.051 0.014 0.012 0.009 0.241 0.007 0.310-0.12 (-0.16; -0.08)-0.02 (-0.08; 0.04)-0.03 (-0.05; -0.002)0.30 (-0.10; 0.70)0.37 (-0.06; 0.80)-0.18 (-0.57; 0.21)-0.01 (-0.06; 0.04)-0.04 (-0.10; 0.02)-0.04 (-0.08; -0.00004) 0.01 (-0.05; 0.07)-0.02 (-0.05; 0.007)-0.05 (-0.14; 0.05) < 0.001 0.468 0.033 0.1430.0900.3610.6340.143 0.0498 0.7720.1540.349Regr.coeff.=Regression coefficient; CI = Confidence interval; BL = baseline; GMHR = General medical health rating; BMI = Body mass index; PSMS = Physical self-maintenance scale; NPI = Neuropsychiatric inventory; NPI-AGI = NPI sub-syndromes agitation/aggression, disinhibition and irritability, NPI-PSY = NPI sub-syndromes delusions and hallucinations; NPI-AFF = NPI sub-syndromes depression and anxiety; CDR-SoB = Clinical dementia* rating scale – sum of boxes; MOBID-2 = The Mobilization-Observation-Behaviour-Intensity-Dementia Pain Scale, overall proxy rating.Unadjusted for measurement error variance; * adjusted for measurement error variance by Blomqvist’s method. ## Discussion At NH admission we observed that about $33\%$ of the NH residents had severe to moderate undernutrition, while $10\%$ were obese. Residents with undernutrition were older, living more often alone before NH admission, had more symptoms of agitation, and a higher mortality rate. Residents with obesity where younger, experienced more pain, had a lower general health state and less cognitive impairment. During the first year of their NH stay, persons with severe to moderate undernutrition had an average increase in BMI of 1.3 kg/m2, while weight changes were either very small or not significant in the other BMI groups. Weight gain was associated with younger age and less agitation. Our findings regarding undernutrition are in line with previous research that reports a prevalence of undernutrition of $30\%$ in Swedish NHs [28]. The findings also reflect well the nutrition states in home-dwelling persons with dementia in Norway, where $29\%$ were found to be underweight at the time of the diagnosis of dementia [11]. Underweight residents had a higher mortality rate during their first year of NH stay, and these results are also in accordance with previous research, possibly indicating that weight loss is part of the natural process of dying of age [5, 29]. Residents with undernutrition represented the sub-cohort with the highest weight gain and 43.6 changed into a higher weight classification group during their first year of NH stay, indicating that the NHs succeeded in providing adequate nourishment. Unfortunately, we lack data whether the follow-up of the residents involved screening for malnutrition and creating customized food plans, as required by the national guidelines, or if the observed weight gain was a more coincidental result of regular meals and the general availability of food or an indicator of general thriving. In addition, we observed that $12.9\%$ of persons with moderate undernutrition at BL were classified as severely undernourished at FU12. These cases would especially warrant a clarification of the reasons for weight loss and individualized interventions. However, our findings contradict quite frequent reports in Norwegian media about insufficient nutrition of NH residents and patients supposedly “starving to death” [30–32]. Rather than created at the NH, malnutrition seems to be a problem that arises prior to NH admission. In the exploratory analyses we found that residents with malnutrition at BL more often were living alone while still home-dwelling, indicating that this group might be especially vulnerable for malnutrition. The linear regression model identified two risk factors for weight loss: age and agitation. While age is a factor beyond intervention, agitation might be approached by trying to identify underlying and treatable health issues like for example pain or urinary tract infections, or by targeted interventions to decrease agitation itself. In residents with overweight, we see that almost one third is classified as obese after one year of NH stay, increasing the share of residents with obesity from 10 to $14\%$. Additionally, we observed that residents who suffered from obesity at BL, were the group with no significant change in BMI. A previous study reported a prevalence for obesity of $18\%$ in a French NH population, with no weight change during the two-month observation period, despite dietary regimes. As in our study, pain was more frequently reported in residents with obesity as compared to the other BMI groups [33]. As described above, obesity represents a risk factor for NH admission, due to the general health risk of overweight and due to increased functional decline, resulting in both a higher rate of NH admissions and at a younger age. However, there has not been observed increased NH mortality, resulting in a longer NH stay for this patient group [4, 29]. As obesity is an increasing public health problem, also in Norway, this might enhance the foreseen need for more NH beds in the decades to come [34, 35]. This raises the question, whether NHs should implement interventions that aim at weight control. According to the current ESPEN guidelines on nutrition and hydration in geriatrics, interventions to lose body weight are not recommended for older people with overweight [36]. Weight loss interventions are only considered beneficial when combined with exercise to retain muscle mass. However, randomized controlled trials with nursing home residents are lacking and most studies include people between 65 and 70 years without severe functional limitations. ## Strengths and limitations of the study We followed a cohort of 1,044 participants in a longitudinal study from NH admission and during the first year of their long-term NH stay, with clinical examinations at BL and after 12 months. For clinical, prospective studies in the NH setting cohorts of about a thousand participants are scarce. The study cohort consisted of NH residents from both urban and rural municipalities. High quality of the data collection was secured by a standardized interview carried out by healthcare workers with adequate training under the supervision of research nurses. Furthermore, the Norwegian health and social system provides a rather homogenous environment for health service research as there are hardly any private sector healthcare providers on the market. The decision of NH admission is made by care workers in the municipality administration, mainly based on the patients’ functional status and with comparable thresholds for admission. Still, the main weakness of this study is that our sample might not be representative of the general NH population in Norway, as less than half of all eligible residents participated in REDIC-NH and SAM-AKS III, and as the stricter inclusion criteria of the present study resulted in an additional attrition of study participants. Further, the BMI was only assessed at two time points one year apart. Thus, we know nothing about the changes in BMI of the deceased participants. The nutritional state was solely assessed by BMI, as we lacked information about comorbidities or possible problems with food intake. The BMI does not always represent the full nutritional status of an older person, as it does not necessarily reflect changes in a person’s percentage of muscle mass and body fat. This might have led to an underestimation of malnutrition. ## Conclusion Malnutrition is a common health challenge at NH admission, with one third of NH residents being moderately to severely underweight and $10\%$ being obese. However, during the first year of NH stay, there was a favourable development for underweight NH residents, as they increased their BMI, and $43.6\%$ changed to a higher weight classification, while we observed no changes in the BMI in residents with obesity. As NH residents are in the last phase of their lives, interventions to prevent malnutrition or overweight should be initiated while still home-dwelling, and then continued in the nursing homes. ## References 1. 1.Maleta K, Undernutrition. Malawi Med J. 2006 Dec;18(4):189–205. PMID: 27529011; PMCID: PMC3345626. 2. 2.WHO. : Obesity and overweight. https://www.who.int/news-room/fact-sheets/detail/obesity-and-overweight. 9 June 2021. (Accessed 19 April 2022) 3. 3.Norwegian Department of Health (Helsedirektoratet). Nasjonale faglige retningslinjer for forebygging og behandling av underernæring. 2.opplag/09/2010 sist endret (pdf). april 2013. https://www.helsedirektoratet.no/retningslinjer/underernaering/Forebygging%20og%20behandling%20av%20underern%C3%A6ring%20%E2%80%93%20Nasjonal%20faglig%20retningslinje.pdf/_/attachment/inline/a69263ff-b40e-4d23-a7c6-67acce28a88e:48ba11770eaa7f7a1d18145aa0e1f30086ca5dcd/Forebygging%20og%20behandling%20av%20underern%C3%A6ring%20%E2%80%93%20Nasjonal%20faglig%20retningslinje.pdf. (Accessed 19 April 2022) 4. 4.Harris JA, Castle NG. Obesity and Nursing Home Care in the United States: A Systematic Review. Gerontologist. 2019 May 17;59(3):e196-e206. doi: 10.1093/geront/gnx128. PMID: 29253135; PMCID: PMC6524472. 5. 5.Vossius C, Selbæk G, Šaltytė Benth J, Bergh S. Mortality in nursing home residents: A longitudinal study over three years. PLoS One. 2018 Sep 18;13(9):e0203480. doi: 10.1371/journal.pone.0203480. PMID: 30226850; PMCID: PMC6143238. 6. 6.Helvik AS, Engedal K, Benth J, Selbæk G. Prevalence and Severity of Dementia in Nursing Home Residents. Dement Geriatr Cogn Disord. 2015;40(3–4):166 – 77. doi: 10.1159/000433525. Epub 2015 Jul 2. Erratum in: Dement Geriatr Cogn Disord. 2015;40(3–4):177. PMID: 26138271 7. 7.Alzheimer’s Disease International (ADI). Nutrition and dementia. London. February 2014. Reprinted. October 2014. https://www.alzint.org/u/nutrition-and-dementia.pdf. (Accessed 19 April 2022) 8. 8.Roberts S, Collins P, Rattray M. Identifying and Managing Malnutrition, Frailty and Sarcopenia in the Community: A Narrative Review.Nutrients. 2021 Jul5;13(7):2316. doi: 10.3390/nu13072316. PMID: 34371823; PMCID: PMC8308465. 9. 9.Tamura BK, Bell CL, Masaki KH, Amella EJ. Factors associated with weight loss, low BMI, and malnutrition among nursing home patients: a systematic review of the literature. J Am Med Dir Assoc. 2013 Sep;14(9):649 – 55. doi: 10.1016/j.jamda.2013.02.022. PMID: 23639716. 10. 10.Bell CL, Tamura BK, Masaki KH, Amella EJ. Prevalence and measures of nutritional compromise among nursing home patients: weight loss, low body mass index, malnutrition, and feeding dependency, a systematic review of the literature. J Am Med Dir Assoc. 2013 Feb;14(2):94–100. doi: 10.1016/j.jamda.2012.10.012. PMID: 23246236. 11. 11.Borda MG, Venegas-Sanabria LC, Garcia-Cifuentes E, Gomez RC, Cano-Gutierrez CA, Tovar-Rios DA, Aarsland V, Khalifa K, Jaramillo-Jimenez A, Aarsland D, Soennesyn H. Body mass index, performance on activities of daily living and cognition: analysis in two different populations. BMC Geriatr. 2021 Mar 12;21(1):177. doi: 10.1186/s12877-021-02127-8. PMID: 33711937; PMCID: PMC7953600. 12. 12.Bergland AK, Dalen I, Larsen AI, Aarsland D, Soennesyn H. Effect of Vascular Risk Factors on the Progression of Mild Alzheimer’s Disease and Lewy Body Dementia. J Alzheimers Dis. 2017;56(2):575–584. doi: 10.3233/JAD-160847. PMID: 28035932. 13. 13.Cao L, Tan L, Wang HF, Jiang T, Zhu XC, Lu H, Tan MS, Yu JT. Dietary Patterns and Risk of Dementia: a Systematic Review and Meta-Analysis of Cohort Studies. Mol Neurobiol. 2016 Nov;53(9):6144–6154. doi: 10.1007/s12035-015-9516-4. PMID: 26553347. 14. 14.Bhaskaran K, Dos-Santos-Silva I, Leon DA, Douglas IJ, Smeeth L. Association of BMI with overall and cause-specific mortality: a population-based cohort study of 3·6 million adults in the UK. Lancet Diabetes Endocrinol. 2018 Dec;6(12):944–53. 10.1016/S2213-8587(18)30288-2. PMID: 30389323; PMCID: PMC6249991. 15. 15.Borda MG, Ayala Copete AM, Tovar-Rios DA, Jaramillo-Jimenez A, Giil LM, Soennesyn H, Gómez-Arteaga C, Venegas-Sanabria LC, Kristiansen I, Chavarro-Carvajal DA, Caicedo S, Cano-Gutierrez CA, Vik-Mo A, Aarsland D. Association of Malnutrition with Functional and Cognitive Trajectories in People Living with Dementia: A Five-Year Follow-Up Study. J Alzheimers Dis. 2021;79(4):1713–1722. doi: 10.3233/JAD-200961. PMID: 33459715. 16. 16.Røen I, Selbæk G, Kirkevold Ø, Engedal K, Testad I, Bergh S. Resourse Use and Disease Couse in dementia - Nursing Home (REDIC-NH), a longitudinal cohort study; design and patient characteristics at admission to Norwegian nursing homes.BMC Health Serv Res. 2017 May22;17(1):365. doi: 10.1186/s12913-017-2289-x. PMID: 28532443; PMCID: PMC5441072. 17. 17.Helvik AS, Bergh S, Šaltytė Benth J, Selbaek G, Husebo BS, Tevik K. Pain in nursing home residents with dementia and its association to quality of life. Aging Ment Health. 2021 Jul 11:1–11. doi: 10.1080/13607863.2021.1947968. Epub ahead of print. PMID: 34251936. 18. 18.Working Group. GLIM criteria for the diagnosis of malnutrition - A consensus report from the global clinical nutrition community. J Cachexia Sarcopenia Muscle. 2019 Feb;10(1):207–17. 10.1002/jcsm.12383. PMID: 30920778; PMCID: PMC6438340. 19. 19.Hughes CP, Berg L, Danziger WL, Coben LA, Martin RL. A new clinical scale for the staging of dementia. Br J Psychiatry. 1982 Jun;140:566 – 72. doi: 10.1192/bjp.140.6.566. PMID: 7104545. 20. 20.O’Bryant SE, Waring SC, Cullum CM, Hall J, Lacritz L, Massman PJ, Lupo PJ, Reisch JS, Doody R. ; Texas Alzheimer’s Research Consortium. Staging dementia using Clinical Dementia Rating Scale Sum of Boxes scores: a Texas Alzheimer’s research consortium study.Arch Neurol. 2008Aug;65(8):1091–5. doi: 10.1001/archneur.65.8.1091. PMID: 18695059; PMCID: PMC3409562. 21. 21.Cummings JL, Mega M, Gray K, Rosenberg-Thompson S, Carusi DA, Gornbein J. The Neuropsychiatric Inventory: comprehensive assessment of psychopathology in dementia. Neurology. 1994 Dec;44(12):2308-14. doi: 10.1212/wnl.44.12.2308. PMID: 7991117. 22. 22.Selbaek G, Kirkevold O, Sommer OH, Engedal K. The reliability and validity of the Norwegian version of the Neuropsychiatric Inventory, nursing home version (NPI-NH).Int Psychogeriatr. 2008Apr;20(2):375–82. doi: 10.1017/S1041610207005601. 23. 23.Lawton MP, Brody EM. Assessment of older people: self-maintaining and instrumental activities of daily living. Gerontologist. 1969 Autumn;9(3):179 – 86. PMID: 5349366. 24. 24.Lyketsos CG, Galik E, Steele C, Steinberg M, Rosenblatt A, Warren A, Sheppard JM, Baker A, Brandt J. The General Medical Health Rating: a bedside global rating of medical comorbidity in patients with dementia. J Am Geriatr Soc. 1999 Apr;47(4):487 – 91. doi: 10.1111/j.1532-5415.1999.tb07245.x. PMID: 10203127. 25. 25.Husebo BS, Strand LI, Moe-Nilssen R, Husebo SB, Snow AL, Ljunggren AE. Mobilization-Observation-Behavior-Intensity-Dementia Pain Scale (MOBID): development and validation of a nurse-administered pain assessment tool for use in dementia. J Pain Symptom Manage. 2007 Jul;34(1):67–80. doi: 10.1016/j.jpainsymman.2006.10.016. PMID: 17509814. 26. 26.Blomqvist N. On the relationship between change and initial value.Journal of the American Statistical Association, 72(360), 1977 27. 27.Chiolero A, Paradis G, Rich B, Hanley JA. Assessing the Relationship between the Baseline Value of a Continuous Variable and Subsequent Change Over Time.Front Public Health. 2013 Aug23;1:29. doi: 10.3389/fpubh.2013.00029. PMID: 24350198; PMCID: PMC3854983. 28. 28.Törmä J, Winblad U, Cederholm T, Saletti A. Does undernutrition still prevail among nursing home residents? Clin Nutr. 2013 Aug;32(4):562-8. doi: 10.1016/j.clnu.2012.10.007. PMID: 23137706. 29. 29.Grabowski DC, Campbell CM, Ellis JE. Obesity and mortality in elderly nursing home residents. J Gerontol A Biol Sci Med Sci. 2005 Sep;60(9):1184-9. doi: 10.1093/gerona/60.9.1184. PMID: 16183961. 30. 30.tb Nyheter. Pappa sultet og tørstet i hjel. 8.12.14 https://www.tb.no/nyheter/pappa-sultet-og-torstet-i-hjel/s/5-76-3122. (Accessed 31.01.23) 31. 31.Stavanger Aftenblad. Astrid tar med seg wienerbrød når hun besøker moren (97) på sykehjemmet. 07.07.22 https://www.aftenbladet.no/lokalt/i/dnX32q/astrid-tar-med-seg-wienerbroed-naar-hun-besoeker-moren-97-paa-sykehjemmet. (Accessed 31.01.23) 32. 32.Pensjonistforbundet. Skremmende tall om underernæring på sykehjem. 29.09.21. https://www.pensjonistforbundet.no/nyheter/skremmende-tall-om-underern%C3%A6ring-p%C3%A5-sykehjem. (Accessed 31.01.23) 33. de Souto Barreto P, Zanandrea V, Lapeyre-Mestre M, Cesari M, Vellas B, Rolland Y. Obesity in nursing home residents: a cross-sectional study. Jour Nurs Home Res (2015.0) 1 6-10 34. 34.Norwegian Institute for Public Health (Folkehelseinstituttet). : Overvekt og fedme blant voksne (Overweight and obesity in adults). Last updated 16 Feb 2021. https://www.fhi.no/fp/overvekt/ (Accessed 19 April 2022). 35. 35.Vossius C, Selbæk G, Ydstebo A, Benth JS, Godager G, Luraas H. Ressursbruk og sykdomsforløp ved demens (REDIC). Sykehuset Innlandet, Report. 2015:1-155. https://sykehuset-innlandet.no/Documents/Rapport%20REDIC-videref%C3%B8ring.pdf. (Accessed 11.May 2022) 36. 36.Volkert D, Beck AM, Cederholm T, Cruz-Jentoft A, Goisser S, Hooper L, Kiesswetter E, Maggio M, Raynaud-Simon A, Sieber CC, Sobotka L, van Asselt D, Wirth R, Bischoff SC. ESPEN guideline on clinical nutrition and hydration in geriatrics. Clin Nutr. 2019 Feb;38(1):10–47. Epub 2018 Jun 18. PMID: 30005900.
--- title: HbA1c is a predictive factor of severe coronary stenosis and major adverse cardiovascular events in patients with both type 2 diabetes and coronary heart disease authors: - Xiaojuan Jiao - Qin Zhang - Ping Peng - Yunfeng Shen journal: Diabetology & Metabolic Syndrome year: 2023 pmcid: PMC10026512 doi: 10.1186/s13098-023-01015-y license: CC BY 4.0 --- # HbA1c is a predictive factor of severe coronary stenosis and major adverse cardiovascular events in patients with both type 2 diabetes and coronary heart disease ## Abstract ### Background Coronary heart disease (CHD) is not only a macrovascular complication of type 2 diabetes mellitus (T2DM). Cardiovascular disease (CVD) is one of the leading causes of mortality among individuals with T2DM. Reducing the risk of adverse cardiovascular events (MACE) is crucial for the management of patients with CHD. This study aimed to investigate the effect of glycemic control on CHD severity and 3-point MACE (3p-MACE) risk in patients with T2DM and CHD. ### Methods 681 patients with both T2DM and CHD throughout October 2017 and October 2021 who were hospitalized in the second affiliated hospital of Nanchang university were included. A total of 300 patients were eventually enrolled in this retrospective cohort research. The severity of CHD in these patients was assessed, and the primary outcome during follow-up was recorded, with the primary result being the 3-point major adverse cardiovascular event (3p-MACE). The correlation between baseline glycated hemoglobin A1c (b-HbA1c) and the severity of CHD was evaluated by logistic regression analysis. The effect of b-HbA1c and follow-up HbA1c (f-HbA1c) levels on the risk of 3p-MACE were investigated by cox regression analysis. ### Results b-HbA1c was positively correlated with the severity of CHD ($r = 0.207$, $$p \leq 0.001$$), and patients with b-HbA1c > $9\%$ were more likely to have severe CHD. The HRs for b-HbA1c and f-HbA1c on the risk of 3p-MACE were 1.24 ($95\%$ CI 0.94–1.64, $$p \leq 0.123$$) and 1.32 ($95\%$ CI 1.02–1.72, $$p \leq 0.036$$), respectively. Patients with f-HbA1c ≥$8.6\%$ had a higher risk of 3p-MACE than f-HbA1c < $8.6\%$ (HR = 1.79, $95\%$ CI 1.16–2.79, $$p \leq 0.009$$). ### Conclusion In patients with both T2DM and CHD, b-HbA1c was an independent predictive factor of severe CHD. f-HbA1c was an independent predictive factor of 3p-MACE. Having the f-HbA1c below $8.6\%$ significantly reduced the risk of 3p-MACE. ### Supplementary Information The online version contains supplementary material available at 10.1186/s13098-023-01015-y. ## Background Diabetes mellitus (DM) is a group of metabolic diseases characterized by hyperglycemia caused by inadequate or impaired insulin secretion or utilization. The global prevalence of DM is increasing at an alarming rate, with the number of adults with DM reaching 537 million ($10.5\%$) worldwide in 2021, an increase of 74 million compared to 2019. The total number of people with DM is expected to increase to 643 million ($11.3\%$) and 783 million ($12.2\%$) worldwide by 2030 and 2045, respectively [1]. Type 2 diabetes mellitus (T2DM) accounts for around $90\%$ of people with DM. Hyperglycaemia in T2DM can damage large vessels, and cardiovascular disease is the most common complication of T2DM with a high incidence [2]. Coronary Heart Disease (CHD) is a macrovascular complication in T2DM patients. It is manifested by atherosclerosis (AS) or (and) spasms of the coronary arteries, resulting in narrowing and occlusion of the cardiovascular lumen and inadequate or interrupted blood supply to the myocardium. Coronary angiography (CAG) is the gold standard for diagnosing CHD. Patients with both T2DM and CHD tend to have diffuse, occlusive lesions throughout the coronary arteries, with a high risk of bleeding, ulceration, and calcification. Approximately two-thirds of DM patients die from cardiovascular disease and stroke [3]. Reducing the risk of adverse cardiovascular events (MACE) is crucial for the management of patients with CHD. Glycosylated hemoglobin A1c (HbA1c) is a marker of the average blood glucose level over the past 8–12 weeks. HbA1c has less variability, is unaffected by acute factors such as stress and exercise, and better represents relative long-term glycemic control [4]. With elevated HbA1c, the relative risk of complications in T2DM increases significantly, including neuropathy, retinopathy, and microangiopathy. Although studies have demonstrated that intensive glycemic control can reduce the occurrence of microangiopathy in T2DM patients, it remains controversial in reducing macrovascular complications and improving clinical outcomes. [ 5, 6] This study aimed to investigate the effect of HbA1c at different times on CHD severity and MACE risk in patients with both T2DM and CHD and guide clinicians to appropriate glucose management in these patients. ## Study population Consecutive hospitalized T2DM patients diagnosed with CHD by CAG at the Second Affiliated Hospital of Nanchang University from October 2017 to October 2021 were included through the hospital medical database. CHD patients included stable angina, unstable angina, myocardial infarction (MI), and those who underwent emergent PCI. This study excluded patients with malignant tumors, renal failure requiring hemodialysis, previous diagnosis of CHD, or other heart diseases without pre-CAG HbA1c (b-HbA1c). Follow-up was performed after CAG, and patients lost to follow-up were excluded. For assessing the level of HbA1c at follow-up, subjects without at least three HbA1c measurements during follow-up were also excluded. ## Definitions Diagnostic criteria for T2DM refer to WHO criteria [7]. Diagnostic criteria for CHD refer to ACC/AHA coronary angiography guidelines: The severity of coronary artery stenosis is assessed by CAG, and CHD is diagnosed when one or more of the coronary arteries have luminal stenosis of more than $50\%$ [8]. Gensini score is a widely used angiographic scoring system for quantifying the severity of CHD [9]. The Gensini score was calculated by two clinicians independently, under the guidance of a cardiologist, following the Gensini score guidelines [10]. Step 1 is to calculate the severity score of the coronary lesion, step 2 is to apply a multiplication factor to each lesion score based upon the location of the lesion in the coronary tree, and step 3 is to sum the lesion severity scores. The detailed calculation of the Gensini score was shown in Additional file 1: Table S1. We were prespecified that the patients would be divided into three groups based on tertiles of the Gensini score (T1: < 29; T2: 29–72; T3: > 73) and defined these three groups as mild, moderate, and severe CHD. To assess the glycemic control of the patients, b-HbA1c was divided into < $6\%$, $6\%$–$7\%$, $7\%$–$8\%$, $8\%$–$9\%$, and ≥ $9\%$ groups according to the different levels. ## Baseline data collection During hospitalization for CAG, baseline characteristics were collected. Demographic information such as age, sex, and body mass index (BMI) was written. Medical histories such as hypertension, stroke, and related medication were recorded. Laboratory tests such as HbA1c, fasting serum glucose (FSG), and fasting total cholesterol (TC) were analyzed. Coronary angiography results and type of coronary artery disease were consulted. HbA1c was measured by High-Performance Liquid Chromatography (HPLC). Some patients would be excluded if they have any condition shortening erythrocyte survival or decreasing mean erythrocyte age (e.g., recovery from acute blood loss, hemolytic anemia, chronic kidney disease). ## Follow-up outcomes and data Follow-up was begun after patients underwent CAG and ended in February 2022. Patients were followed up via the electronic medical record system or telephone, and laboratory tests and outcome events during the follow-up period were recorded. The primary outcome was the 3-points MACE (3p-MACE), defined as cardiovascular death, nonfatal stroke, and nonfatal myocardial infarction. Cardiovascular death was defined as death attributable to an ischemic cardiovascular cause like fatal MI, stroke, or sudden death secondary to a presumed cardiovascular cause in this high‐risk population [11]. The definition of nonfatal stroke refers to ischemic and hemorrhagic strokes [12]. Nonfatal myocardial infarction is an acute myocardial infarction (AMI) that does not result in death. AMI is the presence of acute myocardial injury detected by abnormal cardiac biomarkers in the setting of evidence of acute myocardial ischemia [11]. Laboratory indicators for follow-up included HbA1c, TC, Triglyceride (TG), High-density lipoprotein cholesterol (HDL-C), Low-density lipoprotein cholesterol (LDL-C), and microalbuminuria (MAU). ## Statistical analyses Continuous variables were expressed as the mean ± standard deviation, and categorical variables were expressed as frequencies and percentages. Baseline and follow-up demographic and clinical characteristics were compared with the Pearson χ2 test for categorical variables and analysis of variance for continuous variables. Correlations between two continuous variables were analyzed using Spearman's correlation analysis. The correlation between two ranked variables was assessed using Spearman's Rank Correlation. Patients with mild and moderate CHD were categorized into non-severe CHD groups, thus dividing patients into severe and non-severe CHD groups. Binary logistic regression and ordered logistic regression analyses were used to assess the association of b-HbA1c as a continuous and categorical variable with severe CHD, respectively. To evaluate other risk factors of the severity of coronary artery disease, we included the baseline indicators in Table 1 in logistic regression one by one. The independent risk of HbA1c (b-HbA1c and f-HbA1c) and insulin therapy for 3p-MACE was assessed by age, sex, and history of hypertension in a multivariate cox proportional risk model. Interactions between f-HbA1c and insulin therapy on 3p-MACE outcomes were assessed in a cox model. For each patient, person-months of follow-up were counted from the date of diagnosis of CHD to the date of diagnosis of 3p-MACE or February 2022, whichever came first. Therefore, AMI and stroke leading to death were not counted in the 17 nonfatal strokes and 15 nonfatal infarction events. To analyze the effect of f-HbA1c on 3p-MACE, cumulative event incidence estimates were plotted according to f-HbA1c levels using the Kaplan–Meier technique. Differences between event-free curves were assessed with the log-rank test. Table 1Baseline demographic, clinical, and angiographic data of patientsNOverallNo-event3p-mace eventP value30026931Demographics Age, years65.3 ± 10.265.2 ± 10.165.9 ± 11.10.708 Male193 ($64.3\%$)176 ($65.4\%$)17 ($54.8\%$)0.244 BMI (kg/m2)24.9 ± 2.424.9 ± 2.424.9 ± 2.50.975 Systolic BP, mmHg133.6 ± 22.3133.2 ± 22.6137.5 ± 19.20.316 Diastolic BP, mmHg77.7 ± 13.577.7 ± 13.678.2 ± 12.00.838 Heart Rate,82.1 ± 16.481.8 ± 16.584.9 ± 15.50.314Medical history Diabetes duration, years9.8 ± 7.09.7 ± 7.010.9 ± 6.80.366 Hypertension206 ($68.7\%$)180 ($66.9\%$)26 ($83.9\%$)0.054 Stroke28 ($9.3\%$)24 ($8.9\%$)4 ($12.9\%$)0.471 ACEI/ARB92 ($30.7\%$)80 ($29.7\%$)12 ($38.7\%$)0.305 Insulin98 ($32.7\%$)84 ($31.2\%$)4 ($45.2\%$)0.117Laboratory values HbA1c, (%)8.0 ± 1.87.9 ± 1.78.7 ± 2.40.013 Fasting glucose, mmol/L9.2 ± 4.29.1 ± 3.910.4 ± 5.80.095 TC, mmol/L4.5 ± 1.24.5 ± 1.24.6 ± 1.20.881 TG, mmol/L1.9 ± 1.21.8 ± 1.22.0 ± 1.40.514 HDL-C, mmol/L1.1 ± 0.31.0 ± 0.31.1 ± 0.40.080 LDL-C, mmol/L2.8 ± 0.92.8 ± 0.92.7 ± 0.80.629 eGFR, mL/min/1.73 m283.4 ± 38.884.2 ± 39.276.1 ± 35.10.268 BNP (pg/ml)391.1 ± 680.8395.1 ± 708.4356.8 ± 366.60.767 LVEF, %59.6 ± 13.959.7 ± 14.358.9 ± 9.70.762Angiographic data Gensini score55.6 ± 41.554.8 ± 41.562.5 ± 41.20.328 Myocardial infarction102 ($34.0\%$)92 ($34.2\%$)10 ($32.3\%$)0.829 PCI179 ($59.7\%$)160 ($59.5\%$)19 ($61.3\%$)0.846Severity of CHD0.580 Mild92 ($30.7\%$)84 ($31.2\%$)8 ($25.8\%$) Moderate70 ($23.3\%$)64 ($23.8\%$)6 ($19.4\%$) Sever138 ($46.0\%$)121 ($45.0\%$)17 ($54.8\%$)Values given as mean ± SD or number (percentage), ACEI: angiotensin-converting enzyme inhibitor, ARB: Angiotensin Receptor Blocker, TC: Total Cholesterol, TG: Triglyceride, HDL-C: High-Density Lipoprotein Cholesterol, LDL-C: Low-Density Lipoprotein Cholesterol, eGFR: estimated Glomerular Filtration Rate, LVEF: Left Ventricular Ejection Fraction, PCI: Percutaneous Coronary Intervention Two-tailed p values < 0.05 were considered significant. SPSS software (IBM SPSS Statistics 23) and R software (R 3.6.1) were used for statistical analysis. ## Baseline characteristics and follow-up data of patients A total of 681 patients with both T2DM and CHD were included through the hospital medical database. Patients who had malignant tumors ($$n = 5$$), renal failure requiring hemodialysis ($$n = 6$$), previous diagnosis of CHD ($$n = 145$$), or other heart diseases ($$n = 3$$) without b-HbA1c ($$n = 68$$) were excluded. During the follow-up, 32 patients were lost to follow-up, and patients ($$n = 122$$) without at least three HbA1c measurements were excluded. Ultimately, 300 subjects were included in the study analysis. ( Fig. 1) The mean age of the patients was 65 years, the duration of diabetes was approximately 10 years, the mean b-HbA1c was about $8.0\%$, 193 ($64.3\%$) of them were male, 206 ($68.7\%$) patients had hypertension, 138 ($46.0\%$) patients had severe CHD, 102 ($34.0\%$) patients had acute myocardial infarction (AMI), and 179 ($59.7\%$) patients underwent percutaneous coronary intervention (PCI) (Table 1).Fig. 1Flow chart of the recruitment procedure. T2DM: type 2 diabetes mellitus; CHD: coronary heart disease; CAG: coronary angiography The mean follow-up period was 24 months, and 31 patients ($10.3\%$) experienced 3p-MACE (11 cardiovascular death, 17 nonfatal strokes, and 15 nonfatal infarction). Compared with the event-free patient group, b-HbA1c levels ($8.7\%$ vs$7.9\%$, $$p \leq 0.013$$) and f-HbA1c ($8.6\%$ vs $7.8\%$, $$p \leq 0.008$$) were significantly higher in the 3p-MACE group (Table 2).Table 2Follow-up Laboratory data of patientsNOverallNo-event3P-MACEP value30026931Laboratory values f-HbA1c, (%)7.9 ± 1.77.8 ± 1.68.6 ± 2.30.008 f-TC, mmol/L3.9 ± 1.03.9 ± 1.04.0 ± 1.00.655 f-TG, mmol/L1.6 ± 0.91.6 ± 0.91.7 ± 1.10.510 f-HDL-C, mmol/L1.1 ± 0.31.1 ± 0.31.1 ± 0.30.394 f-LDL-C, mmol/L2.9 ± 5.42.9 ± 5.62.8 ± 3.70.957 f-MAU, mg/L133.2 ± 282.5126.9 ± 247.3187.5 ± 495.90.259f: follow-up ## Distribution characteristics of b-HbA1c and CHD severity Among these patients, $26.0\%$ had a b-HbA1c ≥ $9\%$, next to those with b-HbA1c $6\%$–$7\%$ ($27.3\%$) (Fig. 2A). The group with mild CHD had the highest proportion of patients with a b-HbA1c in the $6\%$–$7\%$ range ($34.8\%$), whereas the group with severe CHD had the highest proportion of patients with a b-HbA1c in the ≥ $9\%$ range ($32.6\%$) (Fig. 2B). As b-HbA1c levels increased, Gensini scores tended to increase. With patients in the b-HbA1c < $6\%$, $6\%$–$7\%$, $7\%$–$8\%$ and $8\%$–$9\%$ ranges all having significantly lower Gensini scores than those with b-HbA1c ≥ $9\%$ ($p \leq 0.05$) (Fig. 2C). The severity of CHD increased gradually with b-HbA1c. Compared to the severe CHD group, patients in the mild CHD group had significantly lower b-HbA1c values ($p \leq 0.001$) (Fig. 2D).Fig. 2Distribution characteristics of b-HbA1c and severity of CHD. A The proportion of different levels of b-HbA1c in patients with both T2DM and CHD; B Percentage of b-HbA1c at different levels in patients with mild, moderate, and severe coronary stenosis, respectively; C: Differences in Gensini score values between different levels of b-HbA1c groups; D: Differences in b-HbA1c values between different grades of coronary stenosis groups; E: Smoothed curve fit for b-HbA1c and Gensini scores; $P \leq 0.05$, $P \leq 0.01$, *$P \leq 0.001$, ns = no significance The distribution characteristics of b-HbA1c and CHD severity in the patients revealed a strong correlation. By plotting the curve fit between b-HbA1c and the Gensini score (Fig. 2E), it could be seen that the two were roughly positively correlated. When b-HbA1c was used as a continuous variable, b-HbA1c was significantly associated with Gensini score values ($r = 0.207$, $$p \leq 0.001$$). When b-HbA1c was used as a categorical variable, b-HbA1c was significantly positive correlated with CHD severity (rs = 0.180, $$p \leq 0.002$$). Additionally, duration of diabetes, fasting glucose, LDL-C, and BNP were risk factors for severe CHD, as shown in Additional file 1: Table S2. ## Baseline -HbA1c is an independent predictive factor of severe CHD Univariate logistic regression showed that b-HbA1c was a risk factor for severe CHD in patients with T2DM (OR = 1.149, $$p \leq 0.037$$). Multivariate logistic regression further demonstrated that b-HbA1c was an independent risk factor for severe CHD in patients with T2DM (OR = 1.151, $$p \leq 0.046$$) (Fig. 3). The risk of severe CHD increased by approximately $15\%$ for every $1\%$ increase in b-HbA1c value. Therefore, b-HbA1c could be considered a predictor of having severe CHD.Fig. 3Forest plots of logistic regression analysis Odds ratios with $95\%$ CI for severe stenosis and Forest plots of cox regression analysis hazard ratios with $95\%$ CI for 3p-MACE.The following items were included as covariates in the multifactorial logistic regression analysis: demographic factors (age, sex, BMI), medical history (hypertension, stroke, diabetes duration), laboratory tests on admission (FBG, LDL-C, HDL-C, TG, CHOL, eGFR, BNP level) in the multifactorial logistic regression analysis. Age, sex and history of hypertension were adjusted in the multifactorial cox regression analysis. &: adjusted the interaction item Insulin f-HbA1c ## Baseline -HbA1c as a risk factor for major adverse cardiovascular events Univariate cox regression analysis showed that b-HbA1c was a risk factor for the prognosis of 3p-MACE in T2DM patients with CHD (HR = 1.25, $95\%$ CI 1.05–1.49, $$p \leq 0.012$$). However, it was not found to be an independent risk factor by multifactorial analysis (HR = 1.24, $95\%$ CI 0.94–1.64, $$p \leq 0.123$$) (Fig. 3). A subgroup analysis of the study population was performed to identify factors influencing the association between b-HbA1c and the risk of prevalence of 3p-MACE. It showed that b-HbA1c was not a risk factor for the 3p-MACE in some subgroups (e.g., age ≥ 65 years, hypertension, duration of diabetes ≥ 12 months, insulin treatment, PCI, severe CHD, etc. $p \leq 0.05$) (Additional file 1: Figure S1.) ## Follow-up HbA1c is an independent predictive factor of major adverse cardiovascular events By univariate (HR = 1.23, $95\%$ CI 1.04–1.45, $$p \leq 0.014$$) and multivariate cox regression analysis (HR = 1.32, $95\%$ CI 1.02–1.72, $$p \leq 0.036$$), f-HbA1c was found to be an independent risk factor for the 3p-MACE (Fig. 3). A curve fit of f-HbA1c to 3p-MACE risk was plotted (Fig. 4A), and a curve cut-off point of $8.6\%$ was found using threshold effects analysis. For patients with f-HbA1c ≥ $8.6\%$, f-HbA1c was an independent risk factor for 3p-MACE (HR = 1.79, $95\%$ CI 1.16–2.79, $$p \leq 0.009$$), whereas for patients with f-HbA1c < $8.6\%$, the association was not statistically significant (HR = 0.59, $95\%$ CI 0.34–1.03, $$p \leq 0.065$$) (Fig. 3). The cumulative incidence of 3p-MACE between groups with different f-HbA1c values was plotted (Fig. 4B). It showed a significant difference in cumulative event rates between the two groups. Those patients with f-HbA1c < $8.6\%$ had a significantly lower risk of 3p-MACE than those with f-HbA1c ≥ $8.6\%$ ($$p \leq 0.014$$).Fig. 4f-HbA1c and the risk of 3p-MACE. A Smoothed curve fit of f-HbA1c values and the risk of 3p-MACE; B Cumulative risk of 3p-MACE between groups with different f-HbA1c values. f-HbA1c GROUP 1 = f-HbA1c ≥ $8.6\%$, f-HbA1c GROUP 2 = f-HbA1c < $8.6\%$ ## Insulin treatment and prognosis of CHD patients Interestingly, insulin treatment was strongly associated with the risk of 3p-MACE in these T2DM with CHD patients by cox regression analysis. Compared with non-insulin treatment, insulin treatment had a higher risk of 3p-MACE (HR = 2.49, $95\%$ CI 1.14–5.42, $$p \leq 0.022$$) (Fig. 3). However, b-HbA1c and f-HbA1c were higher in the insulin-treated group than in the noninsulin-treated group (Table 3). Interactions between f-HbA1c and insulin therapy on 3p-MACE outcomes were assessed. The results show that the interaction was significant, and the P value for the interaction term was reported ($$p \leq 0.038$$). After including the interaction item of insulinf-HbA1c in the adjusted cox model, the HR for f-HbA1c on the risk of 3p-MACE was 1.18 ($95\%$CI 0.97–1.43, $$p \leq 0.090$$). The unadjusted HR was 1.23 ($95\%$CI 1.04–1.45, $$p \leq 0.014$$). For patients with insulin treatment, f-HbA1c was an independent risk factor for 3p-MACE (HR = 1.32, $95\%$ CI 1.01–1.73, $$p \leq 0.047$$), whereas for patients without insulin treatment, the association was not statistically significant (HR = 1.17, $95\%$ CI 0.91–1.51, $$p \leq 0.227$$). ( Fig. 3).Table 3HbA1c levels between insulin-treated and non-insulin-treated groupsNo Insulin treatmentInsulin treatmentZP valueb-HbA1c (%)7.7 (6.5–8.8)7.9 (6.8–9.6)− 1.9860.047f-HbA1c (%)7.4 (6.5–8.3)7.88 (6.9–9.4)− 2.0830.005b-HbA1c baseline HbA1c, f-HbA1c follow-up HbA1c ## Discussion This study found that patients with T2DM with poor glycemic control had a significantly increased risk of severe coronary stenosis compared to those with good glycemic control. b-HbA1c was a risk factor but not an independent risk factor for the 3p-MACE in T2DM with CHD patients. In contrast, f-HbA1c was an independent risk factor for 3p-MACE in these patients. 3p-MACE in patients with both T2DM and CHD was significantly decreased when f-HbA1c < $8.6\%$. T2DM promotes the development and progression of atherosclerosis, including not only traditional risk factors such as genetic factors, hyperglycemia, obesity, lipid metabolism disorders, sex hormone abnormalities, advanced age, and smoking but also nontraditional risk factors such as hyperinsulinemia, insulin resistance, diabetic hypercoagulable state, diabetic endothelial dysfunction, advanced glycosylation end products, oxidative stress, diabetic inflammation, microproteinuria, and hyperhomocysteinemia [13]. HbA1c is not only used to diagnose T2DM but can also be used to identify people at high cardiovascular risk. The higher the HbA1c level, the greater the risk of cardiovascular events in patients with T2DM [14]. Hyperglycaemia damages the cardiovascular system and induces atherosclerosis through several mechanisms, such as endothelial cell damage, oxidative stress, and imbalances in the coagulation and fibrinolytic systems, leading to diffuse coronary artery disease. Although previous studies have shown a positive correlation between high levels of HbA1c and the severity of coronary artery disease, these have been limited to specific types of coronary artery disease [14–16]. The risk of b-HbA1c levels to coronary artery disease in a population of patients not differentiated by CHD type is currently unknown. This study assessed the severity of coronary lesions based on CAG using the Gensini score and found a positive correlation between b-HbA1c levels and the severity of coronary lesions, in line with previous studies [17, 18]. The Gensini score is a scientific evaluation standard of coronary artery lesions, taking into account the number, location, and severity of stenosis of coronary artery lesions [10]. And it is a useful tool for assessing the severity of CHD [19]. Higher levels of b-HbA1c (HbA1c > $9\%$) were associated with more severe CHD when HbA1c was analyzed as a categorical variable, suggesting that elevated b-HbA1c predicts increased CHD severity. Many studies have examined the prognostic impact of baseline blood glucose levels on patients with CHD. Still, most have been limited to patients with AMI or PCI, suggesting a positive association between b-HbA1c and poor prognosis [20–26]. b-HbA1c significantly predicted adverse cardiovascular events at prognosis [27]. A Chinese population-based study also showed no significant difference in prognosis between groups with different b-HbA1c levels in patients with T2DM combined with CHD who underwent PCI [28]. This study found that b-HbA1c was a risk factor for 3p-MACE in T2DM with CHD patients but was not an independent risk factor and was influenced by other factors such as age, duration of diabetes, hypertension, etc. Therefore, in patients with both T2DM and CHD, b-HbA1c alone cannot be used as an indicator to predict the long-term risk of 3p-MACE. The prognosis of patients with CHD is closely related to glycemic control at follow-up. The poorer the glycemic control at follow-up and the need for insulin control, the higher the risk of developing 3p-MACE. It showed a higher incidence of 3p-MACE in the insulin-treated group. Meanwhile, there was relatively poorer glycemic control in the insulin-treated group. The DIGAMI-2 study in 2005 found that intensive glycemic control reduced mortality in heart attack patients [29]. The subsequent DIGAMI-1 in 2014 found that glycemic control by insulin therapy significantly reduced 1-year mortality in patients with acute infarction compared to the conventional treatment group. This finding contradicts the results of this study [30]. We found the risk of 3p-MACE in patients with both T2DM and CHD increased with the higher f-HbA1c in the insulin treatment group. In comparison, it was not significant in the patients without insulin treatment. Furthermore, 3p-MACE increased with higher glycemia when f-HbA1c ≥ $8.6\%$. The results of this study were consistent with the 2020 Chinese guidelines [31], which state that the recommendation of HbA1c < $8.0\%$ for T2DM patients with a long duration of diabetes, a history of cardiovascular disease, or a very high risk of cardiovascular disease is consistent. And patients with f-HbA1c values above the national $8.0\%$ threshold present an increased risk of 3p-MACE. The novelty of our study was that in addition to measuring the pre-CAG HbA1c as the level of the b-HbA1c value, the post-CAG HbA1c was also tested as the f-HbA1c. Thus, it reflects not only the relationship between glycemic control before CAG and CHD severity but also the impact of different stages of glycemic control on the prognosis of patients with both T2DM and CHD. Second, further threshold effect analyses were performed to determine which range of HbA1c was more likely to lead to severe CHD and adverse cardiovascular events and to guide clinicians in developing an individualized glycaemic management strategy for patients when b-HbA1c and f-HbA1c levels were found to correlate with the severity of coronary stenosis and the risk of 3p-MACE. However, there were some limitations in this study. As this was a retrospective cohort study, much of the data were obtained from hospital databases or by reviewing patients' medical records, and some of the data were missing. For example, there were no specific descriptions of smoking and drinking history, data on whether patients had been treated with statins before hospitalization were incomplete, and many patients did not have a urine albumin creatinine ratio, so these indicators were not included in the data analysis. Smoking [32], alcohol history [33], statin treatment [34], and urine microalbumin [35] have been shown to be strongly associated with cardiovascular disease, so the absence of such data at baseline may have had some impact on the results. As study excluded patients on tumor and renal failure dialysis at the time of inclusion, which significantly reduced the number of deaths of patients due to these causes, and only three deaths from other causes were observed at follow-up. So the cox regression model did not include death from other causes as a competing risk, which may have affected the study results. Furthermore, as the assessment of coronary angiograms could only be based on what was available in the medical records, it was not possible to score angiogram images according to the latest 2019 Gensini scoring criteria [36] or to use other scoring metrics, such as the SYNTAX score [37], to assess the severity of CHD. The use of a different scoring system may affect the results. ## Conclusions In conclusion, b-HbA1c was positively associated with the severity of CHD and was a risk factor for adverse cardiovascular events in T2DM with CHD patients, but not an independent risk factor. Whereas f-HbA1c was an independent risk factor. Hence b-HbA1c was an independent predictive factor of severe CHD, and f-HbA1c was an independent predictive factor of 3p-MACE. Patients with f-HbA1c above 8.6 were at the highest risk for 3p-MACE. ## Supplementary Information Additional file 1: Table S1. Step-by-step algorithm for the Gensini Score calculation. Table S2. Other risk factors of severe CHD. Figure S1. A subgroup analysis of the relationship between baseline HbA1c and risk of 3p-MACE. ## References 1. Sun H, Saeedi P, Karuranga S, Pinkepank M, Ogurtsova K, Duncan BB. IDF Diabetes Atlas: Global, regional and country-level diabetes prevalence estimates for 2021 and projections for 2045. Diabetes Res Clin Pract* (2022.0) 183 109119. DOI: 10.1016/j.diabres.2021.109119 2. Preis SR, Hwang SJ, Coady S, Pencina MJ, D'Agostino RB, Savage PJ. Trends in all-cause and cardiovascular disease mortality among women and men with and without diabetes mellitus in the Framingham Heart Study, 1950 to 2005. Circulation (2009.0) 119 1728-1735. DOI: 10.1161/CIRCULATIONAHA.108.829176 3. Kataoka Y, Yasuda S, Morii I, Otsuka Y, Kawamura A, Miyazaki S. Quantitative coronary angiographic studies of patients with angina pectoris and impaired glucose tolerance. Diabetes Care (2005.0) 28 2217-2222. DOI: 10.2337/diacare.28.9.2217 4. Sacks DB. A1C versus glucose testing: a comparison. Diabetes Care (2011.0) 34 518-523. DOI: 10.2337/dc10-1546 5. 5.Intensive blood-glucose control with sulphonylureas or insulin compared with conventional treatment and risk of complications in patients with type 2 diabetes (UKPDS 33). UK Prospective Diabetes Study (UKPDS) Group. Lancet (London, England). 1998;352(9131):837–53. 6. Patel A, MacMahon S, Chalmers J, Neal B, Billot L, Woodward M. Intensive blood glucose control and vascular outcomes in patients with type 2 diabetes. N Engl J Med (2008.0) 358 2560-2572. DOI: 10.1056/NEJMoa0802987 7. 7.World HO, International DFDefinition and diagnosis of diabetes mellitus and intermediate hyperglycaemia: report of a WHO/IDF consultation2006World Health OrganizationWorld Health Organization. Definition and diagnosis of diabetes mellitus and intermediate hyperglycaemia: report of a WHO/IDF consultation (2006.0) 8. 8.Scanlon PJ, Faxon DP, Audet AM, Carabello B, Dehmer GJ, Eagle KA, et al. ACC/AHA guidelines for coronary angiography A report of the American College of Cardiology/American Heart Association Task Force on practice guidelines (Committee on Coronary Angiography) Developed in collaboration with the Society for Cardiac Angiography and Interventions. J Am Coll Cardiol. 1999;33(6):1756–824. 9. Neeland IJ, Patel RS, Eshtehardi P, Dhawan S, McDaniel MC, Rab ST. Coronary angiographic scoring systems: an evaluation of their equivalence and validity. Am Heart J (2012.0) 164 547-52.e1. DOI: 10.1016/j.ahj.2012.07.007 10. Gensini GG. A more meaningful scoring system for determining the severity of coronary heart disease. Am J Cardiol (1983.0) 51 606. DOI: 10.1016/S0002-9149(83)80105-2 11. 11.Thygesen K, Alpert JS, Jaffe AS, Simoons ML, Chaitman BR, White HD. Writing Group on behalf of the Joint ESC/ACCF/AHA/WHF Task Force for the Universal Definition of Myocardial Infarction Third universal definition of myocardial infarction. Global Heart. 2012;7(4):275–95. 12. Judge C, Ruttledge S, Murphy R, Loughlin E, Gorey S, Costello M. Aspirin for primary prevention of stroke in individuals without cardiovascular disease-A meta-analysis. Int J Stroke (2020.0) 15 9-17. DOI: 10.1177/1747493019858780 13. Advances in the mechanisms of coronary atherosclerosis in diabetes. J Clin Pathol Res. (2009.0) 2009 5 14. Hong LF, Li XL, Guo YL, Luo SH, Zhu CG, Qing P. Glycosylated hemoglobin A1c as a marker predicting the severity of coronary artery disease and early outcome in patients with stable angina. Lipids Health Dis (2014.0) 13 89. DOI: 10.1186/1476-511X-13-89 15. Timmer JR, Hoekstra M, Nijsten MW, van der Horst IC, Ottervanger JP, Slingerland RJ. Prognostic value of admission glycosylated hemoglobin and glucose in nondiabetic patients with ST-segment-elevation myocardial infarction treated with percutaneous coronary intervention. Circulation (2011.0) 124 704-711. DOI: 10.1161/CIRCULATIONAHA.110.985911 16. Albashir AAD, Elawad O, Khougali MH. The use of glycosylated hemoglobin (HbA1c) as a predictor of the severity of acute coronary syndrome among diabetic patients. Ir J Med Sci (2021.0) 190 609-614. DOI: 10.1007/s11845-020-02341-0 17. Ravipati G, Aronow WS, Ahn C, Sujata K, Saulle LN, Weiss MB. Association of hemoglobin A(1c) level with the severity of coronary artery disease in patients with diabetes mellitus. Am J Cardiol (2006.0) 97 968-969. DOI: 10.1016/j.amjcard.2005.10.039 18. Saleem T, Mohammad KH, Abdel-Fattah MM, Abbasi AH. Association of glycosylated haemoglobin level and diabetes mellitus duration with the severity of coronary artery disease. Diab Vasc Dis Res (2008.0) 5 184-189. DOI: 10.3132/dvdr.2008.030 19. Wang KY, Zheng YY, Wu TT, Ma YT, Xie X. Predictive Value of gensini score in the long-term outcomes of patients with coronary artery disease who underwent PCI. Front Cardiovasc Med (2022.0) 24 778615. DOI: 10.3389/fcvm.2021.778615 20. Cederberg H, Saukkonen T, Laakso M, Jokelainen J, Härkönen P, Timonen M. Postchallenge glucose, A1C, and fasting glucose as predictors of type 2 diabetes and cardiovascular disease: a 10-year prospective cohort study. Diabetes Care (2010.0) 33 2077-2083. DOI: 10.2337/dc10-0262 21. Kassaian SE, Goodarzynejad H, Boroumand MA, Salarifar M, Masoudkabir F, Mohajeri-Tehrani MR. Glycosylated hemoglobin (HbA1c) levels and clinical outcomes in diabetic patients following coronary artery stenting. Cardiovasc Diabetol (2012.0) 11 82. DOI: 10.1186/1475-2840-11-82 22. Knapik P, Cieśla D, Filipiak K, Knapik M, Zembala M. Prevalence and clinical significance of elevated preoperative glycosylated hemoglobin in diabetic patients scheduled for coronary artery surgery. Eur J Cardio-thor Surg (2011.0) 39 484-489. DOI: 10.1016/j.ejcts.2010.07.037 23. Zhao X, Chang Mei H, Chen L, Jiang L, He M, Chen J. An increased level of haemoglobin A1C predicts a poorer clinical outcome in patients with acute pancreatitis. Clin Endocrinol (2012.0) 77 241-245. DOI: 10.1111/j.1365-2265.2011.04252.x 24. Gustafsson I, Kistorp CN, James MK, Faber JO, Dickstein K, Hildebrandt PR. Unrecognized glycometabolic disturbance as measured by hemoglobin A1c is associated with a poor outcome after acute myocardial infarction. Am Heart J (2007.0) 154 470-476. DOI: 10.1016/j.ahj.2007.04.057 25. Cicek G, Uyarel H, Ergelen M, Ayhan E, Abanonu GB, Eren M. Hemoglobin A1c as a prognostic marker in patients undergoing primary angioplasty for acute myocardial infarction. Coron Artery Dis (2011.0) 22 131-137. DOI: 10.1097/MCA.0b013e328342c760 26. Su G, Mi SH, Tao H, Li Z, Yang HX, Zheng H. Impact of admission glycemic variability, glucose, and glycosylated hemoglobin on major adverse cardiac events after acute myocardial infarction. Diabetes Care (2013.0) 36 1026-1032. DOI: 10.2337/dc12-0925 27. Ueda H, Mitsusada N, Harimoto K, Miyawaki M, Yasuga Y, Hiraoka H. Glycosylated hemoglobin is a predictor of major adverse cardiac events after drug-eluting stent implantation in patients with diabetes mellitus. Cardiology (2010.0) 116 51-57. DOI: 10.1159/000314331 28. Xl Hong. The influence of blood glucose level at different time point on PCI in patients with diabetes mellitus complicated with coronary heart diseases. J Clin Exp Med. (2009.0) 8 19-20 29. Malmberg K, Rydén L, Wedel H, Birkeland K, Bootsma A, Dickstein K. Intense metabolic control by means of insulin in patients with diabetes mellitus and acute myocardial infarction (DIGAMI 2): effects on mortality and morbidity. Eur Heart J (2005.0) 26 650-661. DOI: 10.1093/eurheartj/ehi199 30. Ritsinger V, Malmberg K, Mårtensson A, Rydén L, Wedel H, Norhammar A. Intensified insulin-based glycaemic control after myocardial infarction: mortality during 20 year follow-up of the randomised Diabetes Mellitus Insulin Glucose Infusion in Acute Myocardial Infarction (DIGAMI 1) trial. Lancet Diabetes Endocrinol (2014.0) 2 627-633. DOI: 10.1016/S2213-8587(14)70088-9 31. 31.Chinese Diabetes Society CMA, Endocrinology Branch. Expert consensus on glycated hemoglobin A 1c targets and management algorithm for Chinese adults with type 2 diabetes mellitus. Chin J Diabetes 2020;12(1):1–12. 32. Huxley RR, Woodward M. Cigarette smoking as a risk factor for coronary heart disease in women compared with men: a systematic review and meta-analysis of prospective cohort studies. Lancet (2011.0) 378 1297-1305. DOI: 10.1016/S0140-6736(11)60781-2 33. Biddinger KJ, Emdin CA, Haas ME, Wang M, Hindy G, Ellinor PT. Association of Habitual Alcohol Intake With Risk of Cardiovascular Disease. JAMA Netw Open (2022.0) 5 e223849. DOI: 10.1001/jamanetworkopen.2022.3849 34. Baigent C, Keech A, Kearney PM, Blackwell L, Buck G, Pollicino C. Efficacy and safety of cholesterol-lowering treatment: prospective meta-analysis of data from 90,056 participants in 14 randomised trials of statins. Lancet (2005.0) 366 1267-1278. DOI: 10.1016/S0140-6736(05)67394-1 35. Yudkin JS, Forrest RD, Jackson CA. Microalbuminuria as predictor of vascular disease in non-diabetic subjects. Islington Diabetes Survey. Lancet (London, England). (1988.0) 2 530-3. DOI: 10.1016/S0140-6736(88)92657-8 36. Rampidis GP, Benetos G, Benz DC, Giannopoulos AA, Buechel RR. A guide for Gensini Score calculation. Atherosclerosis (2019.0) 287 181-183. DOI: 10.1016/j.atherosclerosis.2019.05.012 37. Sianos G, Morel MA, Kappetein AP, Morice MC, Colombo A, Dawkins K. The SYNTAX Score: an angiographic tool grading the complexity of coronary artery disease. EuroIntervention (2005.0) 1 219-227. PMID: 19758907
--- title: 'Mid-life cyclists preserve muscle mass and composition: a 3D MRI study' authors: - Martin A. Belzunce - Johann Henckel - Anna Di Laura - Laura M. Horga - Alister James Hart journal: BMC Musculoskeletal Disorders year: 2023 pmcid: PMC10026522 doi: 10.1186/s12891-023-06283-3 license: CC BY 4.0 --- # Mid-life cyclists preserve muscle mass and composition: a 3D MRI study ## Abstract Physical activity and a healthy lifestyle are crucial factors for delaying and reducing the effects of sarcopenia. Cycling has gained popularity in the last decades among midlife men. While the cardiovascular benefits of cycling and other endurance exercises have been extensively proved, the potential benefits of lifelong aerobic exercise on muscle health have not been adequately studied. Our aim was to quantify the benefits of cycling in terms of muscle health in middle-aged men, using magnetic resonance imaging. We ran a cross-sectional study involving two groups of middle-aged male adults (mean age 49 years, range 30–65) that underwent Dixon MRI of the pelvis. The groups consisted of 28 physically inactive (PI) and 28 trained recreational cyclists. The latter had cycled more than 7000 km in the last year and have been training for 15 years on average, while the PI volunteers have not practiced sports for an average of 27 years. We processed the Dixon MRI scans by labelling and computing the fat fraction (FF), volume and lean volume of gluteus maximus (GMAX) and gluteus medius (GMED); and measuring the volume of subcutaneous adipose tissue (SAT). We found that the cyclists group had lower FF levels, a measure of intramuscular fat infiltration, compared to the PI group for GMAX (PI median FF $21.6\%$, cyclists median FF $14.8\%$, $p \leq 0.01$) and GMED (PI median FF $16.0\%$, cyclists median FF $11.4\%$, $p \leq 0.01$). Cyclists had also larger GMAX and GMED muscles than the PI group ($p \leq 0.01$), after normalizing it by body mass. Muscle mass and fat infiltration were strongly correlated with SAT volume. These results suggest that cycling could help preserve muscle mass and composition in middle-aged men. Although more research is needed to support these results, this study adds new evidence to support public health efforts to promote cycling. ### Supplementary Information The online version contains supplementary material available at 10.1186/s12891-023-06283-3. ## Introduction Sarcopenia is the progressive loss of muscle mass and function progressively as part of the natural ageing process [1–3]. Low levels of physical activity have also been associated with increased levels of muscle fat infiltration and progressive muscle weakness [4–6]. Therefore, physical activity and a healthy lifestyle are crucial factors in delaying and reducing the effects of sarcopenia. Moreover, larger muscle mass in early life can help to preserve muscle function at a later stage of life [7]. For this reason, being physically active before and during the onset of sarcopenia can have an important impact on later life stages to protect individuals against the effects of ageing on muscle health [3]. Cycling has gained popularity in the last decade, being one of the main physical activities in middle-aged adults [8], who are taking up cycling not only due to its physical health benefits and low impact, but also because of its effects on mental well-being, as shown by Glackin and Beale [9]. The benefits of cycling in terms of cardiovascular health and fitness have been widely studied [10–17]. For example, commuter cycling has been associated with improvements in cardiovascular fitness, reduction of all-cause mortality, cancer risk, overweight, and obesity among middle-aged individuals [10]. However, the impact of long-term cycling on muscle health has not been thoroughly explored [18, 19]. Consequently, it is essential to determine if cycling can help prevent sarcopenia and to estimate its impact on muscle mass and composition, which are markers associated with strength and mobility [20–23]. These two important muscle health markers can be quantified by measuring muscle volume and intramuscular fat (IMF) content from magnetic resonance imaging (MRI) [24, 25]. In this work, we aimed to study the benefits of cycling in terms of muscle health by comparing muscle health markers of two middle-aged men groups with different lifestyles: a group that has adopted cycling as their main recreational physical activity and a group of physically inactive subjects. We obtained Dixon magnetic resonance images of the pelvic region of each subject and computed the IMF content, muscle mass, and lean muscle mass of the gluteus maximus (very involved in cycling) and gluteus medius (less involved in cycling). ## Study design This was a cross-sectional study involving two matched groups of middle-aged adults who underwent MRI. The first group consisted of trained male cyclists that had cycled more than 7000 km in the preceding year. The second group consisted of physically inactive (PI) men (defined as men doing less than 1 h of physical exercise per week) ready to start the UK NHS (National Health Service) Couch to 5 K (Cto5K) programme, a running plan for absolute beginners. The inclusion criteria for this group were less than 1 h of physical exercise per week and registration to start the Cto5K programme. Common inclusion criteria for both groups were the absence of injuries and other health problems, no contraindication to MRI, and 30–65 years of age. We recruited a total of 56 subjects, 28 for the physically inactive group and 28 for the cyclists group, who met the inclusion criteria. The median cycling experience for the latter group was 12 years. Demographic data for each group are presented in Table 1. The volunteers underwent MRI and filled out a structured questionnaire regarding their physical activity levels and lifestyle on the scanning day. The following validated questionnaires were used: General Practice Physical Activity Questionnaire (GPPAQ), Warwick-Edinburgh Mental Wellbeing Scales (WEMWBS) [26] for mental health, and Hip disability and Osteoarthritis Outcome Score (HOOS) [27] for hip health as we assess two hip muscles. In addition, cyclists were asked about their cycling experience. All subjects provided written informed consent. The study was approved by the UCL Research Ethics Committee (REC) [Number 13,823 /001]. Table 1Demographics of the two study groups. Mean ± SD values are reportedPhysically InactiveCyclistsp-valueDemographicsSubjectsN = 28 $$n = 28$$Age [years]49.3 ± 10.648.0 ± 9.0Body Mass [kg]94.6 ± 17.777.2 ± 7.7Height [cm]179.1 ± 6.5180.8 ± 6.8BMI [kg/m2]29.4 ± 5.023.7 ± 2.5General Health QuestionnairesPhysical Activity (GPAQ)$I = 12$, MI = 4, MA = 9, $A = 5$ $A = 28$WEWBMS†48.3 ± 8.652.7 ± 7.7p = 0.03HOOS§ Pain95.1 ± 7.997.7 ± 6.2p = 0.06HOOS§ Function, Daily Living95.6 ± 7.898.8 ± 4.7p = 0.01HOOS§ Function, Sports92.4 ± 11.797.6 ± 5.6p = 0.10 General Practice Physical Activity Questionnaire; † Warwick-Edinburgh Mental Wellbeing Scales; § Hip disability and Osteoarthritis Outcome Score ## MRI acquisition All subjects underwent a standardized MRI protocol. The MR images were acquired on a 3T scanner (Siemens Magneton Vida, Erlangen, Germany) using a body coil. The scanning protocol consisted of axial PD TSE Dixon and axial T1-weighted images with a field of view (FOV) that covered from 2 cm below the lesser trochanter to the top of the L1 lumbar spine vertebra. The PD TSE Dixon sequence had the following parameters: slice thickness 2.6 mm, spacing between slices 2.6 mm, repetition time (TR) 5590 msec, echo time (TE) 51 msec, number of excitations 1, number of echoes 14, flip angle 150°. The voxel size was 0.55 × 0.55 × 2.6 mm3. ## Muscle health assessment with MRI We used gluteus maximus (GMAX) and gluteus medius (GMED) muscles to evaluate general muscle health, as they are essential to maintain an active lifestyle and are involved in a wide range of physical activities. Furthermore, GMAX is highly involved during the hip extension phase of the pedalling cycle [28, 29] but not GMED. Hence, we compared the health of a muscle that is directly trained by cycling with a muscle not very relevant in this sport. For each muscle, we computed three MRI-based muscle health metrics: intramuscular fat (IMF) content, muscle mass and lean muscle mass following a similar process to what we have done in previous studies [6, 30]. To measure the aforementioned metrics, we labelled the left and right GMAX and GMED muscles (see Fig. 1) using an in-house tool [25, 31] that runs on Simpleware ScanIP (Version 2021.3; Synopsys, Inc., Mountain View, USA). The tool has already been validated and used in other studies [6, 30]. The intramuscular fat (IMF) content was quantitatively measured by computing the mean fat fraction (FF) on each label from the FF Dixon MR images [32–34]. Muscle mass was estimated by summing up all voxels within a label and multiplying the results by the voxel size. Lean muscle mass was estimated as volume multiplied by (1-FF). Both volumetric measurements were normalized by the body mass of each subject. All the MRI scans were cropped at the tip of the lesser trochanter (LT) to avoid volume differences due to FOV mismatches. Additionally, size profiles were computed from the cross-sectional areas (CSA) of each axial slice that forms a muscle label. CSAs were also normalized by body mass. Profiles for FF and lean CSA were also computed. All the profiles (with a different number of slices for each subject) were resampled into 50 fixed slices or sampling points by applying a linear interpolation as described in a previous work [30]. We computed the median and the IQR for each slice of the resampled CSAs profiles and then estimated the relative percentage difference between the two groups. Fig. 1Axial and sagittal views of a physically inactive volunteer (top row) and a well-trained recreational cyclist (bottom row). The labels for GMAX, GMED and SAT are illustrated for both cases. The two subjects had GMAX fat fraction values of $21.8\%$ and $17.6\%$ ## Subcutaneous adipose tissue We measured the amount of subcutaneous adipose tissue (SAT) in the pelvis region by labelling the SAT on the Dixon MRI images and computing its volume (VSAT) and normalized volume (NVSAT) by body mass. The labelling was performed with an automated algorithm that classifies each voxel into three different classes as proposed by Bezrukov et al. [ 35], and then subtracts a convex hull of the non-fat mask from the fat label for each slice. ## Statistical analyses We computed nonparametric descriptive statistics for the FF and volume values for each muscle and group, since their distribution was not normally distributed (Kolmogorov-Smirnov test, $p \leq 0.01$). We compared the FF, volume and lean volume of the GMAX and GMED muscles, and the SAT volume, for the PI and cyclists groups using a Mann-Whitney U test for samples not normally distributed. Effect sizes were computed using the r-value, defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\raisebox{1ex}{$Z$}\!\left/ \!\raisebox{-1ex}{$\sqrt{N}$}\right.$$\end{document}, where Z is the standardized value for the U-value of the test [36]. Effect sizes were classified in low (r < 0.3), medium (0.3 < r < 0.5) and large (r > 0.5). We performed a linear regression analysis between cycling (as a categorical variable) and FF and NV. In addition, a multiple regression analyses were used to adjust for potential covariates. The variables tested were BMI, age, weight, NVSAT, hip health (using three HOOS scores) and levels of physical activity as defined by the GPAQ. We used a level of statistical significance (α) of 0.05 for all the tests. ## Results The PI group had a larger body mass (median 92.5 kg; $p \leq 0.01$) and a higher BMI (median 28.5 kg/m2; $p \leq 0.01$) than the cyclists group (median body mass 76.0 kg, median BMI 23.7 kg/m2). In the PI group, 16 volunteers were classified as inactive, 9 as moderately active and 5 as active using the GPAQ scores. All participants reported good hip function and health as assessed with the HOOS questionnaire, where we only found differences between the PI and the cyclists groups for the scores “HOOS Function, Daily Living” (Table 1). We found that the cyclists group had lower levels of fat infiltration for the two muscles under analysis compared to the PI group, and had larger GMAX and GMED muscles after normalizing the muscle volume by body mass. In Table 2, the median (IQR) values of fat fraction, volume, normalized volume and normalized lean volume are shown for each group, as the well as SAT volume and normalized volume. Table 2Median (IQR) values for GMAX and GMED muscles for the PI and Cyclists groupsPhysically InactiveCyclistsp-valueFat Fraction [%]GMAX21.6 (19.4–25.0)14.8 (13.3–16.2)$p \leq 0.01$GMED†16.0 (14.8–17.1)11.4 (10.5–12.8)$p \leq 0.01$Volume [cm3]GMAX804.7 (696.8-914.4)791.3 (707.6-869.1)$$p \leq 0.79$$GMED414.5 (373.1-484.5)390.2 (359.2-412.4)$$p \leq 0.09$$Normalized Volume [cm3/kg]GMAX8.6 (8.0-9.2)10.2 (9.5–11.0)$p \leq 0.01$GMED4.5 (4.3–4.7)5.0 (4.8–5.2)$p \leq 0.01$Lean Normalized Volume [cm3/kg]GMAX6.5 (5.8–7.5)8.6 (8.1–9.5)$p \leq 0.01$GMED3.7 (3.5-4.0)4.4 (4.2–4.6)$p \leq 0.01$SAT§ Volume [cm3]5071 (3454–6642)2158 (1858–2791)$p \leq 0.01$SAT Normalized Volume [cm3/kg]53.2 (42.7–62.7)29.3 (24.9–34.4)$p \leq 0.01$ Gluteus Maximus; † Gluteus Medius; § Subcutaneous Adipose Tissue ## Intramuscular fat In Fig. 2A, we show boxplots of FF for each group. The FF values were lower for cyclists for GMAX ($p \leq 0.01$, large effect size $r = 0.61$) and GMED ($p \leq 0.01$, large effect size $r = 0.69$). FF was correlated with BMI (R2 = 0.588, $p \leq 0.01$ for GMAX; R2 = 0.496, $p \leq 0.01$ for GMED), the categorical variable PI/Cyclists (R2 = 0.369, $p \leq 0.01$ for GMAX; R2 = 0.357, $p \leq 0.01$ for GMED) and NVSAT (R2 = 0.607, $p \leq 0.01$ for GMAX; R2 = 0.582, $p \leq 0.01$ for GMED). The multivariate model with highest prediction power included BMI and the PI/Cyclists variable as predictors. The NVSAT was highly correlated with both predictor variables and was excluded from the analysis to avoid collinearity. Age, levels of physical activity (as measured with the GPAQ) and hip health were not predictors of FF. The multivariate models for FF prediction were: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{l}F{F_{GMAX}}[\%] = - 1.3 + 0.7BMI + 2.5PI\,\\\,({R^2} = 0.629,\,{p_{BMI}} < 0.01,\,{p_{PI}} = 0.02)\end{array}$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{l}F{F_{GMED}}[\%] = 4.7 + 0.3BMI + 2.4PI\,\\\,({R^2} = 0.601,\,{p_{BMI}} < 0.01,\,{p_{PI}} = 0.01)\end{array}$$\end{document} Fig. 2Boxplots of FF (A), normalized volume (B) and normalized lean volume (C) values of GMAX and GMED muscles for each group. On each box, the central mark is the median and the edges of the box are the 25th and 75th percentiles. Outliers are plotted individually with circles ## Muscle mass In Fig. 2B C, we show boxplots of NV and LNV (B) for each group. NVs were larger for the cyclists than for the PI group for both GMAX ($p \leq 0.01$, large effect size r=-0.72) and GMED ($p \leq 0.01$, large effect size r=-0.55). The same was observed for the LNV (GMAX, large effect size r=-0.7391; GMED, large effect size $r = 0.6515$). Normalized muscle volume was correlated with the PI/Cyclists categorical variable (R2 = 0.439, $p \leq 0.01$ for GMAX; R2 = 0.294, $p \leq 0.01$ for GMED) and the NVSAT (R2 = 0.607, $p \leq 0.01$ for GMAX; R2 = 0.582, $p \leq 0.01$ for GMED), and weakly correlated with BMI (R2 = 0.233, $p \leq 0.01$ for GMAX; R2 = 0.226, $p \leq 0.01$ for GMED). The multivariate model with highest prediction power included the NVSAT and PI/Cyclists variables as predictors of the GMAX normalized volume: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{l}N{V_{GMAX}} = 1148 - 32NSAT\,[c{m^3}] - 1256PI\,\\\,({R^2} = 0.629,\,{p_{NSAT}} = 0.04,\,{p_{PI}} < 0.01)\end{array}$$\end{document} The correlation coefficients between all the tested variables and between the main output variables (FF and NV) and the predictors are illustrated in Figure S.1 and Figure S. 2 of the Supplementary Material. An exploratory data analysis of these variables is shown in Figure S. 3, where the differences between the two groups under study can be easily seen. ## Intramuscular fat and CSA profiles In Fig. 3, we show axial profiles of GMAX (A) and GMED (B) for the PI and Cyclists groups, where the median FF, CSA and lean CSA are shown for each axial slice. The error bars represent the IQR in each slice. The absolute percentage difference between the two groups is shown in a dashed line, where the higher FF of the PI group is uniform along both muscles. In terms of muscle size, the differences were more substantial in the inferior section near the lesser trochanter for GMAX and in the superior region for GMED. Fig. 3Axial profiles with median values and IQR error bars for GMAX fat fraction (A), normalized cross-sectional areas (B) and normalized lean cross-sectional areas (C) for the PI (blue) and cyclists (red) groups. In a purple dashed line and using the left y-axis, the relative percentage difference between the two groups is shown for each slice. The profiles go from the origin of GMAX (slice 1) to the level of the lesser trochanter (slice 50, the most inferior slice) ## Discussion This was a cross-sectional study in which two matched groups of middle-aged men were compared. Our objective was to quantify of cycling on muscle health in midlife men. We showed that the cyclists group had lower levels of intramuscular fat and greater muscle mass for both GMAX and GMED muscles than the physically inactive middle-aged men, with large effect sizes. These are relevant findings, as they suggest that cycling, an activity that is increasingly popular among middle-aged men [8], could be effective in slowing the degradation of muscle composition and the loss of muscle mass that is typically observed in the ageing population. Although our results are somehow expected and that more research is needed to understand how much cycling is needed to observe these outcomes, we provide important evidence supporting that lifelong aerobic exercise can slow the loss of muscle mass and function. These are novel results as the research in the prevention of sarcopenia has been mainly focused on resistance training as intervention instead of aerobic exercise [19]. A previous study showed that the thigh muscle mass of highly trained master cyclists was comparable to healthy young adults [37], which agrees with our findings regarding muscle mass preservation. In addition, our quantitative metrics from Dixon MRI offer reference values that can be used to study other groups in the future (i.e. commuter cyclists instead of the highly/moderately trained midlife cyclists of our study). We focused on only men due to the high cost of MRI scans, which allowed us to achieve a good sample size for two well-matched groups. Muscle health was assessed for GMAX, greatly involved in cycling, and GMED to determine if the benefits of cycling were only seen in the muscles targeted by this sport. We used Dixon FF as a quantitative measure of intramuscular fat, and muscle volume normalized by body mass as a measure of muscle mass. IMF levels were associated with a larger volume of SAT in the pelvis area, a higher BMI, and not being in the cyclists group. The high IMF and pelvic SAT observed in the PI group could be potentially associated with metabolic impairment [38]. Despite the lower levels of IMF content of the Cyclists groups compared to the PI groups, the former had higher levels of fat infiltration compared to previously published reference data of the gluteal muscles in healthy active individuals [30, 39]. This could be explained by the fact that in our study the participants were considerably older. GMAX and GMED muscles had different FF range in agreement with previously reported values [30, 39]. The larger muscle mass of the cyclists is an expected effect of training as muscle volume is associated with strength and power [39–42]. Using CSA profiles, we found that the size difference was located mainly in the inferior section of GMAX, which could be explained by the fact that GMAX is heavily involved during the hip extension phase of the pedaling cycle [28]. The inferior section of GMAX is mainly active during hip extension, while the superior section of GMAX is mainly active for abduction and external rotation that are not relevant for cycling [40]. Although GMED is not particularly targeted during cycling, we found meaningful differences between the two groups in the superior region of the muscle, which could be explained by GMED being active when using a more posterior pedal position [28]. The combination of lower IMF and larger GMAX and GMED mass in the cyclists groups, translated in even larger differences for lean muscle mass (normalized by body mass), a measure that combines muscle size (defined as the volume within the muscle fascia) and composition. The effect sizes on lean muscle mass of not being a cyclist were large for both GMAX and GMED, although slightly higher for the former. A limitation of this work is that we assessed the impact of cycling only in the gluteal muscles, which is an important muscle group associated with good mobility and a reduced risk of falls in the elder population [41, 42], but further research is needed to study if this muscle group is representative of the overall muscle health of middle-aged individuals. A second limitation of this study is that the two groups were recruited according to their current levels of physical activity. However, the volunteers of the cyclists group had been practicing this sports for a mean time of 12 years, and most of the PI subjects reported a lifelong physical inactivity. Therefore, this study compared two groups of midlife men with different long-standing lifestyles, although self-reported. A third limitation was that the recruiting criteria were based on self-reported physical inactivity, but half of the participants in the PI group were classified as moderately active or active using the GPAQ questionnaire. ## Conclusion We observed that well-trained midlife recreational cyclists had lower levels of fat infiltration and greater muscle mass for the two main gluteal muscles when compared to physically inactive individuals of the same age. This suggests that, in addition to other previously reported benefits, cycling could help preserve muscle health in middle-aged men. Although more research is needed to know at what level and how many years of cycling are required to see its positive impact on muscle health, this study adds new evidence to support public health efforts to promote cycling. ## Electronic supplementary material Below is the link to the electronic supplementary material. Supplementary Material 1 ## References 1. Vitale JA, Messina C, Albano D, Fascio E, Galbusera F, Corbetta S, Sconfienza LM, Banfi G. Appendicular muscle Mass, Thigh Intermuscular Fat Infiltration, and risk of fall in postmenopausal osteoporotic elder women. Gerontology (2021.0) 67 415-24. DOI: 10.1159/000513597 2. 2.Cruz-Jentoft AJ, Bahat G, Bauer J, Boirie Y, Bruyère O, Cederholm T, Cooper C, Landi F, Rolland Y, Sayer AA, Schneider SM, Sieber CC, Topinkova E, Vandewoude M, Visser M, Zamboni M. Writing Group for the European Working Group on Sarcopenia in Older People 2 (EWGSOP2), and the Extended Group for EWGSOP2 (2019) Sarcopenia: revised European consensus on definition and diagnosis.Age Ageing48:16–31. 3. McPhee JS, Cameron J, Maden-Wilkinson T, Piasecki M, Yap MH, Jones DA, Degens H. The contributions of Fiber atrophy, Fiber loss, in situ specific force, and Voluntary activation to weakness in Sarcopenia. Journals Gerontol Ser A (2018.0) 73 1287-94. DOI: 10.1093/gerona/gly040 4. Manini TM, Clark BC, Nalls MA, Goodpaster BH, Ploutz-Snyder LL, Harris TB. Reduced physical activity increases intermuscular adipose tissue in healthy young adults. Am J Clin Nutr (2007.0) 85 377-84. DOI: 10.1093/ajcn/85.2.377 5. Goodpaster BH, Chomentowski P, Ward BK, Rossi A, Glynn NW, Delmonico MJ, Kritchevsky SB, Pahor M, Newman AB. Effects of physical activity on strength and skeletal muscle fat infiltration in older adults: a randomized controlled trial. J Appl Physiol (2008.0) 105 1498-503. DOI: 10.1152/japplphysiol.90425.2008 6. 6.Belzunce MA, Henckel J, Di Laura A, Hart A. (2021) Intramuscular fat in gluteus maximus for different levels of physical activity. Sci Reports 2021 111 11:1–10. 7. Pearson SJ, Young A, Macaluso A, Devito G, Nimmo MA, Cobbold M, Harridge SDR. Muscle function in elite master weightlifters. Med Sci Sports Exerc (2002.0) 34 1199-206. DOI: 10.1097/00005768-200207000-00023 8. Aldred R, Woodcock J, Goodman A. Does more Cycling Mean more diversity in Cycling?. Transp Rev (2016.0) 36 28-44. DOI: 10.1080/01441647.2015.1014451 9. Glackin OF, Beale JT. The world is best experienced at 18 mph’. The psychological wellbeing effects of cycling in the countryside: an interpretative phenomenological analysis. Qual Res Sport Exerc Heal (2018.0) 10 32-46. DOI: 10.1080/2159676X.2017.1360381 10. Oja P, Titze S, Bauman A, de Geus B, Krenn P, Reger-Nash B, Kohlberger T. Health benefits of cycling: a systematic review. Scand J Med Sci Sports (2011.0) 21 496-509. DOI: 10.1111/j.1600-0838.2011.01299.x 11. De Geus B, Van Hoof E, Aerts I, Meeusen R. Cycling to work: influence on indexes of health in untrained men and women in Flanders. Coronary heart disease and quality of life. Scand J Med Sci Sports (2008.0) 18 498-510. DOI: 10.1111/j.1600-0838.2007.00729.x 12. de Hartog JJ, Boogaard H, Nijland H, Hoek G. Do the health benefits of cycling outweigh the risks?. Environ Health Perspect (2010.0) 118 1109-16. DOI: 10.1289/ehp.0901747 13. Ried-Larsen M, Rasmussen MG, Blond K, Overvad TF, Overvad K, Steindorf K, Katzke V, Andersen JLM, Petersen KEN, Aune D, Tsilidis KK, Heath AK, Papier K, Panico S, Masala G, Pala V, Weiderpass E, Freisling H, Bergmann MM, Verschuren WMM, Zamora-Ros R, Colorado-Yohar SM, Spijkerman AMW, Schulze MB, Ardanaz EMA, Andersen LB, Wareham N, Brage S, Grøntved A. Association of Cycling with all-cause and Cardiovascular Disease Mortality among persons with diabetes: the european prospective investigation into Cancer and Nutrition (EPIC) study. JAMA Intern Med (2021.0) 181 1196-205. DOI: 10.1001/jamainternmed.2021.3836 14. Gordon-Larsen P, Boone-Heinonen J, Sidney S, Sternfeld B, Jacobs DR, Lewis CE. Active commuting and cardiovascular disease risk: the CARDIA study. Arch Intern Med (2009.0) 169 1216-23. DOI: 10.1001/archinternmed.2009.163 15. Nordengen S, Andersen LB, Solbraa AK, Riiser A. Cycling and cardiovascular disease risk factors including body composition, blood lipids and cardiorespiratory fitness analysed as continuous variables: part 2—systematic review with meta-analysis. Br J Sports Med (2019.0) 53 879-85. DOI: 10.1136/bjsports-2018-099778 16. Lerssrimongkol C, Wisetborisut A, Angkurawaranon C, Jiraporncharoen W, Lam KBH. Active commuting and cardiovascular risk among health care workers. Occup Med (Chic Ill) (2016.0) 66 483-7. DOI: 10.1093/occmed/kqw029 17. 17.Hoevenaar-Blom M p., Wendel-Vos W, Spijkerman AM, Kromhout D, Verschuren W. (2011) Cycling and sports, but not walking, are associated with 10-year cardiovascular disease incidence: The MORGEN Study. Eur J Prev Cardiol 18:41–47. 18. Macaluso A, Young A, Gibb KS, Rowe DA, De Vito G. Cycling as a novel approach to resistance training increases muscle strength, power, and selected functional abilities in healthy older women. J Appl Physiol (2003.0) 95 2544-53. DOI: 10.1152/japplphysiol.00416.2003 19. Laurin JL, Reid JJ, Lawrence MM, Miller BF. Long-term aerobic exercise preserves muscle mass and function with age. Curr Opin Physiol (2019.0) 10 70-4. DOI: 10.1016/j.cophys.2019.04.019 20. Addison O, Marcus RL, Lastayo PC, Ryan AS. Intermuscular fat: a review of the consequences and causes. Int J Endocrinol doi (2014.0). DOI: 10.1155/2014/309570 21. Marcus RL, Addison O, Dibble LE, Foreman KB, Morrell G, Lastayo P. Intramuscular adipose tissue, sarcopenia, and mobility function in older individuals. J Aging Res doi (2012.0). DOI: 10.1155/2012/629637 22. Hilton TN, Tuttle LJ, Bohnert KL, Mueller MJ, Sinacore DR. Excessive adipose tissue infiltration in skeletal muscle in individuals with obesity, diabetes Mellitus, and Peripheral Neuropathy: Association with performance and function. Phys Ther (2008.0) 88 1336-44. DOI: 10.2522/ptj.20080079 23. Yoshida Y, Marcus RL, Lastayo PC. Intramuscular adipose tissue and central activation in older adults. Muscle Nerve (2012.0) 46 813-6. DOI: 10.1002/mus.23506 24. Ogawa M, Lester R, Akima H, Gorgey A. Quantification of intermuscular and intramuscular adipose tissue using magnetic resonance imaging after neurodegenerative disorders. Neural Regen Res (2017.0) 12 2100. DOI: 10.4103/1673-5374.221170 25. Belzunce MA, Henckel J, Fotiadou A, Di Laura A, Hart A. Automated measurement of fat infiltration in the hip abductors from Dixon magnetic resonance imaging. Magn Reson Imaging (2020.0) 72 61-70. DOI: 10.1016/j.mri.2020.06.019 26. Tennant R, Hiller L, Fishwick R, Platt S, Joseph S, Weich S, Parkinson J, Secker J, Stewart-Brown S. The Warwick-Edinburgh Mental Well-being scale (WEMWBS): development and UK validation. Health Qual Life Outcomes (2007.0) 5 63. DOI: 10.1186/1477-7525-5-63 27. Nilsdotter AK, Lohmander LS, Klässbo M, Roos EM. Hip disability and osteoarthritis outcome score (HOOS) – validity and responsiveness in total hip replacement. BMC Musculoskelet Disord (2003.0) 4 10. DOI: 10.1186/1471-2474-4-10 28. Timmer CAW. Cycling biomechanics: a literature review. J Orthop Sports Phys Ther (1991.0) 14 106-13. DOI: 10.2519/jospt.1991.14.3.106 29. So RCH, Ng JKF, Ng GYF. Muscle recruitment pattern in cycling: a review. Phys Ther Sport (2005.0) 6 89-96. DOI: 10.1016/j.ptsp.2005.02.004 30. Belzunce MA, Henckel J, Di Laura A, Hart AJ. Reference values for volume, fat content and shape of the hip abductor muscles in healthy individuals from Dixon MRI. NMR Biomed (2022.0) 35 e4636. DOI: 10.1002/nbm.4636 31. Belzunce MA, Henckel J, Fotiadou A, Di Laura A, Hart A. Automated multi-atlas segmentation of gluteus maximus from Dixon and T1-weighted magnetic resonance images. Magn Reson Mater Physics Biol Med (2020.0) 33 677-88. DOI: 10.1007/s10334-020-00839-3 32. Dixon WT. Simple proton spectroscopic imaging. Radiology (1984.0) 153 189-94. DOI: 10.1148/radiology.153.1.6089263 33. Wokke BH, Bos C, Reijnierse M, Van Rijswijk CS, Eggers H, Webb A, Verschuuren JJ, Kan HE. Comparison of dixon and T1-weighted MR methods to assess the degree of fat infiltration in duchenne muscular dystrophy patients. J Magn Reson Imaging (2013.0) 38 619-24. DOI: 10.1002/jmri.23998 34. Grimm A, Meyer H, Nickel MD, Nittka M, Raithel E, Chaudry O, Friedberger A, Uder M, Kemmler W, Engelke K, Quick HH. Repeatability of Dixon magnetic resonance imaging and magnetic resonance spectroscopy for quantitative muscle fat assessments in the thigh. J Cachexia Sarcopenia Muscle (2018.0) 9 1093-100. DOI: 10.1002/jcsm.12343 35. Bezrukov I, Mantlik F, Schmidt H, Schölkopf B, Pichler BJ. MR-based PET attenuation correction for PET/MR imaging. Semin Nucl Med (2013.0) 43 45-59. DOI: 10.1053/j.semnuclmed.2012.08.002 36. Tomczak M, Tomczak E. The need to report effect size estimates revisited. An overview of some recommended measures of effect size. TRENDS Sport Sci (2014.0) 1 19-25 37. Abe T, Nahar VK, Young KC, Patterson KM, Stover CD, Lajza DG, Tribby AC, Geddam DAR, Ford MA, Bass MA, Loftin M. Skeletal muscle mass, bone mineral density, and walking performance in masters cyclists. Rejuvenation Res (2014.0) 17 291-6. DOI: 10.1089/rej.2013.1538 38. Marcus RL, Addison O, Kidde JP, Dibble LE, Lastayo PC. Skeletal muscle fat infiltration: impact of age, inactivity, and exercise. J Nutr Heal Aging (2010.0) 14 362-6. DOI: 10.1007/s12603-010-0081-2 39. Marcon M, Berger N, Manoliu A, Fischer MA, Nanz D, Andreisek G, Ulbrich EJ. Normative values for volume and fat content of the hip abductor muscles and their dependence on side, age and gender in a healthy population. Skeletal Radiol (2016.0) 45 465-74. DOI: 10.1007/s00256-015-2325-z 40. 40.Selkowitz DM, Beneck GJ, Powers CM. Comparison of EMG Activity of the Superior and Inferior Portionsof the Gluteus Maximus Muscle during Common Therapeutic Exercises. 41. Tran V, De Martino E, Hides J, Cable G, Elliott JM, Hoggarth M, Zange J, Lindsay K, Debuse D, Winnard A, Beard D, Cook JA, Salomoni SE, Weber T, Scott J, Hodges PW, Caplan N. Gluteal muscle atrophy and increased intramuscular lipid concentration are not mitigated by Daily Artificial gravity following 60-Day head-down Tilt Bed Rest. Front Physiol (2021.0). DOI: 10.3389/FPHYS.2021.745811 42. Kiyoshige Y, Watanabe E. Fatty degeneration of gluteus minimus muscle as a predictor of falls. Arch Gerontol Geriatr (2015.0) 60 59-61. DOI: 10.1016/j.archger.2014.07.013
--- title: 'Development of a Survey Tool: Understanding the Patient Experience With Personalized 3D Models in Surgical Patient Education' journal: Cureus year: 2023 pmcid: PMC10026534 doi: 10.7759/cureus.35134 license: CC BY 3.0 --- # Development of a Survey Tool: Understanding the Patient Experience With Personalized 3D Models in Surgical Patient Education ## Abstract Background: Three-dimensional (3D) printing has been increasingly utilized in the healthcare sector for many applications including guiding surgical procedures, creating medical devices, and producing custom prosthetics. As personalized medicine becomes more accessible and desired, 3D printed models emerge as a potential tool in providing patient-specific education. These personalized 3D models are at the intersection of technological innovation and medical education. Our study group utilized a modified Delphi process to create a comprehensive survey tool assessing patient experience with personalized 3D models in preoperative education. Methods: A rigorous literature review was conducted of prior patient education survey tools in surgical cases across specialties involving personalized 3D printed models. Through categorization and mapping, a core study team reviewed individual questions, removed duplicates, and edited them into generalizable form. A modified Delphi process was then used to solicit feedback on question clarity and relevance from both 3D printing healthcare experts and patients to create a final survey. Results: 173 survey questions from the literature were evaluated by the core study team, yielding 31 unique questions for further review. After multiple rounds of feedback, a final survey containing 18 questions was developed. Conclusion: 3D printed models have the potential to be helpful tools in surgical patient education, and there exists a need to standardize the assessment of patient experience with these models. This survey provides a standardized, generalizable way to investigate the patient experience with personalized 3D-printed models. ## Introduction Establishing clear communication and understanding between physicians and patients to allow for shared decision making can be difficult in complex medical situations such as informed consent for surgery, discussing treatment plans, or clinical trials. One review found only around $50\%$ of clinical trial patients had an adequate understanding of the elements of informed consent, including the study’s aims, risks, and benefits [1]. More broadly, the Institute of Medicine reports that almost half of American adults - approximately 90 million - have difficulty understanding and utilizing health information [2]. Higher rates of health literacy can lead to better clinical outcomes. This has been demonstrated across a variety of medical conditions including diabetes mellitus, chronic obstructive pulmonary disease, and heart failure [3]. For example, increased health literacy in patients with diabetes is correlated to better glycemic control and poor health literacy in patients with asthma was the strongest predictor of improper use of inhalers [4,5]. Thus, there is a need for strategies to improve the information exchange between physicians and patients. Currently, physicians largely employ conversation and informational documents to explain health concepts and procedures to patients in the informed consent process, occasionally using visual tools such as sketches, medical imaging, or videos for communication [6-8]. Advances in technology and affordability, along with an increased acceptance by the medical community, have expanded the use of three-dimensional (3D) models in medicine [9]. Compared to the physician experience with custom 3D models for patient care, the patient experience with personalized 3D models is less reported. Published research has reported generally positive experiences utilizing 3D models for patient education and informed consent processes. For example, within the field of urology, patient-specific 3D models have been shown to increase patient understanding of anatomy, disease, and procedures [10]. Similar types of models have been used in preoperative patient consultations for neurosurgical, otolaryngologic, and orthopedic procedures with positive results [11-13]. Across these studies, researchers used a variety of measures to evaluate patients’ experiences with personalized 3D models ranging from subjective measures of satisfaction and understanding of the diagnosis to objective measures, such as anatomical knowledge and surgical approach questions. Because there exists no standardized set of questions to assess patient experiences with 3D models in pre-surgical information exchange, it is difficult to compare overall utility across specialties. Moreover, existing surveys are often highly medicalized, specialty-specific, and only readable at an advanced level. The majority do not involve lay individuals (i.e. patients) in the process to ensure their understanding and priorities are reflected. The purpose of this research was to collect the prior body of survey questions presented to patients exposed to 3D printing in surgical applications as a basis for developing a standardized, easily understood tool. By utilizing a modified Delphi method, the survey went through rounds of anonymous feedback from patients and clinicians. This survey can be utilized to measure the benefit of patient-specific 3D printed models in patient education around surgical planning that reflects both patient and expert feedback. ## Materials and methods Study team The core study team consisted of five individuals including a physician, physician-scientist, social scientist, engineer/medical student, and 3D printing lab manager. The 3D printing expert team consisted of nine individuals with previous experience working with personalized 3D printed models as part of clinical and/or surgical care at the study team’s institution. The patient team consisted of 18 non-medical individuals who had previous experience with a medical procedure. Participants were recruited through email as a convenience sample. This study was reviewed and approved by the Thomas Jefferson University Institutional Review Board (approval #20G.044). Literature search A literature search was conducted using the electronic databases Pubmed, Ovid, and SCOPUS. The following keywords were searched in the title and abstract: model OR planning OR training OR education OR teaching OR assessment OR skills OR simulation AND "3D print" OR "3D printing" OR "3D Printed" OR "three-dimensional print" OR "three-dimensional printing" OR "three-dimensional printed". Articles were screened by reviewing title and abstract to determine if they met inclusion criteria and warranted full-text review. Inclusion criteria required the article be 1) published in the English language, 2) a primary article, 3) use patient-specific 3D models for patient education, and 4) administer questions assessing patient experience with 3D printed models. Articles that initially passed inclusion criteria underwent data extraction. Data extraction Articles were examined in a systematic way and data extraction included study year, specialty, total questions administered, use of objective and subjective questions, and type of rating scale used. When available, full-text questions from surveys were extracted. Categorization To understand the scope of questions being asked in prior studies and create a shorter generalizable survey, the full-text questions were assigned to one of six categories - anatomy, communication, complication, diagnosis, experience, procedure, visualization - based on the focus of the question, as determined by the core study team through consensus (Table 1). Based on matching categories, similar questions were grouped and merged into one representative question. All questions, both unique and representative merged questions, were modified by two study team members to remove references to specific diagnoses or procedures resulting in a general, specialty-independent form. Table 1 \| Study \| Original question \| Generic question \| Generic question.1 \| Category \| Category.1 \| Unnamed: 6 \| Unnamed: 7 \| Generic Question \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Komai 2016 [14] \| Compared to 2D or 3D CT imaging, did the 3D model help understand the risks of operation-related complications such as bleeding and urinary leakage \| Compared to 2D or 3D CT imaging, did the 3D model help understand the risks of operation-related complications \| Complication \| Complication \| Merged \| Merged \| The model was useful for enhancing knowledge of possible complications \| The model was useful for enhancing knowledge of possible complications \| \| Yoon 2019 [15] \| I understand that potential complications includes death, and hospital stay can be lengthened. \| I understand the potential complications \| Complication \| Complication \| Merged \| Merged \| The model was useful for enhancing knowledge of possible complications \| The model was useful for enhancing knowledge of possible complications \| \| Alshomer 2019 [16] \| The information presented gave a clear vision about the risks and complications of the intervention \| The information presented gave a clear vision about the risks and complications of the intervention \| Complication \| Complication \| Merged \| Merged \| The model was useful for enhancing knowledge of possible complications \| The model was useful for enhancing knowledge of possible complications \| \| Klosterm-an 2018 [17] \| In the future I would recommend it [3D model] to family and friends \| In the future I would recommend it [3D model] to family and friends \| Experience \| Experience \| Merged \| Merged \| I would recommend a 3D model to others \| I would recommend a 3D model to others \| \| Bizotto 2016 [18] \| Would you suggest to other patients to request a 3D printed model of their fracture before the surgery? \| Would you suggest to other patients to request a 3D printed model of their diagnosis before the surgery? \| Experience \| Experience \| Merged \| Merged \| I would recommend a 3D model to others \| I would recommend a 3D model to others \| Mapping To further condense the survey to a manageable length, the core study team reviewed all questions remaining after the categorization and generalization process to determine dominant themes (i.e. anatomy/pathology, communication, procedure, and user experience) and grouped the questions accordingly (Figure 1). Questions that were considered too narrow in focus to apply broadly across specialties/procedures as determined by consensus were removed. Each endpoint of the diagram was converted into a short, generalizable question. The core study team reviewed these questions in the context of known features and clinical use of patient-specific 3D models and considered survey question supplementation if indicated. Figure 1: Diagram of Mapping Process Delphi process Round One A modified Delphi process was used to review the list of questions generated post mapping (Appendices). For the first round of feedback, the survey was sent to a team of healthcare-associated 3D printing experts. For each question, anonymous feedback on question clarity and relevance was obtained and experts noted if it should be retained as-is, reviewed, or removed. Clarity and relevance were graded using a 5-point Likert scale and retention on a 3-point scale (3- retained as-is, 2- reviewed, or 1- removed). Questions that received < 1.5 (i.e. scored more toward removal) on retention were automatically removed. Questions that received ≥ 3.0 (i.e. neutral) for relevance and clarity were kept for subsequent feedback rounds either as written, or reworded/combined with another similar question based on reviewer comments by at least two members of the core study team through consensus discussion. All other questions not meeting the above criteria underwent review by at least two members of the core study team and these questions were either reworded or combined with another similar question for inclusion in a subsequent feedback round or removed from further consideration based on reviewer comments (Table 2). Table 2 \| Original survey question \| The model improved my confidence (Strongly disagree = 1, Disagree = 2, Neutral = 3, Agree = 4, Strongly agree = 5) \| \| --- \| --- \| \| Feedback questions \| This question is clear (Strongly disagree = 1, Disagree = 2, Neutral = 3, Agree = 4, Strongly agree = 5) This question is relevant (Strongly disagree = 1, Disagree = 2, Neutral = 3, Agree = 4, Strongly agree = 5) This question should be… (Retained as-is, Reviewed, Removed) Comments \| \| Feedback results \| This question is clear – 66% of experts strongly agreed the question was clear (average = 4.75) This question is relevant – 25% of experts strongly agreed the question was relevant (average = 3.5) This question should be … - 50% of the experts believed the question should be retained as-is (retention score = 2.5) \| \| Feedback comments \| “improved my confidence in what?” “confidence in decision making” \| \| Outcome \| Original survey question edited to clarify keywords and address comments: The model improved my confidence in medical decisions. \| Round Two In the next round of feedback, a group of patients reviewed the list of questions generated from the expert review (Appendices). In a similar anonymous fashion, the patients were asked to review each question for relevancy, clarity, and retention. All questions underwent the same process as before based on the retention, relevance, and clarity scores. The resultant patient survey, both instructions written by the study team and individual questions, were evaluated for readability using the Flesch-Kincaid grade level [19]. Figure 2 displays the entire modified Delphi process for developing the survey. The instructions and questions were then modified to ensure readability at an 8th-grade reading level. Length of survey was calculated to ensure appropriate brevity of under five minutes [20]. Figure 2: Flow Diagram of Modified Delphi Process ## Results Literature search A total of 7141 articles were found on initial inquiry. After removing duplicates, 3705 records remained and underwent title and abstract face validity screening based on inclusion criteria. Sixty-three full text articles were selected for full content review; however upon complete review, only 30 of the 63 articles met inclusion criteria. Eight specialties were represented through these articles with orthopedic surgery being the most frequent specialty represented ($$n = 9$$), followed by urology ($$n = 6$$), neurosurgery ($$n = 5$$), general surgery ($$n = 3$$), cardiology ($$n = 3$$), plastic surgery ($$n = 2$$), otolaryngology ($$n = 1$$), and obstetrics and gynecology ($$n = 1$$). Categorization and mapping results A total of 173 full-text survey questions were extracted from articles. These questions were split into categories based on general focus including diagnosis ($$n = 47$$), procedure ($$n = 43$$), anatomy ($$n = 24$$), communication ($$n = 22$$), experience ($$n = 22$$), complications ($$n = 13$$), and visualization ($$n = 2$$). Following merging of similar/duplicate questions, 53 unique generic questions were generated (13 anatomy, 4 communication, 2 complication, 7 diagnosis, 15 experience, 11 procedure, and 1 visualization). Using the process of mapping questions to dominant themes, 31 key themes were identified and a unique question for each was generated from the questions in that group, yielding 31 survey questions (11 anatomy, 2 communication, 8 procedure, 10 experience). Modified Delphi process Ten healthcare-associated 3D printing experts submitted anonymous feedback on the questions. Of the 11 anatomy questions, two questions were removed due to low retention scores (mean < 1.5). One question received an average score of ≥ 3.0 for relevancy and clarity and was retained but reworded based on expert feedback. Based on review of the remaining question for relevance and clarity scores and discussing expert comments, only one other question was retained after rewording. Of the communication questions, one question received an average score of ≥ 3.0 for relevancy and clarity and was retained as written. The second question received a score of ≥ 3.0 for relevancy but was below the cutoff for clarity so it was retained but reworded. A total of eight questions relating to procedure were reviewed by the experts. Three received an average score of ≥ 3.0 for relevancy and clarity and were retained but reworded based on expert feedback. Of the remaining procedure questions, two pairs of questions were found to be very similar and were retained but combined based on expert feedback. The final question was also retained but reworded for clarity. Ten questions relating to user experience with the patient-specific 3D models were reviewed by the experts and all 10 received an average score of ≥ 3.0 for relevance and clarity and were retained. One was kept as-is, two were combined/reworded, and seven were reworded for clarity. At the end of this round of expert feedback, the survey contained 18 questions (2 anatomy, 2 communication, 6 procedure, 8 experience). In the second round of the Delphi process, 18 patients submitted feedback for the survey. There were no questions that scored below a 1.5 for retention and therefore none were removed. All questions scored above 3.0 for both relevance and clarity and were retained but underwent review by at least two members of the core study team for clarity using qualitative patient feedback as a guide. Seven questions were retained as-is and 11 questions were reworded for clarity/readability. The final survey therefore contained 18 questions (3 anatomy, 2 communication, 6 procedure, 6 experience, 1 comment) with a Flesch Kincaid Grade Level of 7.8, indicating a below 8th-grade comprehension level (Table 3). Table 3 \| Anatomy \| Unnamed: 1 \| \| --- \| --- \| \| \| The model was easy to understand. \| \| \| The model’s colors helped me identify the affected part of my body. \| \| \| The model’s scale (being life-sized) helped me understand the affected part of my body. \| \| Communication \| \| \| \| The model helped me communicate with my doctor. \| \| \| Compared to visits with no model, having the model at my appointment improved the experience. \| \| Procedure \| \| \| \| The model helped me understand my medical diagnosis. \| \| \| The model helped me understand the likely progression of my condition. \| \| \| The model helped me understand my treatment options. \| \| \| The model helped me understand the doctor’s plan. \| \| \| The model helped me understand possible risks during my procedure. \| \| \| The model helped me understand possible complications after my procedure. \| \| Experience \| \| \| \| The model improved my confidence in making medical decisions. \| \| \| I want my doctor to use 3D models in my future care. \| \| \| I am more likely to choose a doctor who uses 3D models compared to one who does not. \| \| \| I would recommend the use of similar 3D models to other patients. \| \| \| I would be willing to pay for a personalized 3D model. \| \| \| Overall, I liked having a 3D model used as part of care. \| \| \| Comment: Do you have any comments about having a 3D model used in your care? \| ## Discussion The strength of this research comes from the multi-faceted approach used to generate a specialty-agnostic, comprehensive set of survey questions that can be presented to patients. Importantly, we chose the landscape of prior survey-based research studies in orthopedic surgery, otolaryngology, general surgery, urology, and others as the foundation for our work. The questions on which our survey is based stem from the prior work of 30 other research groups across nine specialties, allowing us to take into consideration the values of each specialty. From the beginning, we prioritized creating a high-quality survey that takes into account aspects such as sentence and survey length in addition to accessible vocabulary. Given the known issue of survey fatigue and its contributing factors including survey length, topic, and question complexity, we aimed to develop a survey with less than 20 short, concise questions [21]. To condense the 173 questions that were initially extracted, questions were split into categories based on the focus of the question and similar questions were merged. To condense them further, a round of mapping overarching themes yielded 31 questions. By extracting the most basic contents of each question, a generic but comprehensive survey could be derived from complex specialty-specific questions. The Delphi method is a natural way to integrate the perspectives and values of both 3D printing experts and patients in an anonymous way into the survey [22]. In the category of anatomy, the healthcare-associated 3D printing experts found many questions to be too technical, related to the accuracy of the model and its relationship to other structures. These are common ways of thinking about anatomy in the healthcare field, but based on clinician experience with explaining medical procedures, felt to be difficult and less relevant. 3D printing experts’ feedback guided questions to be reworded when necessary and removed if no longer deemed relevant, thereby creating a more patient-centered survey. Patient feedback provided valuable insight for rephrasing questions and was overwhelmingly positive with most questions rated highly in both relevance and clarity. The study team found it critical to include patient perspectives while developing the questionnaire to empower patients to promote the aspects of their pre-surgical educational experience that they value most. The predominance of procedure and experience-related questions in the final survey may reflect the aspects of patient education that these personalized 3D models can address. Asking questions about patient experience may be the first step in learning about the value that patients place on having this educational tool as part of their preoperative appointment. Of course, the patient's experience with the model will be influenced by the educator using the 3D model. Additionally, the complexity of the anatomy or the time the physician spends explaining it could influence the model's ability to help patients understand. Therefore it is important to do this work to determine if patients find these personalized 3D models helpful and what they learn from the model. A well-known barrier to effective physician-patient communication is the use of language and reading levels that are too complex [23]. While the average American reads between a 7th- and 8th-grade level, educational tools like 3D models do not rely heavily on language alone [24,25]. Despite recommendations that patient health information should be delivered at a low readability level, many studies have shown the healthcare system has not done so yet [26]. The use of visual aids, like personalized 3D models, holds the potential to provide an alternative way to explain difficult concepts to patients. To our knowledge, this is the first study to describe a methodology for developing a comprehensive survey tool to assess patient perspectives on the use of 3D models in physician-patient communications for surgical applications. While previous researchers have captured patient feedback on 3D models, such feedback stems from independently generated surveys relevant to their specific specialties and surgical procedures, making it highly variable and difficult to compare across applications. Thus, a complete picture of the value of such models remains out of reach. The survey is now being utilized at our institution across specialties. In the future, the study team plans to validate the use of this survey and review the responses to determine the utility of 3D personalized models in addressing the communication barrier between clinicians and patients. The use of the Delphi process incorporated both 3D printing expert and patient feedback to better reflect the values of patient education researchers and the opinions and needs of patients. This survey can be used to gather data to improve the patient experience using personalized 3D models. Additionally, by utilizing this survey in a standardized fashion, we may determine that some model types or those used in specific specialties have more value than others. Our study did have a few limitations. Because patients involved in survey development were recruited through email as a convenience sample, selection bias may be an issue. In addition, due to the coronavirus pandemic they were unable to hold a model while giving their feedback. To fill this deficit, numerous pictures of 3D models that have been used in pre-surgical education were shown to the non-medical participants prior to collecting their feedback. Due to our decision to utilize both clinician and patient panels for feedback, we gained rounds of anonymous feedback from multiple shareholders. This meant the same group of people did not get the opportunity to evaluate the survey more than once. ## Conclusions Personalized 3D-printed models have the potential to overcome communication barriers by providing a visual tool to allow for better patient education. The purpose of this study was to use a strong base of previously published literature in combination with a modified Delphi process to work with multiple stakeholders to develop a broadly applicable survey to measure the potential of personalized 3D models in patient education. Taking into account years of literature across many specialties ensures the longevity and generalizability of this survey, providing valuable data on this tool in patient education. ## References 1. Falagas ME, Korbila IP, Giannopoulou KP, Kondilis BK, Peppas G. Informed consent: how much and what do patients understand?. Am J Surg (2009) 198 420-435. PMID: 19716887 2. Nielsen-Bohlman L. Health literacy: a prescription to end confusion. (2004) 3. Gazmararian JA, Williams MV, Peel J, Baker DW. Health literacy and knowledge of chronic disease. Patient Educ Couns (2003) 51 267-275. PMID: 14630383 4. Schillinger D, Piette J, Grumbach K. Closing the loop: physician communication with diabetic patients who have low health literacy. Arch Intern Med (2003) 163 83-90. PMID: 12523921 5. Williams MV, Baker DW, Honig EG, Lee TM, Nowlan A. Inadequate literacy is a barrier to asthma knowledge and self-care. Chest (1998) 114 1008-1015. PMID: 9792569 6. Faury S, Koleck M, Foucaud J, M'Bailara K, Quintard B. Patient education interventions for colorectal cancer patients with stoma: a systematic review. Patient Educ Couns (2017) 100 1807-1819. PMID: 28602564 7. McMullen S, Szabo S, Halbert RJ. Tools for communication: novel infrastructure to address patient-perceived gaps in oncology care. Clin J Oncol Nurs (2017) 21 0-48 8. Pentz RD, Lohani M, Hayban M. Videos improve patient understanding of misunderstood chemotherapy terminology. Cancer (2019) 125 4011-4018. PMID: 31418849 9. Aimar A, Palermo A, Innocenti B. The role of 3D printing in medical applications: a state of the art. J Healthc Eng (2019) 2019 5340616. PMID: 31019667 10. Wake N, Rosenkrantz AB, Huang R. Patient-specific 3D printed and augmented reality kidney and prostate cancer models: impact on patient education. 3D Print Med (2019) 5 4. PMID: 30783869 11. Sezer S, Piai V, Kessels RP, Ter Laan M. Information recall in pre-operative consultation for glioma surgery using actual size three-dimensional models. J Clin Med (2020) 9 12. Sander IM, Liepert TT, Doney EL, Leevy WM, Liepert DR. Patient education for endoscopic sinus surgery: preliminary experience using 3D-printed clinical imaging data. J Funct Biomater (2017) 8 13. Zhuang YD, Zhou MC, Liu SC, Wu JF, Wang R, Chen CM. Effectiveness of personalized 3D printed models for patient education in degenerative lumbar disease. Patient Educ Couns (2019) 102 1875-1881. PMID: 31113688 14. Komai Y, Sugimoto M, Gotohda N. Patient-specific 3-dimensional printed kidney designed for “4D” surgical navigation: a novel aid to facilitate minimally invasive off-clamp partial nephrectomy in complex tumor cases. Urology (2016) 91 226-233. PMID: 26919965 15. Yoon SH, Park S, Kang CH, Park IK, Goo JM, Kim YT. Personalized 3D-printed model for informed consent for stage I lung cancer: a randomized pilot trial. Semin Thorac Cardiovasc Surg (2019) 31 316-318. PMID: 30412772 16. Alshomer F, AlFaqeeh F, Alariefy M, Altweijri I, Alhumsi T. Low-cost desktop-based three-dimensional-printed patient-specific craniofacial models in surgical counseling, consent taking, and education of parent of craniosynostosis patients: a comparison with conventional visual explanation modalities. J Craniofac Surg (2019) 30 1652-1656. PMID: 30946228 17. Klosterman T, Romo T II. Three-dimensional printed facial models in rhinoplasty. Facial Plast Surg (2018) 34 201-204. PMID: 29466810 18. Bizzotto N, Tami I, Tami A. 3D Printed models of distal radius fractures. Injury (2016) 47 976-978. PMID: 26876530 19. Readable. (2021) 20. Question Science: How to calculate the length of a survey. (2022) 21. O'Reilly-Shah VN. Factors influencing healthcare provider respondent fatigue answering a globally administered in-app survey. PeerJ (2017) 5 0 22. de Villiers MR, de Villiers PJ, Kent AP. The Delphi technique in health sciences education research. Med Teach (2005) 27 639-643. PMID: 16332558 23. Shah P, Thornton I, Turrin D, Hipskind JE. Informed consent. StatPearls (2022) 24. Kutner M, Greenberg E, Jin Y, Paulsen C. The Health Literacy of America’s Adults: Results from the 2003 National Assessment of Adult Literacy. (2006) 25. Okrainec K, Lau D, Abrams HB. Impact of patient-centered discharge tools: a systematic review. J Hosp Med (2017) 12 110-117. PMID: 28182809 26. Choudhry AJ, Baghdadi YM, Wagie AE. Readability of discharge summaries: with what level of information are we dismissing our patients?. Am J Surg (2016) 211 631-636. PMID: 26794665
--- title: Quantitative evaluation of transdermal drug delivery patches on human skin with in vivo THz-TDS authors: - Xuefei Ding - Gonçalo Costa - A. I. Hernandez-Serrano - Rayko I. Stantchev - Gabit Nurumbetov - David M. Haddleton - Emma Pickwell-MacPherson journal: Biomedical Optics Express year: 2023 pmcid: PMC10026588 doi: 10.1364/BOE.473097 license: CC BY 4.0 --- # Quantitative evaluation of transdermal drug delivery patches on human skin with in vivo THz-TDS ## Abstract Transdermal drug delivery (TDD) has been widely used in medical treatments due to various advantages, including delivering drugs at a consistent rate. However, variations in skin hydration can have a significant effect on the permeability of chemicals. Therefore, it is essential to study the changes in skin hydration induced by TDD patches for better control of the delivery rate. In this work, in vivo terahertz (THz) spectroscopy is conducted to quantitatively monitor human skin after the application of patches with different backing materials and propylene glycol concentrations. Changes in skin hydration and skin response to occlusion induced by other patches are investigated and compared. Our work demonstrates the potential application of in vivo THz measurements in label-free, non-invasive evaluation of transdermal patches on human skin and further reveals the mechanism behind the effect. ## Introduction Transdermal drug delivery (TDD) has become a preferable alternative to conventional methods such as oral delivery and hypodermic injection due to being non-invasive, self-administered, inexpensive and reducing over-dose situations [1,2]. So far, TDD has been utilized in medical treatments for delivering drugs including nicotine, fentanyl and scopolamine [3]. When designing the patches for TDD, the choice of excipient composition and backing material are influential factors to consider. Permeation enhancers are often added to TDD patches to enhance the penetration of the drug through the skin and temporarily decrease the resistance of the skin barrier [4]. Propylene glycol (PG) is a commonly chosen permeation enhancer which also serves as humectant and co-solvent for poorly soluble chemicals, and it has been reported to have stronger ability in improving the transdermal flux for topical drug delivery than other widely used enhancers [5,6]. Backing materials can alter the drug delivery efficacy and the response of skin to the patch through different levels of occlusive features. When the drug contained in the patch is highly potent and toxic, it is essential to prevent molecular exchange with the environment by using a fully occlusive backing [7]. However, long-term occlusion of the skin can cause over-accumulation of water at the skin surface without evaporation and lead to skin irritation, therefore a partially occlusive backing is preferred under certain circumstances [8]. One of the most preferable advantages of TDD patches is that they allow a controlled amount of drugs to be delivered into the human skin at a relatively consistent rate. However, studies have noted that the hydration of the skin area to be treated is an important factor affecting the drug delivery rate, with the permeability of skin increasing significantly with the growth in hydration [9]. The stratum corneum (SC) is the outermost layer of skin, which has a thickness of approximately 10-15µm in the dry state and can swell up-to 40µm when hydrated; by increasing the hydration level of SC, the barrier function of skin can be reduced [3]. Therefore, it is essential to study the influence of TDD patches on skin hydration in order to have better control and understanding of the drug delivery rate. Additionally, analysing the recovery of skin after removing the patches can provide supporting information on the frequency of applying patches and switching the application area [10]. Terahertz (THz) radiation is a novel technique for biomedical examinations with the assets of non-ionizing photon energies and high sensitivity to water [11]. The potential use of ex vivo THz spectroscopy and imaging for studying the contrast between healthy and cancerous tissues [12–14] and to quantify glucose levels in diabetic blood plasma [15,16] has been investigated; subsequently, approaches to enhance THz characterization sensitivity of biological tissues and solutions have been proposed [17,18]. Recently, in vivo THz measurements have aroused increasing interest for its advantage in non-invasively monitoring the response of living skin to certain treatments. Studies have revealed the potential for using THz time-domain spectroscopy (THz-TDS) and THz imaging to assess the depth and severity of burn wounds in vivo [19,20]. Research has also been conducted on in vivo THz sensing of corneal tissue hydration [21,22] and THz imaging for in vivo evaluation of diabetic foot syndrome [23]. THz spectroscopic and imaging techniques have recently been utilized as non-destructive, label-free methods for evaluation of transdermal drug delivery. Kim et al. have demonstrated the ability of THz dynamic imaging in visualizing the spatial distribution and penetration of a topical drug on excised mouse skin [24]. Later on, Wang et al. applied ex vivo THz imaging to compare the efficacy between different TDD methods including the use of microneedles and nanoneedles [25]. Lee et al. reported a new way of quantifying drug delivery rate with THz sensing by measuring nicotine patches before and after application [26]. The above-mentioned studies are either conducted ex vivo on excised skin tissues or verified indirectly by measuring the changes on the patches. To acquire a more comparable result to the actual dynamic process in skin during TDD, in vivo THz measurements on human skin need to be studied. Lindley-Hatcher et al. have conducted in vivo THz point scan on volar forearm to investigate the changes induced in the skin by TDD patches with different backing materials and PG concentrations [10]. The study focused on how different types of patches can influence the drug delivery rate through the changes in skin hydration and skin’s response to occlusion. Their work provided a proof of concept with a small scale of participants and measured with THz point scan observing only a single location on the skin. In this study, we developed the work further by increasing the number of subjects measured and within a larger scale we were able to categorize them into different skin groups according to their original skin condition. We have investigated the effect of different types of patches on the skin hydration level and skin occlusion process [27]; trends were studied in skin type groups as well as in general for all subjects, offering a deeper understanding of the skin’s response to the patches. In addition to the in vivo THz point scan, we also conducted in vivo THz imaging and 3D camera imaging on skin to visualize the results. THz imaging provided spatial information of the skin hydration, while the 3D camera imaging showed changes in the roughness of skin before and after patch application. Both techniques provided supplementary information to THz-TDS revealing the mechanism of the changes induced in the skin by TDD patches. ## THz system and other experimental setup In this study, in vivo THz skin measurements were acquired with a Menlo TERA K15 THz Time-Domain system. The system was set up in reflection geometry with a THz emitter and detector vertically assembled on optical rails at an incident angle of 30° to the quartz imaging window (Fig. 1(a)). To measure the THz response of living skin, each participant was asked to rest their volar forearm on top of the quartz imaging window as shown in Fig. 1(a) for one minute. During that time the THz system recorded approximately 280 reflected THz pulses from a single point on the skin. For the in vivo THz imaging, the K15 spectrometer was connected to an identical reflection setup on a motorized x-y imaging stage (Fig. 1(b)) for raster scanning the skin. Each image has a size of 17mm × 17 mm covering the full area where the patches were applied, taking approximately 3 min to image the area with a resolution of 1 mm in both x-y directions. A reference and baseline measurement are also taken at each point such that the processed waveform (Eq. [ 1]) is calculated individually for each point in the image to account for any inhomogeneity in the imaging window. A high-precision surfaceCONTROL 3D 3500 by Micro-Epsilon camera was used to take the 3D images of the skin (Fig. 1(c)), providing information about the skin surface roughness. The 3D camera can achieve a 2.2 million cloud points per second with a vertical resolution down to 1.0 µm and an x-y resolution of 40 µm. Fig. 1.: (a) Menlo TERA K15 THz spectrometer set up in reflection geometry at an angle of 30°; (b) K15 imaging system; (c) surfaceCONTROL 3D 3500 sensor from Micro Epsilon. ## Materials and protocol for skin measurements Ethical approval was obtained for this study from the Biomedical Scientific Research Ethics Committee, BSREC, (REGO-2018-2273 AM03). The 19 subjects participating in this study were in the age group of 23 to 33 and possessed healthy skin in the regions of interest. Prior to the start of the study, all subjects were informed of the experiments and gave their signed consent. To provide further insight into the significance of the observations in the previous work [10] on the effects of the patches on the skin, woven and film based patches were applied on the volar forearms of the 19 subjects. The purpose of this study was to explore the changes that transdermal drug delivery patches generate in the skin via in vivo THz spectroscopy. As previously stated, in this study patches with two types of backing materials were used: the fully occlusive poly(ethylene terephthalate (PET) film-backed patches and the partially occlusive woven-backed patches; and for each type of patch, an excipient with propylene glycol added at a concentration of $0\%$, $3\%$ and $6\%$ was tested. These are commonly used for the enhancement of TDD rate in such patches, and $10\%$ transcutol was added in all patches as co-solvent. No drugs were present in the patches to allow singling out the changes induced on the skin to be a consequence of the patch material and of the excipient concentration. Figure 2(a) shows three woven-backed patches and three film-backed patches applied to the right and left volar forearm respectively; a control area was also marked on each arm to take into account the natural variation of skin occurring between each set of measurements. All of the 8 areas were marked with surgical skin marking pen in advance to keep track of the areas treated with patches. The measurements were taken in identical time frames: before the application of the patches, then 0 minutes, 30 minutes and 4 hours after removing the patches. Patches were applied and removed in a fixed order to allow each patch being applied on the skin for 24 hours. The ‘0 min’ measurements were immediately conducted after removing each patch, and then the skin areas were measured at 30 min and 4 hours after the time of removal in the same order as patches were removed. The control areas were measured each time along with the treated areas. Fig. 2.: (a) Photo of the areas of interest indicating the locations of patches on the volar forearms. (b) A schematic diagram of the reflected signals of bulk sample in reflection geometry. To further assure repeatability in the measurements, an established protocol for THz in vivo skin measurements was followed, and the pressure of skin contact and environment condition factors were taken into consideration [28]. Previous studies have found that the pressure between the skin and the imaging window affects the THz response of the sample [29]. To mitigate against this effect, pressure sensors were applied on each side of the quartz window to assure the pressure applied on the window was kept constantly in a range of 1.5-2.5 N/cm2. All participants were subjected to a 20-minute period of acclimatization in the lab before starting measurements; and other external factors that might affect the skin properties of each individual were accounted for through data processing. ## Data processing In vivo THz measurements of volar forearm are conducted in reflection geometry with an imaging window in contact with the skin. In this case the detected sample signal Esam contains two pulses (Fig. 2(b)): the first pulse is the reflection from the air-window interface which is defined as the baseline Ebsl; and the second pulse is the actual reflection from the window-sample interface. The reflection from the bare window is also measured as the reference signal Eref. Esam, Ebsl and Eref are all functions of time, t, which is the optical delay in picoseconds (ps). Due to the ringing effect, the baseline needs to be subtracted from the sample and reference signals (we measured the baseline signal from a very thick window where we only see the first reflection). The reference and the baseline were measured on each day during the study to eliminate the subtle variation in system signals. The processed signal is then calculated according to Eq. [ 1], where a double Gaussian filter is applied to remove noise [30,31]. [ 1] processedsignal=iFFT(FFT(filter)×FFT(Esam(t)−Ebsl(t))FFT(Eref(t)−Ebsl(t))) Figure 3(a) presents the processed THz signals throughout a measurement of untreated skin occluded for 60s in which the signals were shifted horizontally for clarity. The peak-to-peak (P2P) variable is defined as the change between the highest and lowest amplitude of the processed signal. A decrease in P2P is observed for the 60s process which was proposed in a previous study as the occlusion effect: when skin is in contact with the quartz imaging window, the surface of skin is occluded leading to accumulation of water, which will then reduce the reflection at quartz-skin interface [32]. Figure 3(b) is the occlusion curve showing how the P2P changes with occlusion time and the data points are fitted with a biexponential function. Here we take a sampled point on the biexponential fit at 52s into occlusion when the curve tends to be stable, and the ΔP2P variable is defined as the difference between the first point into occlusion and the sampled point. Fig. 3.: (a) An example of the processed THz signals measured from untreated skin for 60s of occlusion time. The pulses are plotted every half second for the first two seconds and every five second for the rest; they have been shifted horizontally for clarity. Peak-to-peak (P2P) index is defined. (b) The correlation between the P2P of processed signal and the occlusion time. A biexponential function is fitted to the measured data. A sampled point is taken at 52s into occlusion and the definition of ΔP2P is illustrated. In this study, we measure the skin at different time points throughout a rather long scale of time (28 hours), so it is necessary to consider the natural variation of skin while investigating the effects induced by the patch application. Therefore, the normalized relative change (NRC) is calculated to isolate the change of the THz response of skin that is caused by the patches [28]: [2] NRC(%)=(XTt−XT0)−(XCt−XC0)XT0+(XCt−XC0)×100 where XTt and XCt are the chosen variables of THz responses measured from the treated (T) area and control area (C) of skin at a specific time (t) point after removal of patches, and XT0 and XC0 are the variables measured from those two areas before patch application. Trans-epidermal water loss is relatively constant in the longitudinal direction of the volar forearm [33]. Furthermore since we are calculating the NRC of variables, we are able to make meaningful comparisons between the patches in the different positions within the volar forearm. ## Variation on the original skin condition Among the 19 subjects measured, variation can be seen in their original skin hydration profile without patch application. As illustrated in Fig. 4, the largest P2P variation between subjects is approximately 3-5 times the ΔP2P of one subject throughout the occlusion process for the control area on the dominant arm. As the P2P value is associated with the hydration level of skin in the way that lower P2P represents higher water content in skin, it is therefore important to categorize the subjects into different groups according to their original skin condition given by the control areas. Here we categorize them into three groups: subject 1-6 as the ‘Dry’ group; subject 7-13 as the ‘Average’ group; and subject 14-19 as the ‘Hydrated’ group. Figure 4(b) shows an example of the processed THz signals of all subjects measured at 2s into occlusion, providing more details on the P2P differences between each skin group. Figure 4 is a rough indication of the grouping, whereas the actual categorization is based on a comprehensive analysis of the hydration state of the control area of different subjects at several time points (before, 0 min, 30 min and 4 hours) and also their individual skin response to the patch treatments. The categorization is customized for the group of subjects in this study with a purpose of isolating the unusual results and studying the correlation between the unusual skin responses and the original skin condition. Further research needs to be conducted to give a robust method for the categorization of skin conditions in a general sense. In the following sections, we will be looking at the statistical trends for each group as well as for all subjects in general. Fig. 4.: Occlusion curves of the control area on the dominant arm measured (a) before patch application and (b) 0 min after taking off the patches for all subjects. ## Effect of patch application on the skin hydration level NRC of the sampled P2P is calculated for all subjects at 0 min, 30 min and 4 hours after the removal of the different patches as according to Eq. [ 2]. A negative NRC value of P2P represents an increase in skin hydration level compared to the original state before patch application and a positive value implies a decrease; the hydration level is inversely proportional to the P2P NRC. The average P2P NRC of all subjects is shown in Fig. 5(a) with error bars indicating the standard error in the mean. In general, an increase in skin hydration associated with the decrease in P2P is observed for all patch application areas at 0 min after the patches were removed after being applied for 24 hours. The changes show a declining trend with time indicating the recovery process of the skin, however the hydration effect persists even 4 hours after removal. Changes induced by the film patches are slightly larger than the woven patches and last longer in time. Different excipient compositions are observed to have some influences on both patches, and a higher PG concentration results in more hydrated skin. Fig. 5.: The average normalized relative change (NRC) of the sampled P2P in (a) all 19 subjects, (b) the ‘Average’ group, (c) the ‘Dry’ group and (d) the ‘Hydrated’ group measured at 0 min, 30 min and 4 hours after the removal of different patches (woven patches and film patches with 0%, 3% and 6% propylene glycol). Error bars are acquired by the standard error on the mean. Compared to the ‘All subjects’ group, results given by the ‘Average’ group in Fig. 5(b) show similar trend; more notable differences are observed for skin hydration changes induced by film and woven patches, with areas covered by the film patches being more hydrated than woven patches after 0 min, 30 min and 4 hours of removal respectively. For this skin group, we can still see a positive correlation between PG concentration and skin hydration level for both patches. For subjects with ‘Dry’ skin group, both film and woven patches have smaller impact on the THz response of skin compared to the ‘Average’ group; on the other hand, the skin recovery rate increases in general and some areas appear to be even dryer than the original state after 30 min following the removal of the patches, as shown in Fig. 5(c). When the original skin state of the subjects is more hydrated than average group, we can observe from Fig. 5(d) that film patches induce smaller changes on skin than the woven patches; and the recovery rate of the skin strongly decreases for all areas, leading to the skin staying at a highly hydrated state even 4 hours after removing the patches. A possible explanation is that when the original skin is already hydrated to a certain extent, occlusion effect caused by the fully occlusive film patches is no longer dominant in altering the skin hydration level, while the partially occlusive woven patches allow moisture exchange between skin and air and have more effect on further hydrating the skin. The impact induced by different PG concentrations for the ‘Dry’ group and ‘Hydrated’ group shows similar trend as observed in all subjects. ## Effect of patch application on the skin occlusion process The study shown in the last section provides an idea on how patch application interacts with the general hydration level in skin. Due to the increased hydration levels of the skin after the application of the patches, it is natural to ask if the skin’s response to occlusion has also been affected. As mentioned in Fig. 3, the occlusion process happens on untreated skin once it is in contact with the imaging window, blocking the exchange of water with the environment during the one-minute measurement. This process is also observed as a decrease in the P2P of the THz waveform which can be modelled by a biexponential curve. However, if the patch application disrupts the normal water distribution in skin, changes in the occlusion curve would be expected and the defined ΔP2P can be used as a variable for quantification. Figure 6 presents the NRC of ΔP2P in a box plot, where the red/blue lines inside the boxes indicate the median response and the upper and lower limits show the upper and lower quartiles. A negative NRC in ΔP2P suggests a decline in the variation of P2P during the occlusion process and that the occlusion curve flattens, which can imply that the skin is in a comparable condition of already being occluded and that further occlusion has less impact on it. While a positive NRC in ΔP2P is correlated with an increment in the change of P2P and the occlusion curve steepens. Fig. 6.: Box plots showing the NRC of ΔP2P in (a) all 19 subjects, (b) the ‘Average’ group, (c) the ‘Dry’ group and (d) the ‘Hydrated’ group measured at 0 min and 4 hours after the removal of different patches. The red/blue lines inside the boxes represent the median response; the upper and lower edges of the boxes are the upper and lower quartiles of the responses. As shown in Fig. 6(a-b), the ‘Average’ group presents similar results as taking all subjects into account, where looking at the film patches at 0 min after removal we can see ΔP2P NRC far below the zero-line indicating flattened occlusion curves. This observation further demonstrates that the impact on skin hydration by film patches is through the fully-occlusive feature. Woven patches, on the other hand, seem to have different effect on skin occlusion for different PG concentrations. After 4 hours of removing the patches, ΔP2P NRC for all excipient concentrations goes back to the state before treatment, revealing a recovery process for the water distribution in skin. For the ‘Dry’ group in Fig. 6(c), the impact of the film patches on skin occlusion decreases compared to the ‘Average’ group and barely any impact is seen for woven patches. In Fig. 6(d), we can still observe clear occlusive effect for all the film patches; this indicates that for skin that is already much hydrated, film patches still have an occlusive effect on it although they do not increase the skin hydration level as much (see Fig. 5(d)). ## Statistical significance In addition to the data analysis in the previous sections, the statistical significance of the changes in skin hydration level and skin occlusion process induced by patch application is tested with the one-way analysis of variance (ANOVA) test and the post hoc Dunnett’s test among all the 19 subjects. The NRC of P2P measured at 0 min and 4 hours after removal of 6 types of patches is tested along with the control group which has an NRC value of 0 to check if there is significant difference after treatment. According to the one-way ANOVA test there is a statistically significant difference between groups for both measurement times, and Fig. 7(a) illustrates the results from the Dunnett’s test where the shaded bars represent the $95\%$ confidence intervals. It is observed from Fig. 7(a) that all patches result in a significant change in the P2P of skin at 0 min, while the changes induced by $3\%$ and $6\%$ film patches persist significant after 4 hours. The NRC of ΔP2P is tested in a similar pattern and results are presented in Fig. 7(b). The one-way ANOVA test reveals statistically significant difference between groups for those measured at 0 min; from the Dunnett’s test, the $3\%$ and $6\%$ film patches and the $3\%$ woven patch have significant impact on the ΔP2P of skin immediately after removing the patches, and all changes lose their significance after 4 hours. Fig. 7.: Results of performing the one-way analysis of variance test on (a) the NRC of sampled P2P and (b) the NRC of ΔP2P for all subjects measured at 0 min and 4 hours after the removal of patches. The cross/plus markers indicate the estimated mean of distribution; the shaded bars show the 95% confidence intervals calculated with the Dunnett’s test. The control group is represented by the red dashed line at the value of 0. ## Visualization of changes on skin hydration and surface profile THz imaging is performed for better visualization of the hydration change in the entire region of skin covered by patches. Figure 8 shows the imaging results of the skin areas of one subject from the ‘Average’ group at different times after removing the $6\%$ film patch (Fig. 8(a-c)) and the $6\%$ woven patch (Fig. 8(d-f)). In general, blue areas indicate negative NRC in P2P, associated with an increase in the skin hydration level. We can conclude from the figures that the film patches compared with the woven patches have greater impact on the skin with a larger hydrated area and a deeper hydration level. The skin area applied with film patches stay hydrated for at least 4 hours with the hydration level only decreasing slightly. This imaging result serves as a complement to the quantitative results in Fig. 5, showing the impact of patches on the entire treated area of skin with more spatial information. Fig. 8.: THz imaging results of the skin areas of one subject measured at 0 h, 2 h and 4 hours after the removal of (a-c) 6% film patch and (d-f) 6% woven patch. The 3D camera is able to take images of an object obtaining the height information in the z-axis along an x-y surface, which in this study can be used to measure the roughness of the skin surface as shown in Fig. 9. Our previous discussions indicate that film patches hydrate the skin through occlusion effect, and according to Fig. 8 the hydration is more spatially uniform and persistent than that induced by woven patches. Therefore we used the 3D camera to measure the skin area of one subject from the ‘Average’ group before and after treatment with the $6\%$ film patch and the $6\%$ woven patch. In Fig. 9(a-c), we can directly see the change in skin roughness before and after the application of the film patch; the skin surface is clearly smoother immediately after the patch removal, and its roughness recovers partially after 4 hours yet remains less bumpy than before. To quantify the change in the surface roughness, height information along the white dashed line in each 3D image is acquired and the standard deviation is calculated, where smaller standard deviation is correlated with a smoother surface. From Fig. 9(d) we can see that before treatment σ=31μm, while 0 min after removing the film patch σ decreases to 17μm, and then after 4 hours it goes back to 24μm. In contrary, the skin treated with the woven patch does not have such a visible difference in the surface roughness before and after treatment, as shown in Fig. 9(e-g). The quantitative evaluation of the roughness in Fig. 9(h) indicates that the standard deviation σ changes from 29μm to 20μm immediately after removing the woven patch and then increases to 23μm after 4 hours. Fig. 9.: 3D camera images of the skin area of one subject taken before patch application, 0 min and 4 hours after removing (a-c) the 6% film patch and (e-g) the 6% woven patch. Roughness of the skin surface at different times with the application of (d) the 6% film patch and (h) the 6% woven patch. Roughness is acquired along the black dashed lines on 3D images. Standard deviation σ is calculated. 3D images can display the roughness changes in skin surface before and after patch application with a straightforward visualization. Results show that occlusion induced by film patches changes the skin surface profile by smoothing the surface and reduces the depth of furrows, while accumulating water in the stratum corneum [34]. ## Summary In this work, we have used in vivo THz spectroscopy to quantify the changes in human skin induced by transdermal drug delivery patches with different backing materials and propylene glycol concentrations. We found that the influence of patches on skin hydration and skin’s response to occlusion depends on the initial skin type of the subject as well as the patch backing and excipients used. Patches with a film backing increased the hydration in all but the most hydrated skin subjects, and this effect dominated over the concentrations of excipients used. Patches with a woven backing did not increase the hydration as much and the effect of the concentration changes in excipients was noticeable across most subjects. We have shown how THz sensing can be used to quantitively evaluate the direct impact on skin along with the resultant recovery process caused by different patches. THz imaging and 3D camera imaging complemented the results. Our work provides further insights into the impact of different types of TDD patches on skin and the mechanism behind it. The results demonstrate the potential application of in vivo THz sensing in the design and evaluation process of patches for transdermal drug delivery. ## Funding Royal Society$\frac{10.13039}{501100000288}$ (Wolfson Merit Award); China Scholarship Council$\frac{10.13039}{501100004543}$ (Sophie Ding); Engineering and Physical Sciences Research Council$\frac{10.13039}{501100000266}$ (EP/S$\frac{021442}{1}$, EP/V$\frac{047914}{1}$). ## Disclosures DMH is the Chief Scientific Officer of Medherant Ltd. And GM is an employee of Medherant Ltd. The University of Warwick is a minority shareholder of Medherant Ltd. ## Data availability Data underlying the results presented in this paper are available in Ref. [ 35,36]. ## References 1. Prausnitz M. R., Langer R.. Transdermal drug delivery. Nat. Biotechnol. (2008) 26 1261-1268. DOI: 10.1038/nbt.1504 2. Tombs E. L., Nikolaou V., Nurumbetov G., Haddleton D. M.. Transdermal delivery of ibuprofen utilizing a novel solvent-free pressure-sensitive adhesive (PSA): TEPI® technology. J. Pharm. Innov. (2018) 13 48-57. DOI: 10.1007/s12247-017-9305-x 3. Dhiman S., Singh T. G., Rehni A. K.. Transdermal patches: a recent approach to new drug delivery system. Int J Pharm Pharm Sci (2011) 3 26-34 4. Carrer V., Alonso C., Pont M., Zanuy M., Córdoba M., Espinosa S., Barba C., Oliver M. A., Martí M., Coderch L.. Effect of propylene glycol on the skin penetration of drugs. Arch. Dermatol. Res. (2020) 312 337-352. DOI: 10.1007/s00403-019-02017-5 5. Nair A. B., Gupta S., Al-Dhubiab B. E., Jacob S., Shinu P., Shah J., Aly Morsy M., SreeHarsha N., Attimarad M., Venugopala K. N., Akrawi S. H.. Effective therapeutic delivery and bioavailability enhancement of pioglitazone using drug in adhesive transdermal patch. Pharmaceutics (2019) 11 359. DOI: 10.3390/pharmaceutics11070359 6. Babu R. J., Pandit J. K.. Effect of penetration enhancers on the transdermal delivery of bupranolol through rat skin. Drug Delivery (2005) 12 165-169. DOI: 10.1080/10717540590931936 7. Lane M. E.. The transdermal delivery of fentanyl. Eur. J. Pharm. Biopharm. (2013) 84 449-455. DOI: 10.1016/j.ejpb.2013.01.018 8. Hurkmans J. F. G. M., Boddé H. E., Van Driel L. M. J., Van Doorne H., Junginger H. E.. Skin irritation caused by transdermal drug delivery systems during long-term (5 days) application. Br. J. Dermatol. (1985) 112 461-467. DOI: 10.1111/j.1365-2133.1985.tb02321.x 9. Sharma N., Agarwal G., Rana A., Bhat Z. A., Kumar D. J.. A review: transdermal drug delivery system: a tool for novel drug delivery system. Int. J. Drug Dev. & Res. (2011) 3 70-84 10. Lindley-Hatcher H., Wang J., Hernandez-Serrano A. I., Hardwicke J., Nurumbetov G., Haddleton D. M., Pickwell-MacPherson E.. Monitoring the effect of transdermal drug delivery patches on the skin using terahertz sensing. Pharmaceutics (2021) 13 2052. DOI: 10.3390/pharmaceutics13122052 11. Sun Q., He Y., Liu K., Fan S., Parrott E. P. J., Pickwell-MacPherson E.. Recent advances in terahertz technology for biomedical applications. Quant. Imaging Med. Surg. (2017) 7 345-355. DOI: 10.21037/qims.2017.06.02 12. Sim Y. C., Park J. Y., Ahn K.-M., Park C., Son J.-H.. Terahertz imaging of excised oral cancer at frozen temperature. Biomed. Opt. Express (2013) 4 1413-1421. DOI: 10.1364/BOE.4.001413 13. Bowman T. C., El-Shenawee M., Campbell L. K.. Terahertz imaging of excised breast tumor tissue on paraffin sections. IEEE Trans. Antennas Propag. (2015) 63 2088-2097. DOI: 10.1109/TAP.2015.2406893 14. Wallace V. P., Fitzgerald A. J., Shankar S., Flanagan N., Pye R., Cluff J., Arnone D. D.. Terahertz pulsed imaging of basal cell carcinoma ex vivo and in vivo. Br. J. Dermatol. (2004) 151 424-432. DOI: 10.1111/j.1365-2133.2004.06129.x 15. Chen H., Chen X., Ma S., Wu X., Yang W., Zhang W., Li X.. Quantify glucose level in freshly diabetic’s blood by terahertz time-domain spectroscopy. J. Infrared, Millimeter, Terahertz Waves (2018) 39 399-408. DOI: 10.1007/s10762-017-0462-2 16. Anastasia A. L., Maxim M. N., Maria R. K., Ilia A. M., Vladimir L. V., Vladimir A. A., Elena G. D., Mariya B. C., Yury V. K., Denis A. V., Vladimir V. P., Yulia A. K., Dmitry V. K., Olga P. C., Alexander P. S., Alina Y. B., Olga A. S.. Terahertz spectroscopy of diabetic and non-diabetic human blood plasma pellets. J. Biomed. Opt. (2021) 26 1-14. DOI: 10.1117/1.JBO.26.4.043006 17. Sun Q., Liu K., Chen X., Liu X., Hernandez-Serrano A., Pickwell-MacPherson E.. Utilizing multilayer structures to enhance terahertz characterization of thin films ranging from aqueous solutions to histology slides. Opt. Lett. (2019) 44 2149-2152. DOI: 10.1364/OL.44.002149 18. Ding X., Hernandez-Serrano A. I., Lindley-Hatcher H., Stantchev R. I., Zhou J., Pickwell-MacPherson E.. Optimized multilayer structure for sensitive THz characterization of thin-film glucose solutions. Opt. Express (2022) 30 18079-18089. DOI: 10.1364/OE.451633 19. Tewari P., Garritano J., Bajwa N., Sung S., Huang H., Wang D., Grundfest W., Ennis D. B., Ruan D., Brown E., Dutson E., Fishbein M. C., Taylor Z.. Methods for registering and calibrating in vivo terahertz images of cutaneous burn wounds. Biomed. Opt. Express (2019) 10 322-337. DOI: 10.1364/BOE.10.000322 20. Osman O. B., Jack Tan T., Henry S., Warsen A., Farr N., McClintic A. M., Wang Y.-N., Arbabi S., Arbab M. H.. Differentiation of burn wounds in an in vivo porcine model using terahertz spectroscopy. Biomed. Opt. Express (2020) 11 6528-6535. DOI: 10.1364/BOE.397792 21. Taylor Z. D., Garritano J., Sung S., Bajwa N., Bennett D. B., Nowroozi B., Tewari P., Sayre J. W., Hubschman J. P., Deng S. X., Brown E. R., Grundfest W. S.. THz and mm-Wave Sensing of Corneal Tissue Water Content: In Vivo Sensing and Imaging Results. IEEE Trans. Terahertz Sci. Technol. (2015) 5 184-196. DOI: 10.1109/TTHZ.2015.2392628 22. Ozheredov I., Prokopchuk M., Mischenko M., Safonova T., Solyankin P., Larichev A., Angeluts A., Balakin A., Shkurinov A.. In vivo THz sensing of the cornea of the eye. Laser Phys. Lett. (2018) 15 055601. DOI: 10.1088/1612-202X/aaac76 23. Hernandez-Cardoso G. G., Rojas-Landeros S. C., Alfaro-Gomez M., Hernandez-Serrano A. I., Salas-Gutierrez I., Lemus-Bedolla E., Castillo-Guzman A. R., Lopez-Lemus H. L., Castro-Camus E.. Terahertz imaging for early screening of diabetic foot syndrome: A proof of concept. Sci. Rep. (2017) 7 42124. DOI: 10.1038/srep42124 24. Kim K. W., Kim K.-S., Kim H., Lee S. H., Park J.-H., Han J.-H., Seok S.-H., Park J., Choi Y., Kim Y. I., Han J. K., Son J.-H.. Terahertz dynamic imaging of skin drug absorption. Opt. Express (2012) 20 9476-9484. DOI: 10.1364/OE.20.009476 25. Wang J., Lindley-Hatcher H., Liu K., Pickwell-MacPherson E.. Evaluation of transdermal drug delivery using terahertz pulsed imaging. Biomed. Opt. Express (2020) 11 4484-4490. DOI: 10.1364/BOE.394436 26. Lee G., Namkung H., Do Y., Lee S., Kang H., Kim J.-W., Han H.. Quantitative label-free terahertz sensing of transdermal nicotine delivered to human skin. Curr. Opt. Photon. (2020) 4 368-372 27. Costa G., Ding X., Lindley-Hatcher H., Stantchev R., Hernandez-Serrano A. I., Pickwell-MacPherson E.. Using THz-TDS for assessing the effects of transdermal drug delivery methods on skin. (2022) 1-2 28. Lindley-Hatcher H., Hernandez-Serrano A. I., Sun Q., Wang J., Cebrian J., Blasco L., Pickwell-MacPherson E.. A Robust Protocol for In Vivo THz Skin Measurements. J. Infrared, Millimeter, Terahertz Waves (2019) 40 980-989. DOI: 10.1007/s10762-019-00619-8 29. Wang J., Stantchev R. I., Sun Q., Chiu T.-W., Ahuja A. T., MacPherson E. P.. THz in vivo measurements: the effects of pressure on skin reflectivity. Biomed. Opt. Express (2018) 9 6467-6476. DOI: 10.1364/BOE.9.006467 30. Huang S., Ashworth P. C., Kan K. W. C., Chen Y., Wallace V. P., Zhang Y.-T., Pickwell-MacPherson E.. Improved sample characterization in terahertz reflection imaging and spectroscopy. Opt. Express (2009) 17 3848-3854. DOI: 10.1364/OE.17.003848 31. Chen X., Parrott E. P. J., Ung B. S., Pickwell-MacPherson E.. A Robust Baseline and Reference Modification and Acquisition Algorithm for Accurate THz Imaging. IEEE Trans. Terahertz Sci. Technol. (2017) 7 493-501. DOI: 10.1109/TTHZ.2017.2722981 32. Sun Q., Parrott E. P. J., He Y., Pickwell-MacPherson E.. In vivo THz imaging of human skin: Accounting for occlusion effects. J. Biophotonics (2018) 11 e201700111. DOI: 10.1002/jbio.201700111 33. Gioia F., Celleno L.. The dynamics of transepidermal water loss (TEWL) from hydrated skin. Skin. Res. Technol. (2002) 8 178-186. DOI: 10.1034/j.1600-0846.2002.10342. 34. Sun Q., Chen X., Liu X., Stantchev R. I., Pickwell-MacPherson E.. Exploiting total internal reflection geometry for terahertz devices and enhanced sample characterization. Adv. Opt. Mater. (2020) 8 1900535. DOI: 10.1002/adom.201900535 35. Ding S.. Imaging article. figshare (2022) 36. Ding S.. THz-TDS article: data. figshare (2022)
--- title: Prevalence of and Risk Factors for Back Pain Among Young Male Conscripts During Compulsory Finnish Military Service authors: - Saara Suikkanen - Harri Pihlajamäki - Mickael Parviainen - Hannu Kautiainen - Ilkka Kiviranta journal: Military Medicine year: 2022 pmcid: PMC10026615 doi: 10.1093/milmed/usab375 license: CC BY 4.0 --- # Prevalence of and Risk Factors for Back Pain Among Young Male Conscripts During Compulsory Finnish Military Service ## ABSTRACT ### Introduction Back pain is a major reason for sick leaves and disability pension in primary health care. The prevalence of back pain among adolescents and young adults is believed to be increasing, and back pain during military service predicts unspecified back pain during later life. The aim of this study was to investigate the prevalence and risk factors of back pain among conscripts in compulsory Finnish military service during the period 1987-2005. ### Materials and Methods The Finnish Defence Forces recruit all men aged 18 years for compulsory military service, and new conscripts enter the service twice a year. Before entering the service, all conscripts must pass a medical examination and conscripts entering the service are generally healthy. Health care in Finnish military service is organized by the public Garrison Health Center, and all medical records are stored as part of the Finnish health care operation plan. For this study, we randomly selected 5,000 men from the Finnish Population Register Centre, according to their year of birth from five different age categories (1969, 1974, 1979, 1984, and 1989). ### Results We gathered 4,029 documents for the analysis. The incidence of back pain varied between $18\%$ and $21\%$ and remained unchanged during the examination period. The risk factors for back pain were smoking (risk ratio 1.35, P-value <.001), elementary school only as education (risk ratio 1.55, P-value <.001), and back problems reported before military service (risk ratio 2.03, P-value.002). Half of the back pain incidences occurred during the first months of service. ### Conclusions The prevalence of back pain among male Finnish military service conscripts has not changed in the last 25 years. Twenty percent of conscripts suffer from back-related problems during their military service. The majority of the visits to health centers occurred in the first service months. The risk factors for back pain include smoking, low education level, and musculoskeletal disorders in general. Educating the young people about harms of tobacco and supporting education is a way to influence the back pain prevalence. Strength of this study is a good generalized population sample of young Finnish adult males because of the fact that the Finnish military service is compulsory for all men. All medical records of all visits to the Garrison Health Care Centre were available, and all the conscripts filled the same pre-service questionnaire, minimizing the possibility of selection bias. The sample size was also large. Weakness of this study is that the service time changed during the study period and in the latest conscript group born in 1989, data collection and the data available for this cohort was limited, because nearly half of the conscripts had not yet started their service. The Finnish military service is compulsory only for men and because of the low number of female conscripts, they were excluded from this study. Diagnoses were also missing from $70\%$ of the back-related visits, and these visits were recorded as back pain-related visits according to the reason for seeking care. ## INTRODUCTION Back pain is one of the main reasons for visiting a physician and a major cause of sick leaves.1 The financial burden to society and the humane suffering caused by back pain-related problems is considerable, and prevention methods are of high importance.2 The prevalence of back pain is high among the older population and also among adolescents and young adults.2,3 Finnish military service conscripts are no different in this respect, and back pain during military service predicts unspecified back pain during later life.2,3 The Finnish Defence Forces recruit all men aged 18 years for compulsory military service. One can be exempted for health or religious reasons or can alternatively choose civilian service. New conscripts enter the military service twice a year, in January and in July. Before entering the service, all conscripts must pass a medical examination. Conscripts suffering from diseases that could jeopardize their service are exempted. Conscripts are generally healthy and do not suffer from difficult back problems when military service begins. The service begins with an 8-week basic training period containing 24 h/week of physical activity of increasing intensity.4 At the end of the basic training period, the conscripts gain the fitness level required for military training. After the 8-week basic training period, the physical exercise level is reduced to 15 h/week, but the intensity of the military training increases over the next 4 months. The length of service depends on the assigned position and training. Until 1995, the military service lasted 9 or 11 months, but since then, service periods have been 6, 9, or 12 months. Women have been able to apply for voluntary military service since 1994. The number of female conscripts is less than $3\%$. Nearly $80\%$ of all Finnish men start military service, and about $15\%$ of conscripts discontinue service every year.5 Musculoskeletal disorder and mental problems are the main reasons for discontinuing, and musculoskeletal disorders show an increasing trend.6 The majority of musculoskeletal problems during military service are related to back pain and lower extremity problems.6 Little is known about back pain prevalence among conscripts during military service. Most research has been conducted in recruiter forces 7 or has focused on hospitalization caused by back pain.6 Because nearly all Finnish men participate in military service, the Finnish Defence Forces provide a good data sample for population-based studies. The aim of this investigation was to study the risk factors for back pain and its prevalence among Finnish male conscripts. TABLE I. \| Year of birth \| 1969 \| 1974 \| 1979 \| 1984 \| 1989 \| \| --- \| --- \| --- \| --- \| --- \| --- \| \| Back-related visits to physician during military service, n (%)a \| 166 (21%) \| 165 (18%) \| 179 (20%) \| 167 (19%) \| 104 (20%) \| \| No back-related visits to physician during military service, n (%) \| 642 (79%) \| 752 (82%) \| 720 (80%) \| 712 (81%) \| 422 (80%) \| ## MATERIALS AND METHODS We randomly selected 5,000 men from the Finnish Population Register Centre according to their year of birth (1969, 1974, 1979, 1984, and 1989). Each age cohort comprised 1,000 men. Of the participants selected for this study, 971 were discharged from the data because they did not complete their service during the examination period because of health reasons or they carried out civil service or postponed their service. Finally, we collected the data of 4,029 army conscripts for analysis. When entering military service, all conscripts complete a pre-service questionnaire assessing their socio-economical and general health factors. This questionnaire is described.21 We carefully assessed all visits to the Garrison Health Care Centre during military service from military service documents. Visits to the Garrison Health Care Centre resulting in a back-related diagnosis or when a back problem was the reason for seeking care were recorded as a back-related visit. ## STATISTICAL ANALYSES The data are presented as means with SD or $95\%$ CIs. The $95\%$ confidence interval was calculated using a bootstrap with 1,000 repetitions. We tested the statistical significance of means between two groups using the t-test, and the means between multiple groups using one-way analysis of variance. The Kaplan-Mayer estimate was used for time-to-event analysis. A multivariate Poisson regression model with a robust estimate of variance was used to calculate incidence risk ratios. We tested the significance of the factors using a calculated z-value of standardized normal distribution. The STATA 13.1, StataCorp LP (College Station, TX, USA) statistical package was used for all statistical analyses. The Medical Ethics Committee of the Finnish Defence Forces approved the study plan. The Ethics Committee of the Hospital District of Helsinki and Uusimaa also approved the study (designated $\frac{267}{13}$/$\frac{03}{09}$). ## RESULTS The mean age of the men at the beginning of military service was 19.2 years (SD 1.1), and their body mass index (BMI) 23.3 kg/m2 (SD 3.8 Kg/m2). Altogether 781 of the 4,029 study participants ($19.4\%$) visited the Garrison Health Centre because of back-related problems. The number of men with back problems has remained stable over the years (Table I). The incidence was 268 visits per 1,000 conscript years ($95\%$ CI: 250-287). The total number of visits to the Garrison Health Care Centre because of back pain was 1,694. About $80\%$ of the conscripts did not complain of back-related problems during their service, and less than $10\%$ of the conscripts visited the Garrison Health Care Centre more than once because of back-related problems. Of the 4,029 participating conscripts, 191 ($5\%$) visited the Garrison Heath Centre more than twice because of back pain. This $5\%$ of the conscripts contributed to $55\%$ of back pain-related visits to the Garrison Health Centre. Seventy percent of the back-related visits to the Garrison Health Care Centre resulted in no diagnosis. These visits were recorded as back-related visits, according to the reason for care seeking. Thirty percent of visits resulted in a back-related diagnosis, $95\%$ of which were unspecified back pain. One percent of the diagnoses were scoliosis-related back pain and two percent sciatic back pain. The remaining $3\%$ were classified as having osteoarthritis of the spine or other connective tissue disorders. Half of the visits occurred during the first 2 months of training (Fig. 1). The hazard rate of visiting the Garrison Heath Centre because of back problems was highest during the first and second service months, after which it gradually decreased (Fig. 1). FIGURE 1.: Cumulative incidence and hazard rate of back pain during military service. Half of the visits to Garrison Health occurred during the first 2 months of service. The hazard rate of visiting the Garrison Heath Centre because of back problems was highest during the first and second service months. Smoking and low education level before service and back problems reported before military service were associated with back pain during military service predicting visits to the Garrison Health Care Centre. Body mass index had no influence on the incidence of visits (Table II). TABLE II. \| General information \| Incident risk ratio (95% CI) \| P-value (P>\|z\|, two-sided test) \| \| --- \| --- \| --- \| \| BMIa \| 1.00 (0.99 to 1.03) \| .41 \| \| Elementary school education only \| 1.55 (1.30 to 1.84) \| <.001 \| \| Exercise over two times/week \| 1.10 (0.93 to 1.29) \| .27 \| \| Smoking \| 1.35 (1.14 to 1.60) \| <.001 \| \| Self-reported previous accident \| 1.00 (0.79 to 1.26) \| .99 \| \| Musculoskeletal symptoms reported before military service \| 0.97 (0.66 to 1.43) \| .88 \| \| Respiratory symptoms reported before military service \| 1.04 (0.86 to 1.25) \| .69 \| \| Gastrointestinal symptoms reported before military service \| 1.14 (0.50 to 1.18) \| .59 \| \| Psychological disorders reported before service \| 0.77 (0.50 to 1.18) \| .23 \| \| Back problems reported before military service \| 2.03 (1.29 to 3.19) \| .002 \| $93\%$ of the conscripts had BMI 30 or less, and $76\%$ of the conscripts were considered normal or underweight (BMI 25 or less). Visits to Garrison Heath Care because of low back pain did not differ between different BMI categories (Table III). TABLE III. \| BMI category \| BMI <18 (underweight) \| BMI 18-25 (normal weight) \| BMI 25-30 (overweight) \| BMI >30 (obesity and severe obesity) \| \| --- \| --- \| --- \| --- \| --- \| \| No back-related visits to Garrison Health Care \| 81% a(89b) \| 81% (2,399) \| 78% (551) \| 79% (209) \| \| One back-related visit to Garrison Health Care \| 9% (10b) \| 11% (316) \| 12% (85) \| 8% (20) \| \| 2-3 back-related visits to Garrison Health Care \| 8% (9) \| 5% (151) \| 7% (49) \| 9% (24) \| \| > 3 back-related visits to Garrison Health Care \| 2% (2) \| 3% (85) \| 3% (19) \| 4% (11) \| ## DISCUSSION This study showed that the number of conscripts suffering from back problems during military service has not changed during the 20-year time-period studied. Conscripts are most vulnerable to back problems during the first 2 months of service, after which the risk gradually decreases. Reported low education level, smoking, and musculoskeletal disorders before entering service, including back-related problems, increase the risk of developing back problems during service. Approximately, $20\%$ of Finnish military service conscripts visit the Garrison Health Centre because of back-related problems. A previous study of 15-64 year-old Finnish citizens showed similar results concerning the prevalence of diagnosed back problems.8 A study focusing on musculoskeletal disorders among Finnish conscripts during military service in 1967-2006 showed a 1.6-fold increase of health service utilization, mainly for lower limb and back problems.6 According to these findings the increase of health service utilization among Finnish conscripts because of back problems is not because of an increased amount of patients, but rather because of an increased number of visits to a physician per one patient. The incidence of back pain during military service was 268 visits per 1,000 conscript years. This is clearly higher than that in the U.S. Military Service of active duty members, which is 40.5 per 1,000 conscript years.7 This can be explained by the difference of the military servants in these armies. In Finland, the military service is compulsory for all men, whereas in the USA, the military service is recruitment based. Most of the conscripts who visit the Garrison Health Care Centre visit the physician only once or twice during their military service because of back pain, and most of these visits take place during the first months of service. The physical condition of conscripts is diverse when they enter military service. Poor physical fitness increases the risk of back pain during military service.9 On the other hand, heavy physical load also increases the risk of back pain10 and the amount of physical activity at the beginning of the Finnish military service for those with a sedentary background can be considered heavy (24 h/week of physical training with increasing intensity). The discrepancy between poor physical condition and physically heavy training is most likely one reason for back-related problems occurring during the first months of service. Low education level increased the risk for back pain 1.35-fold in this study. Low education level is a known risk factor for back pain.10,11 The mechanism of low education level causing back pain is unknown. One explanation is that people with a lower education level tend to work in more physically demanding jobs as they have less work options.12 However, occupation explains only part of the phenomena.12 In Finland, education is compulsory for all Finnish citizens until the age of 17 years. The conscripts entering the military service are relatively young, and most come straight from school or have only been in working life for a few years. Thus, occupational differences cannot fully explain the strong effect of education on back pain in this study. Another explanation for the strong correlation between education level and back pain is that educated people are more aware of healthy lifestyle habits and may consciously select them.13 Education also provide better problem-solving abilities and increase compliance with medical programmes.12 In this study, smoking was a clear risk factor for back pain and previous studies have also shown an association between smoking and back pain.14 Tobacco affects through atherosclerosis of the vessels that nourish the lumbar spine, which causes degenerative changes in the vertebrae and also directly affects cell proliferation and collagen fibrils locally.15,16 In addition to adults, smoking also increases the likelihood of developing back pain among adolescents.17,18 The first radiological signs of aortic atherosclerosis are apparent already in young adulthood, but the vascular alteration to the spine tends to build up later.19 For this reason, effective methods for preventing smoking may reduce the socioeconomical burden caused by back pain in addition to other health benefits. In this study, BMI showed no correlation with back pain. Previous studies among Finnish conscripts and young people have shown a moderate correlation between high BMI and back pain.20–22 Vast majority of the conscripts were normal weight (BMI 25 or less), and only $7\%$ of the conscripts were obese having BMI over 30. One explanation for the missing association between BMI and back pain in this study is that conscripts suffering from severe obesity are discharged from military service for health reasons. Data were not available of these discharged conscripts to verify this assumption. This study contained only young male adults, and the follow-up time was relatively short to reveal the impact of slowly varying variables, such as BMI, on back pain. Back pain alone is a strong risk factor for future pack pain.23,24 In this study, previous back pain doubled the risk of future back pain and the same results have been found in a previous study.24 Unfortunately, the high prevalence of back pain makes prevention methods difficult.24 This study has many strengths. The Finnish military service is compulsory for all men and only $20\%$ of conscripts either discontinue their service or choose civil service instead. This guaranteed a good generalized population sample of young Finnish adult males. The participants were randomly selected, and their medical records of all visits to the Garrison Health Care Centre were available. All the conscripts filled the same pre-service questionnaire, minimizing the possibility of selection bias through the questionnaire form. The sample size was also large. A weakness of this study is that the service time changed during the study period, also causing alternation in the content and physical activities during service. The Finnish military service is only compulsory for men, and the number of women in the service is low. Because of the low number of female conscripts, they were excluded from this study. The coding method for diseases (ICD) also changed during the study period, and this might have had an impact on coding consistency. In addition, diagnoses were missing from $70\%$ of the back-related visits, and these visits were recorded as back pain-related visits according to the reason for seeking care. Because of this, it was not possible to categorize the back pain visits into specific and non-specific back pain visits. Data were available only from Garrison Health Care of conscripts who completed the service. The lacking of data of conscripts discharged from the service may alter the BMI distribution and be reason for the correlation of back pain and BMI in this study. In the latest conscript group born in 1989, data collection and the data available for this cohort were limited, because nearly half of the conscripts had not yet started their service. On average, $80\%$ of the men in this study participated in military service and the enrollment percentage is in line with Finnish long-term statistics.5 In conclusion, the percentage of conscripts with back-related problems in military service has remained stable at around $20\%$ over the last 20 years. Smoking and musculoskeletal problems before military service tend to increase the prevalence of back pain among conscripts, and low education level and smoking are risk factors for developing back pain during service. Therefore, smoking cessation and measures to increase education levels should be among the preventive methods aiming to decrease back pain incidence among young men. ## FUNDING None declared. ## CONFLICT OF INTEREST STATEMENT None declared. ## References 1. Olafsson G, Jonsson E, Fritzell P, Hägg O, Borgström F. Cost of low back pain: results from a national register study in Sweden. Eur Spine J (2018) 27 2875-81. PMID: 30155730 2. Hestbaek L, Larsen K, Weidick F, Leboeuf-Yde C. Low back pain in military recruits in relation to social background and previous low back pain. A cross-sectional and prospective observational survey. BMC Musculoskelet Disord (2005) 26 3. Mattila VM, Sahi T, Jormanainen V, Pihlajamäki H. Low back pain and its risk indicators: a survey of 7,040 Finnish male conscripts. Eur Spine J (2008) 17 64-9. PMID: 17874146 4. Mattila V, Kuronen P, Pihlajamäki H. Nature and risk factors of injury hospitalization in young adults: a follow-up of 135,987 military conscripts. Scand J Public Health (2007) 35 418-23. PMID: 17786806 5. Findikkaattori. (2017) 6. Frilander H, Miranda H, Mutanen P, Mertelin T, Pihlajamäki H, Viikari-Juntura E. Trends in musculoskeletal disorders and related health care utilization among conscripts in Finland, 1967-2006. Mil Med (2012) 177 1069-74. PMID: 23025137 7. Knox J, Orchowski J, Scher D, Owens B, Burks R, Belmont P. The incidence of low back pain in acite duty united states military service members. Spine (2011) 36 1492-500. PMID: 21224777 8. Leino PI, Berg MA, Puska P. Is back pain increasing? Results from national surveys in Finland during 1978/9–1992. Scand J Rheumatol (1994) 23 269-76. PMID: 7973482 9. Taanila HP, Suni JH, Pihlajamäki HK. Predictors of low back pain in physically active conscripts with special emphasis on muscular fitness. Spine J (2012) 12 737-48. PMID: 22297262 10. Latza U, Kohlmann T, Deck R, Raspe H. Influence of occupational factors on the relation between socioeconomic status and self-reported back pain in a population-based sample of german adults with back pain. Spine (2000) 1 1390-7 11. Hoy D, Brooks P, Blyth F, Buchbinder R. The epidemiology of low back pain. Best Pract Res Clin Rheumatol (2010) 24 769-81. PMID: 21665125 12. Suman A, Bostick GP, Schaafsma FG, Anema JR, Gross DP. Associations between measures of socio-economic status, beliefs about back pain, and exposure to a mass media campaign to improve back beliefs. BMC Public Health (2017) 504. DOI: 10.1186/s12889-017-4387-4. 13. Hagen KB, Holte HH, Tambs K, Bjerkedal T. Socioeconomic factors and disability retirement from back pain: a 1983–1993 population-based prospective study in Norway. Spine (2000) 25 2480-7. PMID: 11013500 14. Golberg MS, Scott SC, Mayo NE. A review of the association between cigarette smoking and the development of nonspecific back pain and related outcomes. Spine (2000) 25 995-1014. PMID: 10767814 15. Yanbaeva DG, Denteren MA, Creutzberg EC, Wesseling G, Wouters EF. Systemic effects of smoking. Chest (2007) 131 1557-66. PMID: 17494805 16. Akmal M, Kesani A, Anand B, Singh A, Wiseman M, Goodship A. Effect of nicotine on spinal disc cells: a cellular mechanism for disc degeneration. Spine (2004) 29 568-75. PMID: 15129075 17. Feldman DE, Rossignol M, Shrier I, Abenhaim L. Smoking: a risk factor for development of low back pain in adolescents. Spine (1999) 24 2492-6. PMID: 10626312 18. Feldman DE, Shrier I, Rossignol M, Abenhaim L. Risk factors for the development of low back pain in adolescence. Am J Epidemiol (2001) 154 30-6. PMID: 11427402 19. Kauppila LI. Atherosclerosis and disc degeneration/low-back pain- a systematic review. Eur J Vasc Endovasc Surg (2009) 37 661-70. PMID: 19328027 20. Mikkonen PH, Jaana L, Remes J, Tammelin T. Association between overweight and low back pain. Spine (2013) 38 1026-33. PMID: 23459137 21. Taanila H, Suni J, Pihlajamäki H, Mattila V, Ohrankämmen O, Vuorinen P. Aetiology and risk factors of musculoskeletal disorders in physically active conscripts: a follow-up study in Finnish Defence Forses. BMI Musculoskelet Disord (2010) 5 22. Frilander H, Solovieva S, Mutanen P, Pihlajamäki H, Heliövaara M, Viikari-Juntura E. Role of overweight and obesity in low back disorders among men: a longitudinal study with a life course approach. BMJ Open (2015) 5. DOI: 10.1136/bmjopen-2015-007805 23. Rossignol M, Lortie M, Ledoux E. Comparison of spinal health indicators in predicting spinal status in a 1-year longitudinal study. Spine (1993) 18 54-60. PMID: 8434325 24. Papageorgiou AC, Croft PR, Thomas E, Ferry S, Jayson M, Silman AJ. Influence of previous pain experience on the episode incidence of low back pain: results from the South Manchester Back Pain Study. Pain (1996) 66 181-5. PMID: 8880839
--- title: Two years of approved digital health applications in Germany – Perspectives and experiences of general practitioners with an affinity for their use authors: - Julian Wangler - Michael Jansky journal: The European Journal of General Practice year: 2023 pmcid: PMC10026738 doi: 10.1080/13814788.2023.2186396 license: CC BY 4.0 --- # Two years of approved digital health applications in Germany – Perspectives and experiences of general practitioners with an affinity for their use ## Body KEY MESSAGESGPs with experience in DHAs see overall added value in effects on healthcare, especially compared to health apps. Suitable conditions must prevail before DHAs can act as effective support tools for primary care. This requires informing GPs on the use of DHAs as well as addressing concerns. ## Abstract ### Background Since 2020, physicians in Germany can prescribe approved digital health applications (DHAs) with the costs covered by the health system. There has so far been a lack of studies on attitudes and experiences amongst GPs in using DHAs. ### Objectives The aim was to elucidate the experiences and observations of GPs that have used DHAs in health care and to examine the conditions necessary for DHAs to gain a foothold in primary care according to the GPs. ### Methods In 2022, 96 qualitative semi-standardised interviews were conducted with German GPs with experience in prescribing DHAs. The GPs were all organised in digitalisation-oriented physicians’ associations. Fifty-four interviews were carried out in person and 42 by phone. The data were analysed according to qualitative content analysis. ### Results Unlike health apps, the interviewees saw DHAs as reliable tools for enhancing the relationship between GPs and their patients. They saw the DHAs they had been prescribing as useful and reported various benefits, including improvements in compliance, mobility, information for patients and weight reduction. The physicians also saw room for further improvement (usability, gamification, training, information sources). Interviewees saw the inclusion of DHAs in evidence-based guidelines as a major step forward. ### Conclusion The interviewees rated DHAs favourably regarding healthcare potential and as safer and more reliable than conventional health apps. Many saw benefits to healthcare from using such applications. From the interviewees’ point of view, DHAs can be integrated more effectively into patient care. ## Introduction In Germany, high-quality approved digital health applications (DHAs) have been integrated into standard care since 2020 by law (Improving Healthcare by Digitalisation and Innovation Act, DVG) – a step that was unique in the world at that time and still is [1]. Since then, physicians have been able to prescribe DHAs to patients with costs covered by the national health system. DHAs are set to make disease diagnostics and recognition more effective, support treatment and contribute to prevention [1–8]. Like freely available health apps, DHAs are aimed at reinforcing empowerment, motivation, and compliance as well as informing patients and encouraging a healthy lifestyle [1,3,9–11]. Coverage by the national health system is conditional upon an application being included in the national DHA directory [1,2]. This requires manufacturers to apply for approval during an audit process on various requirements (CE markings for medical devices, data protection, security standards, information quality, usability and robustness in operation, patient safety). They are also required to provide sufficient documentation for the added value of the application in its effect on healthcare [1,10]. Once these criteria have been met, the application can be included in the application directory and prescribed. A fast-track procedure is also available for preliminary inclusion in the directory if only the general criteria have initially been fulfilled [1,2,7]. This gives manufacturers one year to have the application tested pending documentation of a beneficial effect. There are various documentable categories of impact on healthcare, such as pain relief, improved information and empowerment with enhanced disease management. GPs play a key role in successfully establishing DHAs in healthcare [12,13]. A plausible scenario would be for GPs to use such tools in targeted prevention medicine and disease management, while also managing the process of change and receiving health data from their patients regularly [1,3,4,7,12,14–19]. So far, little is known about attitudes and behavioural patterns amongst GPs concerning the use of digital health applications. The same applies to experiences in the practical use of these applications in primary care, so there is a need for a broader investigation. ## Research aims The present study has reached an interim assessment from the viewpoint of GPs that are already interested in mHealth tools and prescribe DHAs in patient care. The focus of the study is twofold. First, it was focussed on determining these GPs’ attitudes, experiences about DHAs and their effects on healthcare. Second, it should be determined to what extent GPs perceive an additional value of DHAs compared to ordinary, freely available health apps. We aimed to deliver results as a basis to deduce the conditions necessary for tapping into the potential of digital health applications in healthcare, especially primary care. ## Study design Against the background of our work with the related topic of health apps [17,19,20], we regard this study as an in-depth study specifically focussed on DHAs. Consequently, there are intersections with previous studies on conventional health apps, however, there are specifics due to the described DHA approach. These were to be examined from a GP's perspective. Thus, a qualitative approach appeared most appropriate. ## Recruitment and sampling This study involved creating a convenience sample. Thirty regional and federal physicians’ associations focussed on digitalisation were included in recruitment by post for a study that was to explore attitudes and impressions from GPs as well as their experiences of using DHAs. The associations we approached had outpatient physicians as members (in most cases mere GP associations, in some cases combined GP and specialist associations), who were engaged in discussion and regular exchanges of experience in the topic, especially in committees on quality. These physicians collaborate and keep each other up to speed in more training programmes to keep abreast of practical implementation and digitalisation and integrate digital tools in their medical practices. We mostly approached the physicians’ associations through their websites. We contacted the medical practices acting as coordination centres for the respective association. We aimed to win only GPs from the associations mentioned. Ninety-six GPs replied, and we finally conducted the interviews with these GPs without offering incentives. We were able to run the interviews with all 90 physicians because it was possible to recruit approximately the same number of GPs in each of the 16 German federal states. The interviews took place between March and October 2022 and were conducted by both authors (general practice researchers), each conducting half of the interviews. Fifty-four interviews were carried out in person, 42 by phone (45 to 85 min). The interviews were recorded. We sent our interviewees an explanation of the topic as well as a written declaration of consent for them to sign before the interview. The first author took care of transcription. Table 1 provides an overview of the participating sample. All GPs included in this study are members of a physicians’ association. Table 1. \| Type of practice \| 56 joint practices, 40 single practices \| \| --- \| --- \| \| Practice setting \| 19 in a village or small town, 36 in a medium-sized town, 41 in a city \| \| Status \| 61 practice owners, 35 GPs in employment \| \| Age \| mean: 51 years, range: 22 \| \| Gender \| 59 male, 37 female \| ## Investigation tools Several quantitative and qualitative research studies by the authors [17,19,20] with various areas of focus on the application areas for using health apps in primary care and specialist clinical settings as well as desk research [3,5,7,10,14,16,18,21,22, inter alia] were used in designing the interview guideline (see Supplementary Appendix). The guideline consisted of 32 open questions with four key areas derived from the authors’ preliminary studies [17,19,20]. The focal points are: Prevalence of mHealth tool use amongst patients; attitudes towards digital health applications; prescription policy and experience in using these applications in healthcare; assessment, further development, and establishment of DHAs. ## Data analysis Both authors evaluated the transcripts using content analysis according to Mayring using the MAXQDA software [19]. This first entailed pinpointing the key statements followed by further abstraction and summarisation, finally leading to a categorised system closely based on the interview guidelines and repeatedly reviewed and modified as necessary during evaluation. Our focus lies on forming logical categories from the various opinions and experiences. Theoretical saturation became apparent after 73 interviews. However, we had set the prior condition that all 96 interviews were to be conducted. ## Ethics The study did not involve collecting patient data or conducting clinical tests. All 96 interviews were strictly anonymised. The Ethics Commission of the State of Rhineland-Palatinate informed us that approval by an ethics committee would not be necessary. The researchers identified the participants and requested their written consent to participate. ## Spreading of the use of mHealth Most interviewees estimated that up to a fifth of their patients used mHealth tools, such as health apps, at least occasionally. However, users comprised a ‘heterogeneous group’ and ‘you can no longer claim it was specifically younger or especially digitally minded people’ using them (I-26f). The lion’s share of respondents estimated the potential for patients generally interested and ready to use mHealth tools to amount to around a third, judging from their own patients at their medical practices. Most saw DHAs as ‘a great step forwards […] in winning increasing numbers of patients over to digital forms of support in healthcare’ (I-44m). ## Assessment of the DHA concept Almost all participants had been introduced to DHAs early due to the digital focus of the physicians’ associations the interviewees belonged to. Some associations provided information on DHAs to the physicians straight after the new DVG law had been passed, which the interviewees described as ‘extremely useful’ (I-4m) as they had ‘a form of support and consultation from the start’ (I-28m). This advantageous head start in information from the association’s involvement helped the vast majority of respondents recognise ‘the clear asset that this new type of health application would be from the get-go’ (I-74f). Around half the sample reported that despite general openness towards DHAs initially, they had ‘not always had favourable experiences with ordinary health apps for various reasons’ in the past (I-52m). DHAs are generally considered reliable tools that physicians can ‘prescribe and recommend without uncertainty or worry’ (I-38f). The same applied to issues regarding legal certainty, although several interviewees still had questions and expressed ‘certain remaining doubts’ (I-44m), especially about the risk of data collection errors and liability issues. Even so, there was still a ‘basic trust’ as ‘digital health applications have a legal framework behind them placing significantly stricter demands on content quality […]’ (I-38f). More than half the interviewees expected DHAs to play an especially significant role or make an outstanding contribution to healthcare and/or convalescence if appropriately used; the others saw a contribution but expected this to be relatively small in what would only be a supporting role. Digital health applications rated noticeably higher than conventional health apps by importance and status in clinical healthcare. Respondents attributed this to the BfArM audit, which ‘brings in a modicum of safety and reliability’ (I-88f). ## Perceived benefits and risks of DHAs Perceived benefits from apps varied by application area. Almost all interviewees thought it would make sense if the applications helped in self-monitoring for risk factors such as weight, blood pressure, and blood sugar; lifestyle changes such as diet, quitting smoking, coping with psychological problems; as well as preventive measures and medication management. Two out of three respondents favoured direct support in monitoring and treating chronic diseases. Respondents especially saw increased motivation and compliance as the greatest benefits of using DHAs in a clinical setting (Table 2). Increased empowerment, health literacy, and reaching new patient types were also significant. Table 2. \| Question: What are the most important benefits of using digital health applications in a clinical setting? Where would you see drawbacks or risks? (n = 96) \| Number of mentions \| \| --- \| --- \| \| Benefits \| \| \| Enhancing motivation \| 71.0 \| \| Increasing empowerment \| 68.0 \| \| Improving compliance (inter alia medication adherence) \| 64.0 \| \| Improving appointment management \| 56.0 \| \| Reinforcing health awareness and education \| 56.0 \| \| Making treatment more individual and effective \| 49.0 \| \| Reaching new patient types \| 46.0 \| \| Earlier detection, diagnosis, and treatment of disease or risk of disease \| 44.0 \| \| Drawbacks \| \| \| Lack of data privacy, personal data protection \| 48.0 \| \| Measurement errors or treatment failure due to complicated use \| 44.0 \| \| Raising or inculcating health anxiety \| 34.0 \| \| Impersonal doctor-patient relationship \| 25.0 \| Some physicians also pointed out the potential effectiveness and efficiency benefits of doctor-patient networking, such as measuring health data using DHAs and transferring them to the medical practice, ideally by integrating them into the office software system. Several interviewees saw this as a possibility for treating diseases and health risks in a more targeted and individual manner. Regarding risks, some respondents reported concerns on lack of safety and specifically data privacy such as from existing data leaks despite the BfArM audit. Some also saw undesirable effects such as measurement errors due to insufficient suitability for certain patient types – ‘especially with complex applications that are not intuitive’ (I-34m). This could lead to incorrect health data being collected or, in extreme cases, treatment failure. One negative consequence might be that using digital applications unwisely could cause worry among patients already suffering from health anxiety. For example, a certain type of user logic in the application could prompt misinterpretation or limit fixation on certain parameters. ## Initial reasons for using DHAs Respondents reported various combinations of causes on why they initially saw DHAs as exciting and worthwhile. They quoted care deficits such as compliance-sensitive treatment support and sustainable lifestyle changes as prevention. Recommendations from colleagues and curiosity about digital tools also played a role. Again, many spoke highly of the reliable information they had been provided by the physicians’ associations. Over half the interviewees reported advising patients to use certain conventional health apps before DHAs were introduced, so there was some experience in using digital health apps, albeit haphazard. The willingness to recommend and use DHAs has changed according to many respondents. Around two-thirds reported that mHealth apps in medical practice had gained some regularity – ‘albeit at a limited level so far’ (I-38f). The increase in trust and reliability from the DVG law and the DHA concept provided the primary justification. ## Areas of application and expectations towards DHAs DHAs were mainly prescribed for prevention and self-control, lifestyle, and promoting exercise as specific application areas. Applications that often came up covered lifestyle changes in type 2 diabetes mellitus and severe obesity as well as prevention through exercise, dealing with depressive episodes, sleep disorders, and tinnitus. The interviewees raised the importance of DHAs being easy to understand and use with a clear design for them to consider recommending one. DHAs should protect personal data the most effectively possible while providing customisation options and encouraging patients to become more health-conscious through gamification. A significant share of the sample emphasised that doctors needed solid and reliable sources of information on the respective application as a further requirement. Some respondents named permanent inclusion in the DHA directory as a mandatory requirement for them to prescribe it. Almost all physicians responded that their prescribed applications had proven useful overall when asked about their general experiences. Benefits to healthcare and/or convalescence were widely observed. These especially applied to factors such as improved compliance and self-management in chronic disease, increased mobility, and noticeable weight reduction (Table 3). DHAs showed the most added value in prevention and self-control, health-oriented lifestyle and exercise promotion. Seven interviewees reported adverse effects – overcomplicated design overwhelming patients trying to use the app and negative impact on patients with health anxiety. Table 3. \| Question: What health effects have you seen from patients using approved digital health applications? (n = 96) \| Number of mentions \| \| --- \| --- \| \| Increased compliance, such as in taking medications and adherence to treatment \| 83 \| \| Improvement in health awareness and education \| 65 \| \| Improvement in self-management, such as in chronic disease \| 64 \| \| Weight reduction such as BMI, abdominal circumference, waist circumference \| 59 \| \| Increased mobility \| 62 \| \| Stable decrease in blood sugar (HbA1c) \| 41 \| \| Prevention of sequelae, such as diabetic foot syndrome and CHD \| 40 \| \| Decrease in complications, such as hypoglycaemia \| 38 \| \| Regression of psychological side effects, such as depression \| 37 \| \| Regression of metabolic syndrome \| 29 \| \| Elimination of the need for more severe treatment options, such as insulin therapy \| 25 \| ## Further development approaches Physicians with user experience outlined various areas of focus when responding to an open question on how to make DHAs more accessible and, therefore, more attractive for use in (general) medical care (Table 4). Many physicians lacked factual, reliable information on DHAs. There was much criticism of the DHA directory as it is not detailed enough and sometimes came too close to manufacturer information. In some cases, this was also linked to a more general criticism on what justifies the fast-track procedure. Some respondents suggested the national health portal as a viable information platform focussing on digital health applications. Table 4. \| Question: Judging from your previous experience with approved digital health applications, how do you think digital health applications could be improved, and what would you like to see? (n = 96) \| Number of mentions \| \| --- \| --- \| \| Reliable information platform from a reliable source focussing on digital health applications ideally managed by the state (one of the suggestions offered: German National Health Portal) \| 67 \| \| Optimisation or further optimisation of usability in digital health applications, especially towards simpler, more intuitive and target group-specific use to prevent measurement errors and incorrect use \| 60 \| \| Training programmes for physicians on using digital health applications, especially in primary care, with sufficient CME-certified training \| 49 \| \| More gamification elements, more interactive and light-hearted approach to patient guidance \| 45 \| \| Appropriate remuneration for medical services and additional effort involved with digital health applications on the German national medical fee schedule \| 45 \| \| Inclusion of digital health applications in (evidence-based) guidelines and other instructions, such as from professional organisations \| 42 \| \| Digital health applications should be designed to prevent health anxiety, such as by eliminating the possibility of misinterpretation by patients or focussing on individual health parameters \| 38 \| \| Increased advice and support in using digital health applications for patients from statutory health insurance organisations \| 36 \| \| Cancellation of the fast-track procedure for temporary application approval, tightening the evaluation procedure \| 32 \| \| Technical aspects of integrating digital health applications in medical practice, such as through cost-neutral functional connection to office software \| 29 \| \| Improvement and further improvement of information and data privacy and protection by drawing up more binding and uniform data protection standards for manufacturers \| 28 \| \| Clear exclusion of liability risks for physicians if, for example, a treatment error occurs due to a bug in a digital health application – responsibility and liability should not lie with service providers or patients \| 25 \| \| Members of all statutory health insurance organisations should receive bonuses or bonus programmes for using certain digital health applications regularly and transferring the data to the respective organisation \| 14 \| Despite the high level of satisfaction with the DHAs used, potential was seen in improving the user experience and usability as well as extending interactivity and gamification. Respondents also saw the inclusion of DHAs in evidence-based guidelines as a major step forward. Interviewees advocated a comprehensive further training programme for physicians towards encouraging more use of these applications in primary care. Many respondents reported that their associations had provided them with extensive information on using DHAs. The problem was that many GPs in Germany have little or no knowledge of the legal framework provided by the DVG law. Interviewees emphasised that DHAs should not lead to misinterpretations amongst patients or a limiting fixation on certain parameters being suggested due to the app’s user logic. Many were concerned that the widespread use of DHAs might trigger a self-medication process in certain patients, resulting in doctors’ advice declining in importance. Many GPs wished for health insurance companies to approach patients with more advice and support in using DHAs. Even as it stands, statutory health insurance organisations will provide DHAs without an explicit doctor’s prescription, given the corresponding indication [9]. ## Main findings Our interviews showed that GPs with experience in health apps rated DHAs favourably regarding healthcare potential and as much safer and more reliable than conventional health apps. Such applications were especially beneficial in supporting prevention, self-control, and lifestyle changes. The same applies to practical experience with DHAs: Respondents mainly reported observing beneficial healthcare effects. Despite the favourable general assessment of DHAs and their potential use, GPs still show limited willingness to use these applications extensively and consistently in patient care. This is mainly justified with a little overview of mHealth tools and their possible applications. The interviewees expressed a need for neutral research sources focussed on health apps. Regardless of the potential areas of application for DHAs, interviewees reported criticism on lacking documentation of efficaciousness and liability issues, especially regarding diagnostic and therapeutic DHAs. A sizeable number of respondents saw a specific problem in the fast-track procedure. To promote the use of DHAs in the GP setting, respondents expressed the desire to see a wide range of professional CME-certified training programmes. With regard to further optimisation, physicians with experience in DHAs emphasised reinforcing motivation-boosting usability with an intuitive user interface. ## Comparison with existing literature Health apps can benefit diseases such as obesity and diabetes mellitus type 2 by documenting symptoms and encouraging changes in lifestyle, such as diet and exercise [3–8]. Studies have shown GPs to see potential benefits in health apps but have so far been reluctant to integrate mHealth tools into patient care due to concerns about safety issues and reliability as well as implementation in everyday clinical practice [12,17–20,23,24]. This comes along with great uncertainty in selecting suitable apps from a dynamic application market. The DVG law aims to create a basis for implementing DHAs in healthcare using clear-cut quality standards [1,7]. So far, Germany is the only European country in which state-approved health apps can be included in the reimbursement of the health system under certain conditions. It can be assumed that other countries might use the implementation of DHAs in Germany as a basis or role model for their own decisions [1,10]. Compared to previous surveys, our interviews showed the image and acceptance of DHAs to be noticeably more pronounced than in ordinary health apps among the GPs interviewed in the present study [17,19,20]. Respondents showed greater overall confidence in DHAs as solid, relatively safe, and potentially effective applications due to the necessary examination for inclusion in the DHA directory and legal framework – a finding also hinted at in a Barmer health insurance survey [25]. This promises favourable conditions for implementation in primary care. Studies have pointed out that a limiting factor in the willingness of GPs to use digital tools and mHealth apps in patient care is related to their assessment of competence about the application of such tools [14–16,19]. Also, the present study has shown that this self-assessment can be demonstrated based on two aspects. First, lack of experience with mHealth programs means that most respondents require confidence in their ability to guide and support patients in using DHAs [18]. Second, reliable information sources are needed [14]. Cementing DHAs in general practice across the board would require informing GPs on the fundamentals of the law (DVG), while addressing concerns and requests. Several studies have shown that GPs are dissatisfied with the transparency and reliability of information sources currently available to support patients with prescription or freely available health apps [12,16,17,19,20,25]. Not only in this study, German GPs suggested the national health portal as a possible information platform with this specific focus [21,23]. Professional organisations and their official publications could provide support with their own information services and discussion on healthcare outcomes from using DHAs. Also an authority is needed to provide an overview of which applications are suitable for which area and what needs consideration when using them. The present study found that physicians’ associations could play a key role in providing and sharing information on mHealth topics. A discussion subject is the fast-track procedure for approving new DHAs criticised by the GPs surveyed in this study. This procedure allows provisional inclusion of apps in the directory without evidence of beneficial healthcare effects. An assessment on healthcare system digitalisation from the German Council of Experts highlighted the essential nature of careful evaluation of an app’s effectiveness and benefit in the audit procedure [26]. However, the short development cycles in DHAs pose a challenge compared to the lengthy periods applicable to established study designs [3]. The benefit assessment and coverage process should, therefore, be planned in such a way as to ensure that the safest possible applications with high quality and proven benefit enter the healthcare system while also providing an incentive for suppliers to invest in developing these applications. Expert reports on the healthcare system digitisation encourage a wide range of professional and comprehensive training programmes to familiarise GPs with opportunities and conditions involved in integrating DHAs into patient care [22,27,28]. To facilitate healthcare professionals implementing an eHealth intervention, Versluis et al. have provided a practical worksheet to target expected or experienced barriers effectively [23]. Houwink et al. advise incorporating of eHealth education into vocational training and CPD activities [28]. It would also be important for statutory health insurance organisations to advise patients consistently and proactively on using DHAs rather than leaving this to the physicians alone [7, 10, 18]. Various authors explicitly advised expanding on gamification in DHAs: Prioritising motivation, such as by integrating gamification in an intuitive user experience, could make more of a success at initiating lifestyle changes and keeping them in the long term [29,30]. ## Strengths and limitations The study was based on preliminary studies, aligning closely with the GP perspective. The sample comprised GPs organised in specific physicians’ associations and interested in digital technology. Due to the convenience sample, selection bias needs to be considered in all interpretations. This means that attitudes or critical statements of the target group included do not necessarily have to be congruent with GPs not connected to physician networks. In addition, it should be borne in mind that physicians’ associations are prevalent in urban regions, where care structures and conditions for doctors and patients differ from those in rural areas. Another critical point worth considering is that some participants in the sample were interviewed by telephone rather than a face-to-face interview. ## Conclusion GPs with experience in health apps rated approved digital health applications (DHAs) favourably regarding healthcare potential and as safer and more reliable than conventional health apps. Many saw benefits to healthcare from using digital health applications. The physicians also saw room for further improvement, especially about usability and gamification as well as extra training programmes and trustworthy information sources for these applications. Interviewees saw the inclusion of DHAs in evidence-based guidelines as a significant step forward. ## Disclosure statement The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper. ## Data availability statement Research data are available upon request. ## References 1. Brönneke JB, Debatin JF, Hagen J. DiGA vademecum. Was man zu digitalen gesundheitsanwendungen wissen muss [DiGA vademecum. What you need to know about digital health applications] (2020) 2. Stern AD, Brönneke J, Debatin JF. Advancing digital health applications: priorities for innovation in real-world evidence generation. Lancet Digit Health (2022) 4 e200-e206. PMID: 35216754 3. Albrecht U-V. (2017) 4. König IR, Mittermaier M, Sina C. Evidence of positive care effects by digital health apps—methodological challenges and approaches. Inn Med (2022) 63 1298-1306 5. Dahlhausen F, Zinner M, Bieske L. Physicians’ attitudes toward prescribable mHealth apps and implications for adoption in Germany: mixed methods study. JMIR Mhealth Uhealth (2021) 9 e33012. PMID: 34817385 6. Ernsting C, Stuhmann LM, Dombrowski SU. Associations of health app use and perceived effectiveness in people with cardiovascular diseases and diabetes: population-based survey. JMIR Mhealth Uhealth (2019) 7 e12179. PMID: 30920383 7. Gerlinger G, Mangiapane N, Sander J.. Digital health applications (DiGA) in medical and psychotherapeutic care. Opportunities and challenges from the perspective of the healthcare providers. Bundesgesundheitsbl (2021) 64 1213-1219 8. Richter JG, Nannen C, Chehab G. Mobile app-based documentation of patient-reported outcomes – 3-months results from a proof-of-concept study on modern rheumatology patient management. Arthritis Res Ther (2021) 23 121. PMID: 33874994 9. Shaw T, McGregor D, Brunner M. What is eHealth? Development of a conceptual model for eHealth: qualitative study with key informants. J Med Internet Res (2017) 19 e324. PMID: 29066429 10. Kuhn E, Rogge A, Schreyer K. Apps on prescription in the medical office, but how? A case-based problem outline of medical-ethical implications of DHA usage. Gesundheitswesen (2022) 84 696-700. PMID: 33957698 11. Mittermaier M, Sina C, Richter JG. Practical use of digital health applications (DiGA) in internal medicine. Internist (2022) 63 245-254. PMID: 35037948 12. Hickey E, McMillan B, Mitchell C.. Practitioners should embrace, not ignore, health apps. BMJ (2015) 350 h2336. PMID: 25953643 13. Leigh S, Ashall-Payne L.. The role of healthcare providers in mHealth adoption. Lancet Digit Health (2019) 1 e58-e59. PMID: 33323231 14. Brandt CJ, Søgaard GI, Clemensen J. General practitioners’ perspective on eHealth and lifestyle change: qualitative interview study. JMIR Mhealth Uhealth (2018) 6 e88. PMID: 29666045 15. Nguyen AD, Frensham LJ, Baysari MT. Patients’ use of mobile health applications: what general practitioners think. Fam Pract (2019) 36 214-218. PMID: 29873708 16. Byambasuren O, Beller E, Glasziou P.. Current knowledge and adoption of mobile health apps among Australian general practitioners: survey study. JMIR Mhealth Uhealth (2019) 7 e13199. PMID: 31199343 17. Wangler J, Jansky M.. Potentials of health apps in primary care – a survey among general practitioners. Präv Gesundheitsf (2021) 16 150-156 18. Peeters JM, Krijgsman JW, Brabers AE. Use and uptake of eHealth in general practice: a cross-sectional survey and focus group study among health care users and general practitioners. JMIR Med Inform (2016) 4 e11. PMID: 27052805 19. Wangler J, Jansky M.. Health apps as instruments of prevention? – a qualitative study on the potential for the primary care setting. Präv Gesundheitsf (2020) 15 340-346 20. Wangler J, Jansky M.. Significance and application potential of health apps in diabetological care – results of a survey study. Diabetologie und Stoffwechsel (2023) 18 69-77 21. Byambasuren O, Beller E, Hoffmann T. Barriers to and facilitators of the prescription of mhealth apps in Australian general practice: qualitative study. JMIR Mhealth Uhealth (2020) 8 e17447. PMID: 32729839 22. Dahlhausen F, Zinner M, Bieske L. There’s an app for that, but nobody’s using it: insights on improving patient access and adherence to digital therapeutics in Germany. Digital Health (2022) 8 205520762211046 23. Versluis A, Van Luenen S, Meijer E. SERIES: eHealth in primary care. Part 4: addressing the challenges of implementation. Eur J Gen Pract (2020) 26 140-145. PMID: 33025820 24. de Wilt T, Versluis A, Goedhart A. General practitioners attitude towards the use of eHealth and online testing in primary care. Clin eHealth (2020) 3 16-22 25. 25Barmer. BARMER-Umfrage zu Gesundheits-Apps – Ärzte stehen digitalen Helfern offen gegenüber [BARMER survey on health apps - doctors are open to digital helpers]; 2020 [cited 2022 Nov 26]. Available from: https://www.barmer.de/presse/presseinformationen/pressemitteilungen/barmer-umfrage-zu-gesundheits-apps–-aerzte-stehen-digitalen-helfern-offen-gegenueber-247444. (2020) 26. 26Sachverständigenrat zur Begutachtung der Entwicklung im Gesundheitswesen [Advisory Council for the Assessment of Developments in the Health Care System]. Digitalisierung für Gesundheit. Ziele und Rahmenbedingungen eines dynamisch lernenden Gesundheitssystems [Digitization for health. Goals and framework conditions of a dynamically learning health system]; 2021 [cited 2022 Nov 26]. Available from: https://www.degam.de/files/Inhalte/Degam-Inhalte/Aktuelles/2021/SVR_Gutachten_2021_online.pdf. (2022) 27. Dufour JC, Grosjean J, Darmoni S. ApiAppS: a project to study and help practitioners in recommending mHealth apps and devices to their patients. Stud Health Technol Inform (2019) 264 1919-1920. PMID: 31438407 28. Houwink EJF, Kasteleyn MJ, Alpay L. SERIES: eHealth in primary care. Part 3: eHealth education in primary care. Eur J Gen Pract (2020) 26 108-118. PMID: 32757859 29. Krisam M, Prieß N., Baas J. Gesundheit im Zeitalter der Plattform-Ökonomie [Health in the age of the platform economy] (2022) 146-155 30. Harst L, Lantzsch H, Scheibe M.. Theories predicting end-user acceptance of telemedicine use: systematic review. J Med Internet Res (2019) 21 e13117. PMID: 31115340
--- title: Impact of Obesity on In-Hospital Morbidity and Mortality Among Patients Admitted for Acute Exacerbations of Chronic Obstructive Pulmonary Disease (COPD) journal: Cureus year: 2023 pmcid: PMC10026755 doi: 10.7759/cureus.35138 license: CC BY 3.0 --- # Impact of Obesity on In-Hospital Morbidity and Mortality Among Patients Admitted for Acute Exacerbations of Chronic Obstructive Pulmonary Disease (COPD) ## Abstract Background Obesity has been considered to be a risk factor for increased morbidity and mortality among patients with cardiopulmonary diseases. The burden of chronic obstructive pulmonary disease (COPD) and obesity is very high in the United States. We aimed to use the National Inpatient Sample (NIS) to evaluate the impact of obesity on the outcomes of patients hospitalized with COPD exacerbation. Materials & Methods This is a retrospective cohort study from the NIS database involving adult patients hospitalized for COPD exacerbation in the year 2019 obtained using the international classification of diseases, 10th revision coding system (ICD-10). Obese and morbidly obese subgroups were identified. Statistical analyses were done using the Stata software, and regression analysis was performed to calculate odds ratios. Adjusted odds ratios (aOR) were calculated after adjusting for potential confounders. Results Among patients hospitalized for COPD exacerbations, mortality rates were lower among obese and morbidly obese patients; aOR 0.72 [0.65, 0.80] and aOR 0.88 [0.77-0.99], respectively. Obese and morbidly obese were more likely to require non-invasive ventilation aOR 1.63 [1.55, 1.7] and aOR 1.93 [1.85-2.05], respectively, and were more likely to require mechanical ventilation aOR 1.25 [1.19, 1.31], and aOR 1.53 [1.44-1.62], respectively. The tracheostomy rate was $1.17\%$, $0.83\%$, and $0.38\%$ among patients with morbid obesity, obesity, and nonobese patients, respectively. Obese (aOR 1.11 [1.07-1.14]) and morbidly obese patients (aOR 1.21 [1.16-1.26]) had higher odds of being discharged on home oxygen and to a skilled nursing facility (SNF), aOR 1.32[1.27-1.38] and aOR 1.37 [1.3-1.43], respectively. Average hospital charges and length of hospitalization were significantly higher for morbidly obese and obese patients as compared to non-obese patients ($p \leq 0.01$). Conclusions Among admissions for COPD exacerbation, the rates of non-invasive ventilation, mechanical ventilation, tracheostomy, discharge with supplemental oxygen, length of hospitalization, hospitalization charges, and discharge to an SNF were higher among obese patients representing a higher morbidity and healthcare utilization in this group. This, however, did not translate into increased mortality among obese patients admitted with COPD exacerbations, and further randomized controlled trials are required to confirm our findings. ## Introduction Obesity is a growing cause of concern in the United States (US) in the past few decades. The annual healthcare cost attributable to obesity exceeds $700 billion each year worldwide, with an annual economic burden of about $100 billion in the US [1]. Obesity affects outcomes in several cardiopulmonary disease processes, especially among critically ill patients [2-4]. Obesity has a significant impact on the pulmonary system, as it is directly involved in the pathophysiology of obesity hypoventilation syndrome and obstructive sleep apnea [5,6]. Chronic obstructive pulmonary disease (COPD) is a leading cause of death, with a significant financial burden on healthcare resources in the United States [7]. Up to $65\%$ of the COPD population is overweight (Body mass index (BMI) 25-29.99) or obese (BMI>30) [8]. Although the impact of obesity on all-cause mortality and morbidity cardiovascular diseases is established, a protective association has been noted in the effect of obesity on COPD patients [9]. The impact of obesity on respiratory pathophysiology and symptom intensity in patients with COPD is not well known. It is hypothesized that patients with COPD who are obese may have some favorable lung mechanics like limited hyperinflation of the lungs, and this may be, in part, the reason that a negative association has been noted between obesity and mortality among them [10]. In addition to mortality, COPD is also associated with morbidity in the form of hospitalization for non-invasive ventilation, oxygen dependence, the requirement of intubation/tracheostomy, and nursing home stay. Our study aims to explore the impact of obesity on the disease severity, in-hospital outcomes, morbidity, and mortality of patients admitted for acute exacerbation of COPD (AECOPD). ## Materials and methods Design and data source This study is a retrospective cohort study involving adult patients (aged>18) hospitalized for exacerbation of chronic obstructive pulmonary disease (COPD) in the US between January 1, 2019, and December 31, 2019. Data were obtained from the Nationwide Inpatient Sample (NIS) database for 2019. The NIS is the largest publicly available database, covering more than $98\%$ of the US population and approximately 1,000 hospitals. It approximates discharges from US community hospitals, excluding rehabilitation and long-term acute hospitals. The database uses the international classification of diseases, 10th revision, and clinical modification/procedure coding system (ICD-10-CM/PCS) [11]. Study population This study included adult patients with a primary discharge diagnosis of acute exacerbation of COPD or a primary diagnosis of acute respiratory failure with a secondary diagnosis of COPD. Obese patients were identified based on the presence of a secondary discharge diagnosis of obesity. Patients were excluded if their age was less than 18 years. Sub-group analysis was done on morbidly obese patients (body mass index >40). Patient characteristics and comorbidities in all groups were recorded. Outcome measures The primary outcome was comparing the mortality; and the incidence of non-invasive (NIV) and mechanical ventilation (MV) or endotracheal intubation as a measure of clinical severity among patients admitted for COPD exacerbation depending on the presence or absence of obesity. Secondary outcomes included length of stay and hospital charges accrued as a measure of healthcare utilization cost among both groups, supplemental oxygen on discharge, incidence of tracheostomy and discharge to a skilled nursing facility (SNF) as a measure of morbidity status, and the incidence of cardiac arrest and septic shock as a measure of complications during the hospital stay. Statistical analysis Data were analyzed using Stata® (Statistics and Data) Version 17 BE software (StataCorp, Texas, USA). All analyses were conducted using the weighted samples for national estimates in adjunct to Healthcare Cost and Utilization Project (HCUP) regulations for using the NIS database. Co-morbidities were calculated as proportions of the cohort, and the Chi-squared test was used to compare the association between the non-obese and the obese subgroups. Multivariate regression analysis was done to adjust for possible confounders while calculating the primary and secondary outcomes. The patient's comorbidities were obtained during the literature review. A univariate screen was done to confirm these factors further. A p-value of 0.05 was set as the threshold for statistical significance in the regression analysis. Odds ratios were calculated for all outcomes and were adjusted for age, sex, race, insurance status, and Charleston comorbidity index (CCI). Ethical considerations The NIS database does not contain any patient identifiers. Since 2012, the NIS has also removed state-level and hospital identifiers. This is in accordance with HIPAA regulations and respects patient protection and anonymity. All NIS-based studies are thus exempt from institutional review board approval. ## Results Baseline characteristics of the study population A total of 674,080 weighted hospitalizations for COPD exacerbation were included in the analysis. Of these, $19.6\%$ ($$n = 132$$,120) were obese, and $80.3\%$ ($$n = 541$$,959) were non-obese. The mean age of obese and non-obese patients was 64.33 and 69.10 years, respectively. $62.91\%$ of the obese and $56.07\%$ of the non-obese patients were females. Caucasians (77.73 vs. $73.45\%$) and Pacific Islanders ($1.29\%$ vs. $0.66\%$) were more prevalent in the non-obese group, whereas African Americans (13.95 vs. $17.87\%$) and Hispanics ($4.73\%$ vs. $5.74\%$) tended to be more prevalent in the obese group. Type 2 diabetes mellitus, chronic kidney disease, atrial fibrillation, and pulmonary hypertension were significantly more prevalent among obese patients, while hypertension and smoking were more prevalent in the non-obese group. Among the obese group, $40.65\%$ ($$n = 74485$$) were morbidly obese and made up the morbidly obese subgroup. The detailed baseline characteristics for the included patients are summarized in Table 1. Table 1 \| Variable \| Non-obese (%) \| Obese (%) \| P-value \| \| --- \| --- \| --- \| --- \| \| Total \| 80.3% (N=541959) \| 19.6 (N=132120) \| \| \| Female \| 56.07 \| 62.91 \| <0.0001 \| \| Age (mean) \| 69.1 years \| 64.33 years \| <0.0001 \| \| Race \| Race \| Race \| <0.0001 \| \| Caucasians \| 77.73 \| 73.45 \| \| \| African americans \| 13.95 \| 17.87 \| \| \| Hispanics \| 4.73 \| 5.74 \| \| \| Pacific Islanders \| 1.29 \| 0.66 \| \| \| Native americans \| 0.56 \| 0.59 \| \| \| Others \| 1.74 \| 1.69 \| \| \| Charleston comorbity index (CCI) \| Charleston comorbity index (CCI) \| Charleston comorbity index (CCI) \| <0.0001 \| \| CCI=1 \| 33.52 \| 17.88 \| \| \| CCI=2 \| 24.1 \| 25.72 \| \| \| CCI > or = 3 \| 42.39 \| 56.4 \| \| \| Income quartile (median household income of the patients ZIP Code) \| Income quartile (median household income of the patients ZIP Code) \| Income quartile (median household income of the patients ZIP Code) \| <0.0001 \| \| 1-47,999$ \| 38 \| 39.12 \| \| \| 48,000-60,999$ \| 27.73 \| 28.13 \| \| \| 61,000-81,999$ \| 20.97 \| 21.02 \| \| \| 82,000$ + \| 13.3 \| 11.74 \| \| \| Insurance \| Insurance \| Insurance \| <0.0001 \| \| Medicare \| 72.23 \| 65.51 \| \| \| Medicaid \| 13.67 \| 18.88 \| \| \| Private insurance \| 11.39 \| 12.71 \| \| \| Self pay \| 2.71 \| 2.9 \| \| \| Type 2 Diabetes mellitus \| 11.5 \| 19.12 \| <0.0001 \| \| Essential Hypertension \| 37.92 \| 32.2 \| <0.0001 \| \| End stage renal disease \| 23.5 \| 22.3 \| 0.2527 \| \| Chronic kidney disease \| 13.23 \| 17.93 \| <0.0001 \| \| Atrial Fibrillation \| 19.89 \| 22.22 \| <0.0001 \| \| Smoking \| 42.23 \| 40.32 \| <0.0001 \| \| Pulmonary Hypertension \| 1.58 \| 2.8 \| <0.0001 \| Clinical and hospital-related outcomes in non-obese, obese, and morbidly obese patients admitted for COPD exacerbation. Among patients hospitalized for COPD exacerbations, mortality rates were lower among obese patients ($2.1\%$ vs. $2.87\%$) with an adjusted odds ratio of 0.72 [0.65, 0.80], $p \leq 0.0001.$ Mortality rates were also lower (aOR, 0.88 [0.77-0.99], $$p \leq 0.03$$) among patients with morbid obesity (BMI > 40 kg/m2) when compared to those with a BMI < 40 kg/m2. Non-invasive ventilation was required in $12.16\%$ of non-obese and $19.7\%$ of obese patients. Mechanical ventilation was required in $7.5\%$ of non-obese and $11.27\%$ of obese patients. Obese patients were more likely to require non-invasive ventilation (aOR 1.63 [1.55, 1.7], $p \leq 0.0001$) and mechanical ventilation (aOR 1.25 [1.19, 1.31], $p \leq 0.0001$). The odds of requiring NIV ($23.3\%$) and MV ($13.85\%$) were even higher among morbidly obese patients at aOR 1.95 [1.85-2.05], $p \leq 0.0001$, and aOR 1.53 [1.44-1.62], $p \leq 0.0001$, respectively. The primary outcomes, like mortality and the incidence of mechanical and non-invasive ventilation in the obese and morbidly obese subgroups, are summarized in Table 2. Table 2 \| Variable \| Non obese (%) \| Obese (%) \| Odds Ratio (OR) \| p-value \| Adjusted OR \| p-value.1 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Mortality \| 2.87 (N= 15554) \| 2.1 (N=775) \| 0.726 \| <0.0001 \| 0.72 [0.65, 0.80] \| <0.0001 \| \| Non Invasive ventilation \| 12.16 \| 19.7 \| 1.77 \| <0.0001 \| 1.63 [1.55,1.7] \| <0.0001 \| \| Mechanical ventilation \| 7.52 \| 11.27 \| 1.56 \| <0.0001 \| 1.25 [1.19, 1.31] \| <0.0001 \| The primary outcomes, like mortality and the incidence of mechanical and non-invasive ventilation in the morbidly obese subgroup, are summarized in Table 3. Table 3 \| Variable \| BMI<40 (%) \| Morbid obesity (%) \| Odds Ratio (OR) \| p-value \| Adjusted OR \| p-value.1 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Mortality \| 2.87 (N= 15554) \| 2.34 (N=1662) \| 0.844 \| 0.004 \| 0.88 [0.77, 0.99] \| 0.043 \| \| Non Invasive Ventilation \| 12.48 \| 23.3 \| 2.13 \| <0.0001 \| 1.95 [1.85,2.05] \| <0.0001 \| \| Mechanical ventilation \| 7.59 \| 13.85 \| 1.95 \| <0.0001 \| 1.53 [1.44, 1.62] \| <0.0001 \| The rate of tracheostomy was $1.17\%$, $0.83\%$, and $0.38\%$ among patients with morbid obesity, obesity, and BMI < 30 mg/kg with higher odds among obese (aOR 1.7 [1.42-2.04], $p \leq 0.0001$) and morbidly obese patients (aOR 2.25 [1.85-2.75], $p \leq 0.0001$). Obese and morbidly obese COPD patients were more likely to be discharged to a skilled nursing facility, with aOR 1.32[1.27-1.38] and aOR 1.37 [1.3-1.43], respectively. The odds of developing a cardiac arrest among obese vs. non-obese patients (aOR 0.89, [0.78, 1.02] $$p \leq 0.09$$) were comparable. There was no increased risk of cardiac arrest among morbidly obese patients as well (aOR 1.05, [0.89, 1.24], $$p \leq 0.5$$). The odds of developing septic shock were comparable among non-obese, obese ($$p \leq 0.14$$), and morbidly obese patients ($$p \leq 0.05$$). Obese (aOR 1.11 [1.07-1.14], $p \leq 0.0001$) and morbidly obese patients (aOR 1.21 [1.16-1.26], $p \leq 0.0001$) had higher odds of being discharged on home oxygen as compared to non-obese patients. Average hospital charges were highest for morbidly obese patients ($59776), high in obese patients ($54849), and lowest for non-obese patients ($44958), $p \leq 0.0001.$ Similarly, the mean length of stay was highest among morbidly obese patients (5.61 days), high among obese patients (5.22 days), and lowest among non-obese patients (4.51 days), and this difference was statistically significant ($p \leq 0.0001$). The secondary outcomes like the rate of tracheostomy, the incidence of adverse events like cardiac arrest and septic shock, discharge with supplemental oxygen, discharge to a skilled nursing facility, average hospital cost, and length of stay in the obese and morbidly obese subgroup are summarized in Table 4. Table 4 \| Variable \| Non obese \| Obese \| Odds Ratio (OR) \| p-value \| Adjusted OR \| p-value.1 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Tracheostomy \| 0.38 \| 0.83 \| 2.2 \| <0.0001 \| 1.7 [1.42, 2.04] \| <0.0001 \| \| Supplemental oxygen on discharge \| 30.53 \| 32.32 \| 1.08 [1.05, 1.12] \| <0.0001 \| 1.11 [1.07, 1.14] \| <0.0001 \| \| Skilled Nursing Facility \| 14.51 \| 16.47 \| 1.15 [1.1, 1.2] \| <0.0001 \| 1.32 [1.27, 1.38] \| <0.0001 \| \| Cardiac arrest \| 1.1 \| 1.19 \| 1.08 \| 0.2 \| 0.89 [0.78, 1.02] \| 0.093 \| \| Septic shock \| 0.55 \| 0.58 \| 1.07[0.9,1.27] \| 0.44 \| 0.87 [0.71, 1.04] \| 0.14 \| \| Length of stay \| 4.51 days \| 5.22 days \| \| \| \| <0.0001 \| \| Charges \| $44958 \| $54849 \| \| \| \| <0.0001 \| The secondary outcomes in the morbidly obese subgroup are summarized in Table 5. Table 5 \| Variable \| BMI<40 \| Morbid obesity \| Odds Ratio (OR) \| p-value \| Adjusted OR \| p-value.1 \| \| --- \| --- \| --- \| --- \| --- \| --- \| --- \| \| Tracheostomy \| 0.38 \| 1.16 \| 3.05 \| <0.0001 \| 2.25 [1.85, 2.75] \| <0.0001 \| \| Supplemental oxygen on discharge \| 30.51 \| 33.98 \| 1.17 [1.13, 1.21] \| <0.0001 \| 1.21 [1.16, 1.26] \| <0.0001 \| \| Skilled nursing facility \| 14.5 \| 18.86 \| 1.37 [1.3, 1.43] \| <0.0001 \| 1.73 [1.64, 1.82] \| <0.0001 \| \| Cardiac arrest \| 10.8 \| 13.8 \| 1.27 [1.09,1.48] \| 0.001 \| 1.05 [0.89, 1.24] \| 0.5 \| \| Septic shock \| 0.54 \| 0.66 \| 1.21 [0.98, 1.5] \| 0.066 \| 0.95 [0.75, 1.19] \| 0.65 \| \| Length of stay \| 4.54 days \| 5.61 days \| \| \| \| <0.0001 \| \| Charges \| $44958 \| $59776 \| \| \| \| <0.0001 \| ## Discussion Obesity is a worldwide concern and has a significant impact on the morbidity and mortality associated with most chronic illnesses, including COPD [12,13]. This study analyzed the largest available US clinical registry from the year 2019, which included over 670,000 COPD-related hospitalizations. The key findings from our contemporary analysis of the NIS are as follows 1) Among COPD-related hospitalizations, obese patients had lower odds of mortality as compared to non-obese patients, 2) Odds of requiring NIV, MV, Tracheostomy were higher among obese patients as compared to non-obese patients admitted for COPD exacerbations. 3) Obese and morbidly obese patients had higher LOS and hospital charges when admitted for COPD exacerbations. 4) Odds of undergoing tracheostomy or requiring supplemental oxygen at discharge were higher among obese patients admitted for AECOPD. We observed that obese COPD patients were younger than the non-obese patients suggesting an earlier onset of morbidity among this group. Females were more prevalent in the obese group, which is consistent with the higher prevalence of obesity in females [14]. We found that mortality rates were lower in obese and morbidly obese patients as compared to nonobese patients. The results of our study were consistent with what has been reported before [10,15]. There have been reports of obese or overweight individuals having favorable survival outcomes. This was hypothesized as being secondary to better-preserved lung function, muscle mass, and exercise tolerance and not from fat accumulation [16]. Another hypothesis states that advanced COPD itself may lead to weight loss among obese patients; thus, patients are thought to be ‘non-obese’ at the time of death. Thus COPD-related mortality may appear more in patients with a normal BMI [9]. The obesity paradox is well documented, as seen in a large study that looks at 180,000 admissions for COPD across multiple states [17-20]. In our study, we assessed in-hospital morbidity represented by the need for mechanical ventilation and non-invasive ventilation. We observed that the odds of requiring both MV and NIV were significantly higher in the obese and morbid groups. Obese individuals had almost twice the odds of needing intubation and mechanical ventilation. NIV has also been increasingly reported to be responsible for favorable outcomes in all COPD patients regardless of BMI [21,22]. The higher rates of NIV and MV in COPD among obese individuals could be partly attributed to the co-existence of other pulmonary comorbidities, such as obstructive sleep apnea or obesity hypoventilation syndrome [23]. The average length of stay was higher among the obese and morbidly obese patients, which is congruent with the increased use of non-invasive and invasive ventilation in these groups. The impact of NIV and mechanical ventilation on a longer length of stay has been reported before [24,25]. This reflects higher morbidity when compared with non-obese individuals. Costs of care increased as the severity of obesity increased among these patients, with morbidly obese patients having the most amount of burden on healthcare utilization and non-obese patients having the least. Clinical adverse events such as cardiac arrest and septic shock were similar among obese and non-obese patients. The incidence of tracheostomy was higher in the obese group and almost twice as higher in the morbidly obese group, which suggests a presumed longer duration of recovery and morbidity in these groups [26]. Obesity was found to have an increased tendency to require supplemental oxygen, also suggesting a higher morbidity status and disease severity in this group [27,28]. Our study focuses primarily on investigating the effects of obesity on the morbidity of COPD and is the first of its kind to our knowledge. The use of NIV and MV are indicators of in-hospital morbidity, and length of stay and hospital charges represent healthcare resource utilization. The incidence of tracheostomy and discharge with supplemental oxygen are parameters of post-discharge morbidity, which suggests that such patients if discharged to rehabilitation centers or nursing homes, could further stretch out health care resources. Our study has several strengths; the NIS is one of the largest databases, and thus, the statistical analyses have high power. We acknowledge that our study also has some inherent limitations. The ICD 10 codes are primarily for billing purposes, so the sensitivity of the clinical information is limited. The parameters of obesity have been established using the secondary diagnosis of obesity; hence, it is difficult to ascertain the degree of obesity within the range. NIS data records hospitalizations and not individuals; hence, an individual with multiple admissions can be counted as multiple encounters. The data does not represent a linear timeline, and the secondary diagnoses could precede or follow the primary reason for hospitalization. NIS studies can only establish association without any comment on causation. We did not include the underweight category. ## Conclusions The presence and degree of obesity affect morbidity, as evidenced by higher rates of NIV, mechanical ventilation, tracheostomy, and discharge with supplemental oxygen among obese individuals admitted for COPD exacerbation. While obesity has been implicated in worse outcomes in many disease processes, it has not been associated with higher mortality in COPD patients. Obese COPD patients have a higher length of stay and hospital charges that suggest that they have a greater burden on the health care system. ## References 1. Panuganti KK, Nguyen M, Kshirsagar RK. Obesity 2. Velazquez G, Gomez TM, Asemota I, Akuna E, Ojemolon PE, Eseaton P. Obesity impacts mortality and rate of revascularizations among patients with acute myocardial infarction: An analysis of the national inpatient sample. Cureus (2020.0) 12 0 3. Bapat SP, Whitty C, Mowery CT. Obesity alters pathology and treatment response in inflammatory disease. Nature (2022.0) 604 337-342. PMID: 35355021 4. Yaegashi M, Jean R, Zuriqat M, Noack S, Homel P. Outcome of morbid obesity in the intensive care unit. J Intensive Care Med (2005.0) 20 147-154. PMID: 15888902 5. Pattipati M, Gudavalli G, Dhulipalla L. The influence of obesity hypoventilation syndrome on the outcomes of patients with diabetic ketoacidosis. Cureus (2022.0) 14 0 6. Locke BW, Lee JJ, Sundar KM. OSA and chronic respiratory disease: Mechanisms and epidemiology. Int J Environ Res Public Health (2022.0) 19 7. Duffy SP, Criner GJ. Chronic obstructive pulmonary disease: Evaluation and management. Med Clin North Am (2019.0) 103 453-461. PMID: 30955513 8. García-Rio F, Soriano JB, Miravitlles M. Impact of obesity on the clinical profile of a population-based sample with chronic obstructive pulmonary disease. PLoS One (2014.0) 9 0 9. Spelta F, Fratta Pasini AM, Cazzoletti L, Ferrari M. Body weight and mortality in COPD: focus on the obesity paradox. Eat Weight Disord (2018.0) 23 15-22. PMID: 29110280 10. Zewari S, Vos P, van den Elshout F, Dekhuijzen R, Heijdra Y. Obesity in COPD: Revealed and unrevealed issues. COPD (2017.0) 14 663-673. PMID: 29210599 11. Agency for Healthcare Research and Quality: Healthcare cost and utilization project (HCUP). (2023.0) 12. Burney PG, Patel J, Newson R, Minelli C, Naghavi M. Global and regional trends in COPD mortality, 1990-2010. Eur Respir J (2015.0) 45 1239-1247. PMID: 25837037 13. Engin A. The definition and prevalence of obesity and metabolic syndrome. Adv Exp Med Biol (2017.0) 960 1-17. PMID: 28585193 14. Clinical guidelines on the identification, evaluation, and treatment of overweight and obesity in adults--the evidence report. National institutes of health. Obes Res (1998.0) 6 Suppl 2 51-209 15. Chittal P, Babu AS, Lavie CJ. Obesity paradox: does fat alter outcomes in chronic obstructive pulmonary disease?. COPD (2015.0) 12 14-18. PMID: 24945083 16. Galesanu RG, Bernard S, Marquis K, Lacasse Y, Poirier P, Bourbeau J, Maltais F. Obesity in chronic obstructive pulmonary disease: is fatter really better?. Can Respir J (2014.0) 21 297-301. PMID: 24791257 17. Goto T, Hirayama A, Faridi MK, Camargo CA Jr, Hasegawa K. Obesity and severity of acute exacerbation of chronic obstructive pulmonary disease. Ann Am Thorac Soc (2018.0) 15 184-191. PMID: 29053337 18. Iyer AS, Dransfield MT. The "obesity paradox" in chronic obstructive pulmonary disease: Can it be resolved?. Ann Am Thorac Soc (2018.0) 15 158-159. PMID: 29388820 19. Franssen FM, O'Donnell DE, Goossens GH, Blaak EE, Schols AM. Obesity and the lung: 5 obesity and COPD. Thorax (2008.0) 63 1110-1117. PMID: 19020276 20. Brigham EP, Anderson JA, Brook RD. Challenging the obesity paradox: extreme obesity and COPD mortality in the SUMMIT trial. ERJ Open Res (2021.0) 7 21. Hess DR. Noninvasive ventilation for acute respiratory failure. Respir Care (2013.0) 58 950-972. PMID: 23709194 22. Chu CM, Piper A. Non-invasive ventilation: A glimpse into the future. Respirology (2019.0) 24 1140-1142. PMID: 31625248 23. Senaratna CV, Perret JL, Lodge CJ. Prevalence of obstructive sleep apnea in the general population: A systematic review. Sleep Med Rev (2017.0) 34 70-81. PMID: 27568340 24. Pokharel P, Lamichhane P, Pant P, Shrestha AB. Factors affecting length of hospital stay in chronic obstructive pulmonary disease patients in a tertiary hospital of Nepal: A retrospective cross-sectional study. Ann Med Surg (Lond) (2022.0) 80 104246. PMID: 36045760 25. Li M, Cheng K, Ku K, Li J, Hu H, Ung CO. Factors influencing the length of hospital stay among patients with chronic obstructive pulmonary disease (COPD) in Macao population: A retrospective study of inpatient health record. Int J Chron Obstruct Pulmon Dis (2021.0) 16 1677-1685. PMID: 34135579 26. Namin AW, Kinealy BP, Harding BC, Alnijoumi MM, Dooley LM. Tracheotomy outcomes in the medical intensive care unit. Mo Med (2021.0) 118 168-172. PMID: 33840862 27. Long term domiciliary oxygen therapy in chronic hypoxic cor pulmonale complicating chronic bronchitis and emphysema. Report of the Medical Research Council Working Party. Lancet (1981.0) 1 681-686. PMID: 6110912 28. Nocturnal Oxygen Therapy Trial Group. Continuous or nocturnal oxygen therapy in hypoxemic chronic obstructive lung disease: A clinical trial. Ann Intern Med (1980.0) 93 391-398. PMID: 6776858
--- title: 'Effectiveness of BNT162b2 and CoronaVac vaccines in preventing SARS-CoV-2 Omicron infections, hospitalizations, and severe complications in the pediatric population in Hong Kong: a case-control study' authors: - Vincent Ka Chun Yan - Franco Wing Tak Cheng - Celine Sze Ling Chui - Francisco Tsz Tsun Lai - Carlos King Ho Wong - Xue Li - Eric Yuk Fai Wan - Joshua Sung Chih Wong - Esther Wai Yin Chan - Ian Chi Kei Wong - Mike Yat Wah Kwan - Patrick Ip journal: Emerging Microbes & Infections year: 2023 pmcid: PMC10026771 doi: 10.1080/22221751.2023.2185455 license: CC BY 4.0 --- # Effectiveness of BNT162b2 and CoronaVac vaccines in preventing SARS-CoV-2 Omicron infections, hospitalizations, and severe complications in the pediatric population in Hong Kong: a case-control study ## ABSTRACT Severe COVID-19 appears to be disproportionately more common in children and adolescents since the emergence of Omicron. More evidence regarding vaccine effectiveness (VE) is urgently needed to assist policymakers in making decisions and minimize vaccine hesitancy among the public. This was a case-control study in the pediatric population using data extracted from the electronic health records database in Hong Kong. Individuals aged 3–17 with COVID-19 confirmed by polymerase chain reaction were included in the study. Each case was matched with up to 10 controls based on age, gender, and index date (within 3 calendar days). The VE of BNT162b2 and CoronaVac in preventing COVID-19, hospitalizations, and severe outcomes were estimated using conditional logistic regression adjusted by patients’ comorbidities and medication history during the outbreak from January to August 2022. A total of 36,434 COVID-19 cases, 2231 COVID-19-related hospitalizations, and 1918 severe COVID-19 cases were matched to 109,004, 21,788, and 18,823 controls, respectively. Compared to the unvaccinated group, three doses of BNT162b2 or CoronaVac was associated with reduced risk of infection [VE: BNT162b2: $56.0\%$ ($95\%$ CI: 49.6–61.6), CoronaVac: $39.4\%$ ($95\%$ CI: 25.6–50.6)], hospitalization [VE: BNT162b2: $58.9\%$ ($95\%$ CI: 36.1–73.6), CoronaVac: $51.7\%$ (11.6–73.6)], and severe outcomes [VE: BNT162b2: $60.2\%$ ($95\%$ CI: 33.7–76.1), CoronaVac: $42.2\%$ ($95\%$ CI: −6.2–68.6)]. Our findings showed that three doses of BNT162b2 or CoronaVac was effective in preventing COVID-19, hospitalizations, and severe outcomes among the pediatric population during Omicron-dominant pandemic, which was further enhanced after a booster dose. ## Key points In this territory-wide case–control study, three doses of BNT162b2 or CoronaVac were associated with reduced risk of infection [Vaccine Effectiveness (VE): BNT162b2: $56.0\%$, CoronaVac: $39.4\%$], hospitalization (VE: BNT162b2: $58.9\%$; CoronaVac: $51.7\%$), and severe outcomes (VE: BNT162b2: $60.2\%$; CoronaVac: $42.2\%$). ## Background In the pre-Delta pandemic, children and adolescents were less likely to experience severe COVID-19 [1]. On the contrary, the emergence of SARS-CoV-2 Omicron variant appears to disproportionately cause more severe symptoms in children and adolescents, including more hospital admissions, convulsion, croup, and multisystem inflammatory syndrome in children (MIS-C) [2,3]. According to the data from the COVID-19-Associated Hospitalization Surveillance Network (COVID-NET), the weekly COVID-19-associated hospitalization rates in younger patients aged 17 or below increased by 4-fold from 1.8 per 100,000 population during the Delta predominance period to 7.1 per 100,000 during Omicron predominance period compared to a 2.5-fold increase in adults from 15.5 to 38.4 per 100,000 [2,4]. Although the side effect profiles of COVID-19 vaccines are now becoming established based on extensive pharmacovigilance studies [5,6], more evidence regarding vaccine effectiveness (VE) is urgently needed, especially during the Omicron era, which will assist policymakers in making decisions and to minimize vaccine hesitancy among the public. Hong Kong Special Administrative Region (HKSAR), China, started a territory-wide vaccination programme in February 2021 using mRNA vaccine BNT162b2 (Comirnaty, BioNTech/Pfizer/Fosun) and inactivated vaccine CoronaVac (Sinovac Biotech HK Limited). The vaccination programme was extended to adolescents aged 12–15 starting in June 2021 and was further extended to children aged 5–11 in January 2022. The minimum age for the CoronaVac vaccination was further lowered to infants aged 3 years in February 2022 and aged 6 months in August 2022. The VE of mRNA vaccines has been demonstrated in children and adolescents in several trials conducted in the early phase of the Omicron pandemic when BA.1 and BA.2 were the dominant subvariants [7–10]. CoronaVac is one of only a few approved vaccines for the pediatric population and is widely used in developing countries. It has been demonstrated to have a VE of around $40\%$ against symptomatic COVID-19 and $60\%$ against hospitalization in children based on two observational studies in Brazil and Chile during the Omicron period. However, these studies did not compare the VE of mRNA and inactivated vaccines, and did not consider the effects of booster doses [11,12]. Therefore, we conducted a territory-wide case–control study to examine the effectiveness of BNT162b2 and CoronaVac vaccines in children and adolescents in Hong Kong during the current Omicron-dominated pandemic, which also considered the number of vaccine doses and their effectiveness on a range of clinical outcomes. ## Data sources Clinical data on COVID-19 vaccinations in the pediatric and adolescent population in Hong Kong were obtained from the electronic health records database of the Hospital Authority (HA), the vaccination records of the Department of Health (DH), and COVID-19-confirmed case records from the Centre of Health Protection (CHP). Anonymized unique patient identifiers were used to integrate these databases. The HA is a statutory administrative organization that manages all public inpatient services and most of the public outpatient services in Hong Kong. The electronic health records database contains data on patient demographics, diagnoses, prescriptions, and laboratory tests, which provides real-time information to support routine clinical management across all clinics and hospitals in the HA. The DH maintains a vaccination records database of all individuals in Hong Kong. The CHP maintains a database of all confirmed COVID-19 cases based on both mandatory and voluntary reporting of positive polymerase chain reaction (PCR) and rapid antigen test (RAT) results in Hong Kong. These population-based databases have been used in studies on the risk of adverse effects of COVID-19 vaccinations and in other COVID-19 pharmacovigilance studies [5,6,13–27]. ## Study design and population This is a case-control study conducted on children and adolescents aged 3–17 years in Hong Kong. The study period was from 1 January to 15 August, 2022. Individuals with an incident COVID-19-related outcome during the study period were identified as cases. Controls were selected from all other individuals without a COVID-19-related outcome who attended HA services during the study period. The index date was the date of COVID-19-related outcomes for cases and the attendance date for controls. Subjects in the control group who reported a positive RAT result on the online voluntary reporting platform were excluded. The matching procedures were conducted separately for each of the COVID-19-related outcomes. Each case was matched with up to 10 controls according to age, sex, index date (within 3 calendar days), and Charlson Comorbidity Index (0, 1-2, 3-4, ≥5) [28]. ## Definitions of vaccine exposure Two COVID-19 vaccines, BNT162b2 and CoronaVac, are provided by the Hong Kong government free of charge in its mass vaccination programme. The BNT162b2 vaccine was first made available to individuals aged ≥16 in March 2021, and was extended to adolescents aged 12–15 in June 2021 and to children aged 5–11 in February 2022. The CoronaVac vaccine was first made available to adolescents aged 12–17 in November 2021 and was extended to children aged 5–11 in January 2022. Details of the full rollout schedule is listed in Supplementary Table 1. Individuals have a choice of the first vaccine dose between BNT162b2 or CoronaVac, but are then restricted to the same vaccine for the second dose. For the booster vaccination, individuals have a choice of either vaccine. In this study, COVID-19 vaccination status was classified into eight mutually exclusive groups based on the type and number of vaccine doses administered: (i) 1 dose BNT162b2, (ii) 1 dose CoronaVac, (iii) 2 doses BNT162b2, (iv) 2 doses CoronaVac, (v) 3 doses (all BNT162b2), (vi) 3 doses (all CoronaVac), (vii) 3 doses (2 doses BNT162b2 and CoronaVac booster), and (viii) 3 doses (2 doses CoronaVac and BNT162b2 booster). Individuals who received a fourth dose or with incomplete vaccination records were excluded from the study. ## Definitions of COVID-19 and outcomes The outcomes investigated in this study were (i) COVID-19 diagnosis; (ii) COVID-19-related hospitalization within 28 days after COVID-19 diagnosis; and (iii) severe COVID-19 defined as any diagnosis of complications or requiring procedures (including ventilatory support) listed in Supplementary Table 2, prescription of tocilizumab, methylprednisolone or intravenous immunoglobulin G, admission to an intensive care unit (ICU), or death within 28 days after COVID-19 diagnosis. COVID-19 diagnosis was defined as a positive PCR result obtained from the CHP of the HKSAR government and/or HA databases. A positive PCR result is recognized as the gold-standard test for COVID-19 given its high specificity of >$99\%$ [29]. The Hong Kong government has implemented extensive PCR testing for SARS-CoV-2 in public hospitals and clinics for those presenting with COVID-like symptoms and for close contacts of confirmed cases. The government has also set up territory-wide community testing centres to screen asymptomatic individuals and provide regular testing for staff at high risk of exposure to SARS-CoV-2, such as those working in nursing homes. A sensitivity analysis was conducted where voluntarily reported positive RAT cases were also included in the definition for COVID-19. Information regarding all-cause mortality was extracted from the Hong Kong Deaths Registry, the official governmental registry covering all registered deaths in Hong Kong. ## Statistical analysis Conditional logistic regressions adjusted for pre-existing asthma, diabetes, epilepsy, and use of immunosuppressants within 90 days were applied to calculate the crude and adjusted odds ratio (OR) with $95\%$ confidence interval [30]. Vaccine effectiveness (VE) was estimated by (1 – adjusted OR) × $100\%$. Subgroup analyses were conducted and were stratified by age (3–11 and 12–17 years). To evaluate waning VE, additional analyses were conducted at five time-since-vaccination intervals (0–13, 14–60, 61–120, 121–180, and ≥180 days) after receiving each vaccine dose. For each time-since-vaccination interval, only eligible matched pairs, in which both cases and controls were either unvaccinated or fell within the specific time-since vaccination interval, were included to derive the corresponding estimates. Two sensitivity analyses were conducted. First, cases and controls who received their last vaccine dose for more than 90 days were excluded. Second, both PCR and RAT positive cases were recognized in the definition of COVID-19. All statistical tests were two-sided and P values less than 0.05 were considered statistically significant. Statistical analyses were conducted using R version 4.0.3 (www.R-project.org). Two investigators (VY and FC) conducted the statistical analyses independently for quality assurance. STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) statement checklists were followed to guide the transparent reporting in this case–control study. ## Ethical approval This study was approved by the Central Institutional Review Board of the Hospital Authority of Hong Kong (CIRB-2021-005-4) and the Department of Health Ethics Committee (LM$\frac{171}{2021}$). ## Results From 1 January to 15 August 2022, a total of 36,434 COVID-19 cases, 2231 COVID-19-related hospitalizations, and 1918 severe COVID-19 cases were matched to 109,004, 21,788 and 18,823 controls, respectively (Figure 1). The baseline characteristics of cases and controls are summarized in Table 1. During the study period, eight COVID-19-related deaths and 99 cases of COVID-19-related ICU admission or ventilatory support were identified, which are summarized in Supplementary Table 3. Among the eight COVID-19-related deaths (mean [SD] age of 8.75 [5.06] years and $50\%$ male), six ($75\%$) were unvaccinated, one ($12.5\%$) was vaccinated with one dose of BNT162b2, and one ($12.5\%$) was vaccinated with one dose of CoronaVac. Figure 1.Selection of cases and controls. Table 1.Baseline characteristics of cases and controls. OutcomesInfectionHospitalizationSevere COVID-19CasesControlsCasesControlsCasesControlsNumber of individuals36,434109,004223121,788191818,823Age, years (mean (SD))10.35 (4.00)10.35 (4.00)8.70 (4.38)8.67 (4.38)8.53 (4.31)8.50 (4.30)Sex, male (%)20766 (57.0)62211 (57.1)1283 (57.5)12561 (57.7)1106 (57.7)10871 (57.8)Charlson Comorbidity Index (mean (SD))0.03 (0.19)0.03 (0.18)0.07 (0.34)0.03 (0.20)0.05 (0.30)0.03 (0.19)Time since recent dose (mean (SD))88.72 (80.59)75.75 (80.79)85.07 (67.14)81.05 (72.60)84.41 (66.22)79.38 (69.90)Comorbidities – no. (%) Asthma789 (2.2)1955 (1.8)53 (2.4)467 (2.1)40 (2.1)333 (1.8) Diabetes45 (0.1)251 (0.2)9 (0.4)46 (0.2)7 (0.4)32 (0.2) Epilepsy237 (0.7)1961 (1.8)83 (3.7)200 (0.9)77 (4.0)180 (1.0)Medication use within 90 days – no. (%) Immunosuppressants36 (0.1)223 (0.2)13 (0.6)43 (0.2)9 (0.5)37 (0.2) A dose–response relationship was observed in the VE against COVID-19, in which VE increased with more vaccine doses (Table 2 and Figure 2). In children and adolescents who received two vaccine doses, VE ($95\%$ CI) was $31.3\%$ (27.8; 34.7) for BNT162b2 and $21.7\%$ (17.0; 26.2) for CoronaVac. In children and adolescents who received three vaccine doses, VE was observed to be higher at $56.0\%$ (49.6; 61.6) for BNT162b2 and $39.4\%$ (25.6; 50.6) for CoronaVac. Figure 2.Vaccine effectiveness among children and adolescents with different vaccination status. Table 2.Vaccine effectiveness among children and adolescents with different vaccination status. Vaccination statusCaseControlCrude OR ($95\%$ CI)Adjusted OR ($95\%$ CI)VE% ($95\%$ CI)COVID-19 Unvaccinated17,82548,738(Ref)(Ref)(Ref)1 dose BNT162b2410013,2410.733 (0.700–0.769)0.721 (0.688–0.756)27.9 (24.4; 31.2) CoronaVac601120,0160.800 (0.772–0.828)0.793 (0.766–0.821)20.7 (17.9; 23.4)2 doses All BNT162b2540117,1210.704 (0.670–0.741)0.687 (0.653–0.722)31.3 (27.8; 34.7) All CoronaVac252275950.793 (0.748–0.842)0.783 (0.738–0.830)21.7 (17.0; 26.2)3 doses All BNT162b238916600.449 (0.392–0.514)0.440 (0.384–0.504)56.0 (49.6; 61.6) All CoronaVac1304630.619 (0.505–0.759)0.606 (0.494–0.744)39.4 (25.6; 50.6) B-B-C551610.674 (0.492–0.922)0.656 (0.480–0.899)34.4 (10.1; 52.0) C-C-B190.229 (0.029–1.814)0.225 (0.028–1.783)77.5 (−78.3; 97.2)COVID-19-related hospitalization Unvaccinated118811,181(Ref)(Ref)(Ref)1 dose BNT162b215716310.758 (0.619–0.928)0.796 (0.650–0.975)20.4 (2.5; 35.0) CoronaVac29029930.902 (0.781–1.042)0.919 (0.795–1.062)8.1 (−6.2; 20.5)2 doses All BNT162b220224940.588 (0.480–0.720)0.624 (0.509–0.766)37.6 (23.4; 49.1) All CoronaVac34327531.147 (0.975–1.348)1.175 (0.998–1.383)−17.5 (−38.3; 0.2)3 doses All BNT162b2294780.385 (0.248–0.599)0.411 (0.264–0.639)58.9 (36.1; 73.6) All CoronaVac122130.477 (0.261–0.872)0.483 (0.264–0.884)51.7 (11.6; 73.6) B-B-C10431.511 (0.727–3.140)1.563 (0.751–3.250)−56.3 (−225.0; 24.9) C-C-B02–––Severe COVID-19 Unvaccinated10449956(Ref)(Ref)(Ref)1 dose BNT162b212813070.780 (0.623–0.976)0.809 (0.646–1.013)19.1 (−1.3; 35.4) CoronaVac24725960.897 (0.767–1.049)0.912 (0.779–1.068)8.8 (−6.8; 22.1)2 doses All BNT162b216520790.586 (0.469–0.732)0.609 (0.488–0.762)39.1 (23.8; 51.2) All CoronaVac29423061.231 (1.029–1.474)1.249 (1.042–1.497)−24.9 (−49.7; −4.2)3 doses All BNT162b2223600.379 (0.228–0.631)0.398 (0.239–0.663)60.2 (33.7; 76.1) All CoronaVac121820.584 (0.318–1.072)0.578 (0.314–1.062)42.2 (−6.2; 68.6) B-B-C6351.107 (0.444–2.764)1.120 (0.448–2.799)−12.0 (−179.9; 55.2) C-C-B02–––OR: odds ratio, VE: vaccine effectiveness, CI: confidence interval, B-B-C: two doses of BNT162b2 followed by CoronaVac, C-C-B: two doses of CoronaVac followed by BNT162b2. A similar dose–response relationship was observed for the VE of BNT162b2 against COVID-19-related hospitalizations and severe COVID-19. In children and adolescents who received BNT162b2, VE ($95\%$ CI) was $37.6\%$ (23.4; 49.1) against COVID-19-related hospitalization and $39.1\%$ (23.8; 51.2) against severe COVID-19 after two doses, which increased to $58.9\%$ (36.1; 73.6) and $60.2\%$ (33.7; 76.1) after three doses. For CoronaVac, there was a significant reduction in the risk for COVID-19-related hospitalizations and a trend toward a reduction for severe COVID-19 after three doses, with VE ($95\%$ CI) of $51.7\%$ (11.6; 73.6) and $42.2\%$ (−6.2; 68.6), respectively. However, there was no observed risk reduction after two doses (adjusted OR [$95\%$ CI] for hospitalization: 1.175 [0.998–1.383]; severe COVID-19: 1.249 [1.042–1.497]). A significant waning of effectiveness against all outcomes was observed for both vaccines over time (Table 3 and Figure 3). For BNT162b2, VE against COVID-19 and severe COVID-19 peaked at 0–13 days after the second dose, which remained effective up to 61–120 days. For CoronaVac, VE against COVID-19, hospitalization, and severe COVID-19 peaked at 14–60 days, with no significant risk reduction by 61–120 days. However, the trend of waning effectiveness after the third dose was less clear due to the limited number of children and adolescents receiving a third dose for more than 60 days. When waning effectiveness was accounted for by restricting the analyses to those who received their last vaccine dose within 90 days, there was a significant risk reduction in COVID-19-related hospitalizations after two and three doses of CoronaVac (VE [$95\%$ CI] two doses: 21.5 [2.1; 37.1], three doses: $50.9\%$ [2.7; 75.2]), and a significant risk reduction against severe COVID-19 with three doses of CoronaVac (VE [$95\%$ CI]: $49.4\%$ [1.5; 74.0]) (Supplementary Table 4). Figure 3.Vaccine effectiveness for different time-since-vaccination intervals after second dose. Table 3.Vaccine effectiveness for different time-since-vaccination intervals. Days since vaccine doseVE % ($95\%$ CI)0–1314–6061–120121–180≥180COVID-19 Unvaccinated(Ref)(Ref)(Ref)(Ref)(Ref)1 dose BNT162b257.9 (53.7–61.8)28.9 (21.3–35.8)−6.0 (−26.1–10.8)−34.1 (−50.4 to −19.5)−5.6 (−21.7–8.4) CoronaVac35.1 (32.0–38.0)−6.3 (−12.5 to −0.5)29.3 (−1.7–50.8)26.9 (−6.2–49.6)–2 doses All BNT162b258.7 (47.8–67.3)39.7 (29.0–48.7)20.0 (3.2–33.9)−7.4 (−20.7–4.4)16.5 (8.7–23.6) All CoronaVac43.0 (35.2–49.8)29.5 (20.1–37.7)−10.3 (−26.5–3.8)−35.2 (−63.3 to −12.0)–3 doses All BNT162b278.3 (63.4–87.1)39.1 (13.1–57.3)24.8 (−2.3–44.7)15.6 (−26.8–43.8)– All CoronaVac38.6 (−26.9–70.3)−1.4 (−57.6–34.8)33.9 (−49.4–70.8)76.7 (−97.5–97.3)– B-B-C49.3 (−53.3–83.2)54.8 (−21.7–83.2)−13.6 (−118.7–41.0)11.0 (−158.8–69.4)– C-C-B–––––COVID-19-related hospitalization Unvaccinated(Ref)(Ref)(Ref)(Ref)(Ref)1 dose BNT162b219.2 (−25.2–47.8)45.7 (18.2–64.0)0.0 (−64.1–39.1)−0.4 (−42.7–29.4)26.8 (−52.6–64.9) CoronaVac7.5 (−13.7–24.7)−3.4 (−27.3–15.9)55.1 (9.7–77.6)70.6 (34.7 to −86.8)–2 doses All BNT162b281.0 (35.2–94.4)45.5 (−0.2–70.3)27.1 (−14.9–53.7)30.4 (0.4–51.3)33.6 (5.5–53.3) All CoronaVac22.0 (−20.0–49.3)55.8 (34.9–70.0)−39.0 (−77.4 to −8.9)−66.9 (−137.3 to −17.3)–3 doses All BNT162b291.9 (36.4–99.0)58.9 (−15.4–85.4)46.7 (−11.0–74.4)62.0 (−43.3–89.9)– All CoronaVac22.4 (−190.8–79.3)69.5 (12.8–89.3)−91.0 (−745.5–56.8)–– B-B-C––––– C-C-B–––––Severe COVID-19 Unvaccinated(Ref)(Ref)(Ref)(Ref)(Ref)1 dose BNT162b210.9 (−40.6–43.6)47.8 (15.0–68.0)23.2 (−32.7–55.6)−6.9 (−55.4–26.5)20.2 (−80.3–64.7) CoronaVac3.0 (−20.2–21.7)−6.5 (−35.6–16.3)39.8 (−22.7–70.5)65.7 (22.0–84.9)–2 doses All BNT162b270.5 (−9.4–92.1)67.3 (34.1–83.8)48.0 (13.4–68.8)21.4 (−13.6–45.7)29.2 (−4.9–52.2) All CoronaVac6.5 (−53.4–43.0)57.6 (32.8–73.2)−42.5 (−84.6 to −10.1)−48.8 (−120.0 to −0.7)–3 doses All BNT162b2–73.3 (20.5–91.0)75.9 (43.6–89.7)76.2 (−37.9–95.9)– All CoronaVac58.8 (−58.5–89.3)75.7 (23.7–92.3)−33.8 (−463.6–68.2)–– B-B-C––––– C-C-B–––––OR: odds ratio, VE: vaccine effectiveness, CI: confidence interval, B-B-C: two doses of BNT162b2 followed by CoronaVac, C-C-B: two doses of CoronaVac followed by BNT162b2. Consistent results were observed between those aged 3–11 and 12–17 years (Table 4), and when RAT-positive cases were included in the definition of COVID-19 (Supplementary Table 5). Table 4.Subgroup analyses by age. Vaccination statusAged 3–11Aged 12–17CaseControlVE % ($95\%$ CI)CaseControlVE% ($95\%$ CI)COVID-19 Unvaccinated14,56040,045(Ref)32658693(Ref)1 dose BNT162b2579325655.0 (50.6; 59.0)3521998511.9 (6.3; 17.1) CoronaVac476616,06922.3 (19.1; 25.3)1245394717.7 (11.1; 23.8)2 doses All BNT162b217474245.3 (34.1; 54.5)522716,37923.1 (18.4; 27.5) All CoronaVac1571458717.5 (10.5; 24.0)951300823.5 (16.3; 30.1)3 doses All BNT162b200–389166052.0 (44.8; 58.2) All CoronaVac6725939.1 (19.3; 54.1)6320435.6 (13.2; 52.2) B-B-C00–5516128.0 (1.3; 47.5) C-C-B12−16.0 (−1184.2; 89.5)07–COVID-19-related hospitalization Unvaccinated10099817(Ref)1791364(Ref)1 dose BNT162b23646023.3 (−10.1; 46.5)121117123.1 (0.2; 40.8) CoronaVac24926126.3 (−9.7; 20.0)4138119.4 (−16.9; 44.4)2 doses All BNT162b21527344.7 (3.4; 68.4)187222140.3 (23.6; 53.3) All CoronaVac2702225−23.9 (−49.8; −2.5)735283.8 (−34.0; 31.0)3 doses All BNT162b200–2947861.9 (39.3; 76.0) All CoronaVac714151.9 (−5.3; 78.0)57254.0 (−19.5; 82.3) B-B-C00–1043−45.9 (−207.1; 30.7) C-C-B00–02–Severe COVID-19 Unvaccinated8998826(Ref)1451130(Ref)1 dose BNT162b23237817.1 (−22.0; 43.7)9692920.3 (−6.8; 40.6) CoronaVac21722796.0 (−11.4; 20.7)3031727.8 (−11.2; 53.2)2 doses All BNT162b21427648.4 (8.1; 71.0)151180339.5 (20.6; 53.9) All CoronaVac2351936−24.4 (−53.1; −1.1)59370−20.0 (−75.7; 18.1)3 doses All BNT162b200–2236060.7 (32.6; 77.0) All CoronaVac712747.6 (−15.3; 76.2)55535.5 (−70.8; 75.6) B-B-C00–635−8.7 (−177.6; 57.5) C-C-B00–02–OR: odds ratio, VE: vaccine effectiveness, CI: confidence interval, B-B-C: two doses of BNT162b2 followed by CoronaVac, C-C-B: two doses of CoronaVac followed by BNT162b2. ## Discussion Our study evaluated the real-world effectiveness of mRNA (BNT162b2) and inactivated virus (CoronaVac) COVID-19 vaccines among the pediatric population in Hong Kong during the Omicron-dominant pandemic. The results showed a clear dose–response relationship between the number of vaccine doses received and the level of vaccine effectiveness against COVID-19, hospitalizations, and severe symptoms. There was a reduction in VE starting from 60 days after the second dose of both vaccines, which highlights the need for booster doses in children and adolescents. Our VE estimates during the Omicron predominant wave were generally lower than those in other studies, even after booster doses. For BNT162b2, we observed a VE of $31.3\%$ against infection after two doses and a VE of $56.0\%$ after a third dose compared to the reported VE ranging from $48\%$ to $65\%$ in two-dose BNT162b2 recipients aged 5–11 in the Singapore and Israel studies [7,10]. Our results for the VE against COVID-19-related hospitalizations were also lower than those reported by other studies. In the Singapore study, VE against hospitalization was reported to be higher than $80\%$. A similar study in the US that stratified individuals based on variants and age groups also reported higher VE estimates against hospitalization (Aged 12-17: $40\%$ vs. $23.1\%$; Aged 5-11: $68\%$ vs. $45.3\%$) [8]. For CoronaVac, our results showed VE against infection and hospitalization were $21.7\%$ and −$17.5\%$ after two doses, and $39.4\%$ and $51.7\%$ after three doses, whereas VE estimates after two doses were $40\%$ and $60\%$ in the Brazil and Chile studies, respectively [11,12]. One potential explanation for the lower VE levels in our study is the waning protection of both BNT162b2 and CoronaVac, which has been observed in both pediatric and adult populations [9,31]. The VE against infection peaked during the first 60 days after vaccination and then declined thereafter regardless of the number of vaccines doses. Previous studies demonstrated sustained protection against severe outcomes despite waning protection against infection in the general population [31]. However, our results revealed the protection against hospitalization and severe outcomes started to diminish 2 months after vaccination, with more rapid waning in those receiving CoronaVac compared to BNT162b2. Another potential explanation was the introduction of the Vaccine Pass in Hong Kong, which limits unvaccinated individuals from visiting high-risk premises. As a result, the incidence of COVID-19 was underestimated in the unvaccinated group, and thus VE was also underestimated. The mechanism of immune protection may account for a more rapid reduction in VE against COVID-19 over time, but to a lesser extent against COVID-19-related complications. Although humoral immunity mediated by antibodies blocks SARS-CoV-2 from entering host cells and thus prevents infection [32], specific CD4+ and CD8+ T cells appear to be responsible for limiting disease severity [33]. As a result, despite the rapid reduction of serum antibody titers, memory T cells are more durable and may contribute to the protection from severe disease. The main strength of this study is that it is one of the first to provide real-world evidence on the effectiveness of both mRNA (BNT162b2) and inactivated virus (CoronaVac) vaccines against different Omicron subvariants among the pediatric population. Our findings highlight the importance of booster doses for reducing the risk of COVID-19 and preventing subsequent COVID-19-related complications including hospitalization and severe outcomes. Nonetheless, the findings of this study need to be interpreted with the following caveats. First, given the limited number of individuals receiving heterologous booster following the primary vaccine course, the VE of a booster dose against several outcomes could not be fully evaluated in this study. Second, a negative case-control study design is not feasible because only positive PCR or RAT results are reported to the DH. Moreover, there is a possibility that people with asymptomatic COVID-19 could be misclassified as controls, leading to a bias in the estimates towards the null. Asymptomatic COVID-19 cases cannot be captured unless nationwide screening is conducted, which at present is not possible in most countries. Furthermore, it is also possible that people with COVID-19 were misclassified due to false negative PCR results, but this is likely to be minimal due to the high specificity, and even less likely for severe cases. Thirdly, although the time-since-vaccination interval was not adjusted in our primary analysis, the assessment of waning vaccine effectiveness was stratified. Lastly, only patients who attended HA services could be included as control in our study. However, the selection bias was minimized during the matching process and with the adjustment of comorbidities during conditional logistic regression. In conclusion, three doses of either BNT162b2 or CoronaVac vaccine is effective in preventing COVID-19, hospitalizations, and severe outcomes in children and adolescents during the Omicron variant-dominant pandemic. Our findings highlight the need for booster doses to enhance vaccine effectiveness in children and adolescents. ## Conflict of interest FTTL has been supported by the RGC Postdoctoral Fellowship under the Hong Kong Research Grants Council and has received research grants from the Food and Health Bureau of the Government of the Hong Kong Special Administrative Region, outside the submitted work. XL has received research grants from the Food and Health Bureau of the Government of the Hong Kong Special Administrative Region; research and educational grants from Janssen and Pfizer; internal funding from the University of Hong Kong; and consultancy fees from Merck Sharp & Dohme, unrelated to this work. EYFW has received research grants from the Food and Health Bureau of the Government of the Hong Kong Special Administrative Region, and the Hong Kong Research Grants Council, outside the submitted work. CKHW reports receipt of research funding from the EuroQoL Group Research Foundation, the Hong Kong Research Grants Council, and the Hong Kong Health and Medical Research Fund; outside of the submitted work. EWYC reports honorarium from Hospital Authority; and grants from Research Grants Council (RGC, Hong Kong), Research Fund Secretariat of the Food and Health Bureau, National Natural Science Fund of China, Wellcome Trust, Bayer, Bristol-Myers Squibb, Pfizer, Janssen, Amgen, Takeda, and Narcotics Division of the Security Bureau of the Hong Kong Special Administrative Region, outside the submitted work. CSLC has received grants from the Food and Health Bureau of the Hong Kong Government, Hong Kong Research Grant Council, Hong Kong Innovation and Technology Commission, Pfizer, IQVIA, MSD, and Amgen; and personal fees from PrimeVigilance; outside the submitted work. ICKW reports research funding outside the submitted work from Amgen, Bristol-Myers Squibb, Pfizer, Janssen, Bayer, GSK, Novartis, the Hong Kong Research Grants Council, the Food and Health Bureau of the Government of the Hong Kong Special Administrative Region, National Institute for Health Research in England, European Commission, and the National Health and Medical Research Council in Australia; has received speaker fees from Janssen and Medice in the previous 3 years; and is an independent non-executive director of Jacobson Medical in Hong Kong. All other authors declare no competing interests. PI has received funding from the Hong Kong Research Grants Council, the Food and Health Bureau of the Government of the Hong Kong Special Administrative Region, and Hong Kong Jockey Club Charities Trust. ## Data sharing statement Data will not be available for others as the data custodians have not given permission. ## Disclosure statement FTTL has been supported by the RGC Postdoctoral Fellowship under the Hong Kong Research Grants Council and has received research grants from the Food and Health Bureau of the Government of the Hong Kong Special Administrative Region, outside the submitted work. XL has received research grants from the Food and Health Bureau of the Government of the Hong Kong Special Administrative Region; research and educational grants from Janssen and Pfizer; internal funding from the University of Hong Kong; and consultancy fees from Merck Sharp & Dohme, unrelated to this work. EYFW has received research grants from the Food and Health Bureau of the Government of the Hong Kong Special Administrative Region, and the Hong Kong Research Grants Council, outside the submitted work. CKHW reports receipt of research funding from the EuroQoL Group Research Foundation, the Hong Kong Research Grants Council, and the Hong Kong Health and Medical Research Fund; outside of the submitted work. EWYC reports honorarium from Hospital Authority; and grants from Research Grants Council (RGC, Hong Kong), Research Fund Secretariat of the Food and Health Bureau, National Natural Science Fund of China, Wellcome Trust, Bayer, Bristol-Myers Squibb, Pfizer, Janssen, Amgen, Takeda, and Narcotics Division of the Security Bureau of the Hong Kong Special Administrative Region, outside the submitted work. CSLC has received grants from the Food and Health Bureau of the Hong Kong Government, Hong Kong Research Grant Council, Hong Kong Innovation and Technology Commission, Pfizer, IQVIA, MSD, and Amgen; and personal fees from PrimeVigilance; outside the submitted work. ICKW reports research funding outside the submitted work from Amgen, Bristol-Myers Squibb, Pfizer, Janssen, Bayer, GSK, Novartis, the Hong Kong Research Grants Council, the Food and Health Bureau of the Government of the Hong Kong Special Administrative Region, National Institute for Health Research in England, European Commission, and the National Health and Medical Research Council in Australia; has received speaker fees from Janssen and Medice in the previous 3 years; and is an independent non-executive director of Jacobson Medical in Hong Kong. All other authors declare no competing interests. PI has received funding from the Hong Kong Research Grants Council, the Food and Health Bureau of the Government of the Hong Kong Special Administrative Region, and Hong Kong Jockey Club Charities Trust. ## References 1. Ludvigsson JF.. Systematic review of COVID-19 in children shows milder cases and a better prognosis than adults. Acta Paediatr (2020) 109 1088-1095. PMID: 32202343 2. Delahoy MJ, Ujamaa D, Whitaker M. Hospitalizations associated with COVID-19 among children and adolescents — COVID-NET, 14 states, March 1, 2020–August 14, 2021. MMWR Morb Mortal Wkly Rep. (2021) 70 1255-1260. PMID: 34499627 3. Tso WWY, Kwan MYW, Wang YL. Severity of SARS-CoV-2 Omicron BA.2 infection in unvaccinated hospitalized children: comparison to influenza and parainfluenza infections. Emerg Microbes Infect (2022) 11 1742-1750. PMID: 35730665 4. Taylor CA, Whitaker M, Anglin O. COVID-19–associated hospitalizations among adults during SARS-CoV-2 delta and Omicron variant predominance, by race/ethnicity and vaccination status — COVID-NET, 14 states, July 2021–January 2022. Morb Mortal Wkly Rep (2022) 71 466 5. Chua GT, Kwan MYW, Chui CSL. Epidemiology of acute myocarditis/pericarditis in Hong Kong adolescents following comirnaty vaccination. Clin Infect Dis (2022) 75 673-681. PMID: 34849657 6. Lai FTT, Li X, Peng K. Carditis after COVID-19 vaccination with a messenger RNA vaccine and an inactivated virus vaccine: a case-control study. Ann Intern Med (2022) 175 362-370. PMID: 35073155 7. Cohen-Stavi CJ, Magen O, Barda N. BNT162b2 vaccine effectiveness against Omicron in children 5 to 11 years of age. N Engl J Med (2022) 387 227-236. PMID: 35767475 8. Price AM, Olson SM, Newhams MM. BNT162b2 protection against the Omicron variant in children and adolescents. N Engl J Med (2022) 386 1899-1909. PMID: 35353976 9. Sacco C, Del Manso M, Mateo-Urdiales A. Effectiveness of BNT162b2 vaccine against SARS-CoV-2 infection and severe COVID-19 in children aged 5-11 years in Italy: a retrospective analysis of January-April, 2022. Lancet (2022) 400 97-103. PMID: 35780801 10. Tan SHX, Cook AR, Heng D. Effectiveness of BNT162b2 vaccine against Omicron in children 5 to 11 years of age. N Engl J Med (2022) 387 525-532. PMID: 35857701 11. Florentino PTV, Alves FJO, Cerqueira-Silva T. Vaccine effectiveness of CoronaVac against COVID-19 among children in Brazil during the Omicron period. Nat Commun (2022) 13 4756. PMID: 35963844 12. Jara A, Undurraga EA, Zubizarreta JR. Effectiveness of CoronaVac in children 3-5 years of age during the SARS-CoV-2 Omicron outbreak in Chile. Nat Med (2022) 28 1377-1380. PMID: 35605637 13. Cheng FWT, Fan M, Wong CKH. The effectiveness and safety of mRNA (BNT162b2) and inactivated (CoronaVac) COVID-19 vaccines among individuals with chronic kidney diseases. Kidney Int (2022) 102 922-925. PMID: 35964798 14. Chui CSL, Fan M, Wan EYF.. Thromboembolic events and hemorrhagic stroke after mRNA (BNT162b2) and inactivated (CoronaVac) covid-19 vaccination: a self-controlled case series study. eClinicalMedicine (2022) 50 101504. PMID: 35770253 15. Lai FTT, Chua GT, Chan EWW. Adverse events of special interest following the use of BNT162b2 in adolescents: a population-based retrospective cohort study. Emerg Microbes Infect (2022) 11 885-893. PMID: 35254219 16. Lai FTT, Huang L, Chui CSL. Multimorbidity and adverse events of special interest associated with Covid-19 vaccines in Hong Kong. Nat Commun (2022) 13 411. PMID: 35058463 17. Leung TYM, Chan AYL, Chan EW. Short- and potential long-term adverse health outcomes of COVID-19: a rapid review. Emerg Microbes Infect (2020) 9 2190-2199. PMID: 32940572 18. Li X, Tong X, Yeung WWY. Two-dose COVID-19 vaccination and possible arthritis flare among patients with rheumatoid arthritis in Hong Kong. Ann Rheum Dis (2022) 81 564-568. PMID: 34686479 19. Li X, Gao L, Tong X. Autoimmune conditions following mRNA (BNT162b2) and inactivated (CoronaVac) COVID-19 vaccination: A descriptive cohort study among 1.1 million vaccinated people in Hong Kong. J Autoimmun (2022) 130 102830. PMID: 35461018 20. Sing CW, Tang CTL, Chui CSL. COVID-19 vaccines and risks of hematological abnormalities: nested case-control and self-controlled case series study. Am J Hematol (2022) 97 470-480. PMID: 35080247 21. Wan EYF, Chui CSL, Lai FTT. Bell’s palsy following vaccination with mRNA (BNT162b2) and inactivated (CoronaVac) SARS-CoV-2 vaccines: a case series and nested case-control study. Lancet Infect Dis (2022) 22 64-72. PMID: 34411532 22. Wan EYF, Chui CSL, Wang Y. Herpes zoster related hospitalization after inactivated (CoronaVac) and mRNA (BNT162b2) SARS-CoV-2 vaccination: A self-controlled case series and nested case-control study. Lancet Reg Health West Pac (2022) 21 100393. PMID: 35128500 23. Wan EYF, Wang Y, Chui CSL. Safety of an inactivated, whole-virion COVID-19 vaccine (CoronaVac) in people aged 60 years or older in Hong Kong: a modified self-controlled case series. Lancet Healthy Longev (2022) 3 e491-e500. PMID: 35813276 24. Wong CKH, Lau KTK, Xiong X. Adverse events of special interest and mortality following vaccination with mRNA (BNT162b2) and inactivated (CoronaVac) SARS-CoV-2 vaccines in Hong Kong: A retrospective study. PLoS Med (2022) 19 e1004018. PMID: 35727759 25. Wong CKH, Mak LY, Au ICH. Risk of acute liver injury following the mRNA (BNT162b2) and inactivated (CoronaVac) COVID-19 vaccines. J Hepatol (2022) 77 1339-1348. PMID: 35817224 26. Xiong X, Wong CKH, Au ICH. Safety of inactivated and mRNA COVID-19 vaccination among patients treated for hypothyroidism: a population-based cohort study. Thyroid (2022) 32 505-514. PMID: 35216517 27. Ye X, Ma T, Blais JE. Association between BNT162b2 or CoronaVac COVID-19 vaccines and major adverse cardiovascular events among individuals with cardiovascular disease. Cardiovasc Res (2022) 118 2329-2338. PMID: 35732274 28. Huang YQ, Gou R, Diao YS. Charlson comorbidity index helps predict the risk of mortality for patients with type 2 diabetic nephropathy. J Zhejiang Univ Sci B (2014) 15 58-66. PMID: 24390745 29. Miller TE, Garcia Beltran WF, Bard AZ. Clinical sensitivity and interpretation of PCR and serological COVID-19 diagnostics for patients presenting to the hospital. FASEB J (2020) 34 13877-13884. PMID: 32856766 30. Pearce N.. Analysis of matched case-control studies. BMJ (2016) 352 i969. PMID: 26916049 31. Chemaitelly H, Tang P, Hasan MR. Waning of BNT162b2 vaccine protection against SARS-CoV-2 infection in Qatar. N Engl J Med (2021) 385 e83. PMID: 34614327 32. Wherry EJ, Barouch DH.. T cell immunity to COVID-19 vaccines. Science (2022) 377 821-822. PMID: 35981045 33. Rydyznski Moderbacher C, Ramirez SI, Dan JM. Antigen-Specific adaptive immunity to SARS-CoV-2 in acute COVID-19 and associations with age and disease severity. Cell (2020) 183 996-1012.e19. PMID: 33010815
--- title: 'Movement matters: short-term impacts of physical activity on mood and well-being' authors: - Loree T. Pham - Raymond Hernandez - Donna Spruijt-Metz - Jeffrey S. Gonzalez - Elizabeth Ann Pyatak journal: Journal of Behavioral Medicine year: 2023 pmcid: PMC10026784 doi: 10.1007/s10865-023-00407-9 license: CC BY 4.0 --- # Movement matters: short-term impacts of physical activity on mood and well-being ## Abstract Few studies have investigated the short-term, momentary relationships between physical activity (PA) and well-being. This study focuses on investigating the dynamic relationships between PA and affective well-being among adults with type 1 diabetes. Participants ($$n = 122$$) wore an accelerometer and completed daily EMA surveys of current activities and affective states (e.g., happy, stressed, excited, anxious) via smartphone over 14 days. Within-person, increased sedentary time was associated with less positive affect (r = − 0.11, $p \leq 0.001$), while more PA of any intensity was associated with greater positive affect and reduced fatigue, three hours later. Between-person, increased light PA was associated with increased stress ($r = 0.21$, $$p \leq 0.02$$) and diabetes distress ($r = 0.30$, $$p \leq 0.001$$). This study provides evidence that positive affect and fatigue are predicted by previous activity regardless of the different activities that people engaged in. Positive affect increased after engaging in PA. However, participants with higher amounts of light PA reported higher stress ratings. ## Introduction Physical activity is associated with numerous and substantial health benefits (Warburton & Bredin, 2017; Warburton et al., 2006), but while the prevalence of meeting high aerobic physical activity guidelines has increased over the past decade (Whitfield et al., 2021), only $53.3\%$ of U.S. adults meet aerobic physical activity guidelines (CDC/National Center for Health Statistics, 2021). Excessive sedentary behavior may influence the development of chronic conditions such as cardiovascular disease (Chomistek et al., 2013; Patterson et al., 2018), type 2 diabetes (Bowden Davies et al., 2018; Paing et al., 2018; Patterson et al., 2018; Sardinha et al., 2017), certain cancers (Cong et al., 2014; Schmid & Leitzmann, 2014), and low back pain (Citko et al., 2018; Subramanian & Arun, 2017). It is especially important for individuals with type 1 diabetes to engage in physical activity (Riddell et al., 2017) as a means of reducing the risk of complications and comorbid conditions. Hence, there continues to be a need for the promotion of physical activity and an effort to reduce physical inactivity and sedentary activity. Besides reducing the risk of chronic health conditions, there is substantial, evidence-based support that physical activity improves psychological well-being (Warburton & Bredin, 2017). While numerous behavior change theories (e.g. Michie et al., 2014) help to explain why people adopt physical activity behaviors, it is equally if not more important to consider what may help people sustain behaviors over time. For example, people are more likely to maintain behavior change if they are satisfied with the new behavior after weighing the relative costs and benefits (Rothman et al., 2011). Therefore, it is important to consider physical activity’s impact on mood and well-being, as the promotion of short-term positive mood states (Shiota et al., 2021; Van Cappellen et al., 2018) and well-being variables (Razazian et al., 2020; Stults-Kolehmainen & Sinha, 2014) may be beneficial in promoting sustainable long-term behavior change. Similarly, considering activity type and its importance to the individual may be useful when aiming to get people to move more, since certain activities may hold more meaning and impact compared to others. For instance, men spend a greater amount of time engaged in physical activity than women (Hamrik et al., 2014; Saffer et al., 2013), particularly in group-based occupations (e.g., sports), while women are more likely to participate in individual-level occupations (e.g., walking or housework) (Azevedo et al., 2007). Contextual factors may also play a role (Craft et al., 2014). For example, women and older adults are more likely to exercise at home, men are more likely to exercise outdoors or at work, younger adults are more likely to exercise with friends, and individuals with lower income are more likely to exercise with family members (Dunton et al., 2008; Welk & Kim, 2015). Gaining a better understanding of the impacts on mood may be beneficial in promoting participation in physical activity and improving health and well-being. Though there exists an extensive body of research related to physical activity, there are a number of methodological limitations that need to be addressed (Kanning et al., 2013). For example, a vast majority of physical activity studies rely on retrospective self-report, such as estimating the amount of physical activity engaged in over the past week or month, which may be subject to participant bias and overestimation and may not be reliable (Warburton & Bredin, 2017). Some of these limitations may be addressed through the use of electronic ecological momentary assessment (EMA). EMA allows participants’ behaviors and experiences to be recorded in real-time and in real-world contexts while minimizing retrospective recall and memory biases (Shiffman et al., 2008). Using electronic EMA may be even more advantageous than the traditional paper-and-pencil EMA in terms of reliability and compliance (Berkman et al., 2014; Stone et al., 2002). An additional benefit of using electronic EMA is the ability to record the exact time of survey completion, which is beneficial when pairing the EMA data with other types of data, such as accelerometry data, that is collected in real time. Overall, while there are numerous studies examining the relationship between physical activity and well-being (Marquez et al., 2020), relatively few studies have investigated the short-term, momentary relationships between the two. Furthermore, to our knowledge, activity type and activity importance have not been thoroughly explored in physical activity research. Lastly, there is a need to incorporate underrepresented populations, such as diverse racial and ethnic groups, in order to have a greater public health impact (Marquez et al., 2020). The study sample consists of individuals with type 1 diabetes, who face the same barriers for physical activity as the overall population but who also may face different challenges and levels of healthcare support (Pyatak et al., 2014). To address the present research gaps, this study selected and recruited from clinical sites with a diverse patient population, and used real-time data collection via EMA and accelerometry to examine the relationships between physical activity and well-being. Combined, these momentary data collection methods may offer greater insight into the relationship between physical activity and well-being. The purpose of this study is to investigate the relationship between physical activity and subsequent mood, while controlling for the effect of activity type on mood. ## Participants This study is a secondary analysis from an overarching study investigating relationships between blood glucose, function, and emotional well-being in adults with type 1 diabetes (T1D) (Pyatak et al., 2021). In this case, function refers to self-reported daily life activity performance, objective cognitive function, and physical activity derived from accelerometers. Emotional well-being in this study refers to positive and negative affect, stress, and diabetes distress. Participants were recruited from clinical sites in Los Angeles and New York City metropolitan areas. Eligibility criteria were specified to ensure that participants were able to fully complete the study procedures and do not have any health conditions besides diabetes that could significantly impact blood glucose levels. Participants were required to be able to answer EMA surveys throughout the day, including while at work. Greater detail regarding the parent study methodology and research questions are outlined in (Pyatak et al., 2021). In the parent study, 196 participants were recruited and started the study protocol. ## Data collection Due to COVID-19, recruitment and data collection was completed remotely, through phone calls, videoconferencing, emails, and mailings. Participants who expressed interest in the study completed a screening questionnaire over the phone. Eligible participants completed enrollment paperwork and baseline surveys through an online data capture platform, Research Electronic Data Capture (Harris et al., 2009). Study equipment was shipped directly to participants and included the following: accelerometer, smartphone, participant manual, ClinCard for loading stipends onto, and shipping materials to mail the study equipment back after the completion of data collection. ## Accelerometry The Actigraph, Inc., wGT3X_BT model accelerometer was used to objectively measure participants’ physical activity, providing continuous data that could be used to infer time spent in sedentary, light, moderate, and vigorous physical activity each day. This core study device was attached to an adjustable strap, which participants wore on their non-dominant wrist. Participants were instructed to wear the device continuously for 14 consecutive days, removing the device only for water-based activities, such as bathing. The accelerometer recorded the intensity of participants’ activity at a 30 Hz sampling frequency, with each recording time-stamped. The 30-s epoch was chosen over longer epochs in order to better capture shorter bouts of activity, which may help to reduce misclassification errors of physical activity estimates (Brønd & Arvidsson, 2016; Gabriel et al., 2010). ## Ecological momentary assessment (EMA) Participants completed 14 days of EMA surveys and mobile cognitive tasks via Xiaomi Mi A1 smartphone (AT&T USA, Inc.). The phones came pre-installed with the necessary apps, including mEMA (mobile EMA by Ilumivu) which is a HIPAA-compliant software application that would prompt the EMA surveys at 3-h intervals and store survey responses on a secure cloud-based server for data management (ilumivu: Software for Humanity, n.d.). Participants completed about 5–6 surveys per day that were scheduled depending on the participants’ individual preferences. Depending on the branching logic, participants answered approximately 30 survey items in the first five surveys of the day and 50 items in the evening survey. ## Physical activity Using the data recordings from the accelerometer, activity counts were converted to minutes spent in light, moderate, and vigorous physical activity. Light physical activity was defined as the time that was not spent in either moderate-vigorous activity or sedentary activity (which is considered less than 100 counts/minute; Healy et al., 2008). Consistent with national surveillance studies (Troiano et al., 2008), moderate-vigorous physical activity is defined as 2020 activity counts per minute. Time in physical activity ranges were based on three hour periods with at least $75\%$ of the data present in that period. ## Well-being and activity Positive affect, negative affect, stress, diabetes distress, and fatigue were assessed using EMA survey questions (Table 1). The EMA questions were selected based on validated global measures and/or being successfully used in previous EMA studies (Broderick et al., 2009; Crawford & Henry, 2004; Dunton et al., 2008; Laurenceau, 2013; Merwin et al., 2015; Scott et al., 2017). At each time point, participants rated indicators of well-being as well as identified the type of activity that they were engaging in at the time of the signal. In order to reduce selection and recall bias for activity type, a sampling approach rather than coverage approach was used, where participants selected the type of activity that they were immediately engaged in prior to the survey, rather than selecting the activity that they were engaged in for the longest duration in the time frame period to the survey notification (Shiffman et al., 2008). Possible activity type responses were derived from the Occupational Therapy Practice Framework: Domain and Process-Fourth Edition (American Occupational Therapy Association, 2020b) using lay language agreed upon by the research team with input from pilot testing of the study protocol prior to enrolling participants. The term “activity” was chosen as a more accessible term for the study participants, but the terms “activity” and “occupation” will be used interchangeably in the subsequent discussion. Table 1Ecological momentary assessment survey measuresConstructItem(s)Response Option(s)Well-BeingPositive affect4 items: Average of mood ratings for “happy”, “content”, “enthusiastic”, “excited”For each mood, 0 (not at all) to 100 (extremely)Negative affect4 items: Average of mood ratings for “tense”, “upset”, “sad”, “disappointed”For each mood, 0 (not at all) to 100 (extremely)StressHow stressed are you right now?0 (Not at all stressed) to 100 (Extremely stressed)Diabetes distressHow stressed do you feel about your diabetes or diabetes management right now?0 (Not at all stressed) to 100 (Extremely stressed)FatigueAt this moment, how tired do you feel?0 (Not at all) to 100 (Extremely)PainAt this moment, how much bodily pain do you have?0 (None) to 100 (Extreme pain)ActivityActivity typeWhat were you doing right before starting this survey?Work/school activities (e.g., paid labor, volunteer work, and studying)activity type examples were not included in the survey but were listed in the participant manual and elaborated upon during the baseline trainingTraveling (e.g., driving, riding in a car, walking)Relaxing/chilling (e.g., passive leisure like watching Netflix, listening to music)Sleeping/nappingSocializing (e.g., talking with friends/family)Caring for myself (e.g., eating, dressing, bathing, toileting, personal grooming)Caring for others (e.g., caring for your children and pets, if you’re caring for others as part of work this counts as “work”)Doing housework/errands (e.g., paying bills, washing dishes and clothes, exercising for health)Fun/play/leisure activities (e.g., active leisure like exercising for fun, video games, reading for fun)Other (If chosen, please specify)Activity importanceHow important is this activity to you?0 (Not at all) to 100 (Extremely) ## Statistical analysis In order to be included in the analysis, participants need to have worn the accelerometer for at least $75\%$ of the time block (i.e. 2 h and 15 min out of 3 h). Between and within-person correlations between physical activity, well-being, and activity measures were calculated using the “psych” package in R (Revelle, 2021). The “statsBy” function in this package can decompose observed correlations from longitudinal data into within and between-person components, with the within portion derived from the pooled correlation within groups and between portion from the weighted correlation of the means between groups (Revelle, 2021). The correlations are between measures of well-being at the time of the survey and physical activity metrics calculated from the three-hour time period preceding the survey prompt. To determine if relationships between physical activity and well-being were significant after adjusting for activity type and activity importance, mixed effects models were run using the “lme4” package in R (Bates et al., 2015). Compared to ordinary regression, mixed models are better suited for analyzing longitudinal data as they can account for non-independence of observations (e.g., multiple days nested in an individual) (Bell, 2013). Separate models were run, with positive affect, negative affect, stress, fatigue, or pain as the dependent variable, a single PA metric as the focal predictor (i.e. time sedentary, time light, or time moderate-vigorous), and activity type and/or activity importance as a covariate to control for it. Person-mean centered versions of the PA variables and activity type and importance were used to make sure that only their within-person components (and not between-person) would be represented (Curran & Bauer, 2011). In terms of the mixed model settings, the intercept and slopes of the PA variables were specified as random effects, which allows them to vary according to the clusters (e.g. different individuals). An unstructured covariance matrix was used, which chooses parameters that best fit the data at the cost of a greater number of parameters specified (Kincaid, 2005). ## Descriptive statistics Of 136 total participants at the time analyses were conducted, 14 were excluded due to incomplete data, leaving A diverse sample of 122 total participants included in the analysis (see Table 2 for demographic characteristics). The median EMA completion percentage was $92.86\%$, and four or more EMA surveys were completed on $82.34\%$ of all data collection days across the participants. The final dataset included a total of 8,639 EMA data points across 1,812 days. Table 2Demographic characteristics ($$n = 122$$)CharacteristicnMean (SD) or Percent (%)Age (years)12241.12 (14.83)GenderMale$5444\%$Female$6856\%$Race/EthnicityWhite$4335\%$Latino/x$4638\%$Black$1411\%$Asian$43\%$Multi-ethnic$97\%$Other$43\%$Not provided$22\%$EducationHigh school grad or less$2520\%$Some college, no degree$2621\%$Associate’s degree$1311\%$Bachelor’s degree$3730\%$Graduate degree$1916\%$Not provided$22\%$Annual household incomeLess than $50,$0004739\%$$50,000–$99,$9992319\%$Above $100,$0002117\%$Did not wish to provide$2420\%$Did not know$76\%$ ## Within-person Increased time spent sedentary over a three hour period was associated with decreased subsequent positive affect (r = − 0.03, $p \leq 0.01$) and increased diabetes stress ($r = 0.04$, $p \leq 0.001$) three hours later. Meanwhile, more physical activity at any intensity was associated with greater positive affect and reduced fatigue three hours later. Similar increases in positive affect were observed with light, moderate, and vigorous PA ($r = 0.10$, 0.10, 0.09 respectively, all $p \leq 0.001$). Fatigue decreases with light PA, moderate PA, and vigorous PA (r = − 0.07, − 0.05, − 0.05 respectively, all $p \leq 0.001$). See Table 3 for full results looking at positive and negative affect, stress, diabetes stress, fatigue, and pain at different activity levels. Based on the significant correlations seen with positive affect and fatigue, additional testing was run to determine if the correlations were due to any other factors, such as the activity just reported. Table 4 illustrates the relationship between physical activity and positive affect before and after adjustment by activity type and activity importance. Table 5 illustrates the relationship between physical activity and fatigue before and after adjustment by activity type and activity importance. Table 6 indicates the various activity types that were significantly associated with positive affect, when using the “relaxing/chilling” activity type as a reference point. Table 3Within-person relationships between physical activity and subsequent well-beingPositive AffectNegative AffectStressDiabetes StressFatiguePainSedentaryr = − 0.11$r = 0.02$r = − 0.01r = 0.02r = 0.06*r = − 0.01Lightr = 0.10r = − 0.01r = 0.01r = − 0.02r = − 0.07$r = 0.01$Moderater = 0.10r = − 0.01r = 0.02r = − 0.02r = − 0.05$r = 0.02$Vigorousr = 0.09r = − 0.02r = 0.00r = − 0.01r = − 0.05$r = 0.00$Moderate or Vigorousr = 0.11r = − 0.02r = 0.01r = − 0.01r = − 0.06*$r = 0.01$ = significant at α = 0.05; = significant at α = 0.01; * = significant at α = 0.001Table 4Within-person physical activity and positive affect after adjustment for activity type and activity importanceBefore adjustmentAfter adjustmentFor every one minute increase in time spent in sedentary activity, positive affect decreased by: − 0.058p <.001 − 0.038p <.001For every one minute increase in time spent in light activity, positive affect increased by:0.193p <.0010.120p <.001For every one minute increase in time spent in moderate or vigorous activity, positive affect increased by:0.069p <.0010.045p <.001Table 5Within-person physical activity and fatigue after adjustment for activity type and activity importanceBefore adjustmentAfter adjustmentFor every one minute increase in time spent in sedentary activity, fatigue decreased by:0.054p <.001 − 0.007p = 0.643For every one minute increase in time spent in light activity, fatigue increased by: − 0.230p <.001 − 0.045p = 0.395For every one minute increase in time spent in moderate or vigorous activity, fatigue increased by: − 0.065p <.0010.009p = 0.629Table 6Within-person relationship between activity type and positive affectEstimatep-valueSedentaryTime spent0.038 < 0.001Activity importance0.17 < 0.001Activity reported (Relaxing/chilling is reference group)Doing Fun/play/leisure activities4.29 < 0.001Doing Sleeping/napping − 7.04 < 0.001Doing Socializing5.43 < 0.001Traveling1.210.166Doing Work/school activities − 2.41 < 0.001LightTime spent0.120 < 0.001Activity importance0.17 < 0.001Activity reported (Relaxing/chilling is reference group)Doing Fun/play/leisure activities4.33 < 0.001Doing Sleeping/napping − 7.00 < 0.001Doing Socializing5.34 < 0.001Traveling1.340.126Doing Work/school activities − 2.40 < 0.001Moderate or vigorousTime spent0.045 < 0.001Activity importance0.17 < 0.001Activity reported (Relaxing/chilling is reference group)Doing Fun/play/leisure activities4.35 < 0.001Doing Sleeping/napping − 7.03 < 0.001Doing Socializing5.55 < 0.001Traveling1.280.143Doing Work/school activities − 2.42 < 0.001 Results from this study suggest that within-person, more sedentary time is associated with subsequent decreased positive affect, while greater physical activity is associated with subsequent increased positive affect. The negative relationship between sedentary activity and positive affect is consistent with previous EMA studies (Smith et al., 2020; Wen et al., 2018). The positive relationship between positive affect and physical activity was found for light, moderate, and vigorous levels of activity. This finding is consistent with previous EMA studies where higher levels of positive affect were found after engagement in moderate-vigorous physical activity (Dunton et al., 2014; Liao et al., 2015; Wen et al., 2018). No significant associations were found between physical activity and negative affect. These null findings are consistent with a meta-analysis of 12 studies that also examined affective responses from physical activity and found no significant relationship between light physical activity and negative affect (Wiese et al., 2018). Numerous meta-analyses across different patient populations demonstrate a relationship between physical activity and the reduction of fatigue (Juvet et al., 2017; Oberoi et al., 2018; Razazian et al., 2020). Activity type and activity importance are shown to have an effect on both positive affect and fatigue, where the types of activities people engage in and the perceived importance of the activities influence the change in positive affect and fatigue that is caused by differing levels of physical activity. In other words, the associations between both physical activity and positive affect and physical activity and fatigue are greater when activity type and activity importance are considered. The low correlations are consistent with other studies examining short-term, momentary relationships (Bennett et al., 2020; Yang et al., 2021). Additionally, while the effect sizes for these relationships are small, they should be considered in the context of a lifetime. Experiencing slight fluctuations in mood on any given day may not seem significant, but these experiences have the potential to be repeated countless times throughout the lifespan. Thus, though the momentary effects are marginal, the overall effect on mood and well-being may be substantial. ## Between-person When examining between-person relationships between physical activity and well-being, increased light PA was associated with increased stress ($r = 0.21$, $$p \leq 0.023$$). See Table 7 for full results. As we hypothesized this finding could be due to work demands that may be stressful and require light PA, we ran an additional test to determine if the increase in stress was moderated by employment status (i.e., working or not); but significant moderation was not found ($$p \leq 0.10$$).Table 7Between-person relationships between physical activity and subsequent well-beingPositive affectNegative affectStressDiabetes distressFatiguePainSedentaryr = 0.14r = − 0.09r = − 0.10r = − 0.10r = 0.10r = 0.05Lightr = − 0.02r = 0.13r = 0.21$r = 0.30$$r = 0.09$r = 0.13Moderater = − 0.13r = 0.10r = 0.11r = 0.20r = − 0.02r = 0.02Vigorousr = − 0.15r = 0.04r = 0.03r = − 0.02r = − 0.15r = − 0.12Moderate or Vigorousr = − 0.16r = 0.07r = 0.06r = 0.05r = − 0.12r = − 0.08* = significant at α = 0.05; = significant at α = 0.01; * = significant at α = 0.001 Between-person, people that spent more time in light activity over the study period also typically had higher average stress over the study period. This may be because light activities may include running errands or other hassles and stressful events that occur in people’s everyday lives that may lead to increased stress states. Previous studies have had mixed results regarding light physical activity and stress. Similar to this study, Jones et al. [ 2017] found that higher light activity was associated with higher stress in real-time. These findings are in contrast to an earlier study that did not find any associations between objectively measured physical activity and stress (Poole et al., 2011). The participant sample from this study is closer in similarity to Jones et al. [ 2017], which may explain the differing results. ## Discussion This study adds to our understanding of how physical activity is associated with subsequent mood. Importantly, while previous studies have investigated objectively measured physical activity impacts on mood, this study examines the effects of activity type and activity importance. Accelerometry measures do not provide contextual information about the type of activity, so previous studies have remarked on the difficulty of disentangling the importance of activity type (Poole et al., 2011), which this study helps to address. Additionally, the study was conducted among an ethnically and socioeconomically diverse population, which strengthens its generalizability. ## Limitations Despite the advantages of using accelerometry to objectively measure activity level and electronic EMA survey methods to assess well-being, this study had a few limitations. To start, the sample consisted of individuals who had type 1 diabetes and had the added experience of participating in the study amidst the COVID-19 pandemic and its related social distancing effects during data collection, hence their experiences may differ from the general population. Furthermore, since the participants were limited to particular geographic areas (Los Angeles and New York), the results may not be applicable to other populations in other areas of the United States or around the world. Replication of this protocol with larger, more diverse samples is needed to confirm the relationships suggested by this study. Additionally, the construct of well-being is a broad, multidimensional concept and all aspects of well-being were not covered in this study. Well-being measures were limited to those included in the overarching study. ## Conclusions This study provides evidence that positive affect is predicted by previous activity and this relationship is still pertinent even when adjusting for the different activities that people were engaged in. Ultimately, when people were more active, they were in a better mood. Additionally, when people were engaged in activities that they enjoy, this also led to a better mood. There still exists an independent effect that suggests a portion of the improved mood was derived from purely the physical activity component and some portion comes from the activities that are considered important and meaningful. This study also suggests that while people may experience increased positive affect and reduced fatigue after engaging in physical activity, people who had increased light physical activity also reported higher stress states. Results from this study highlight the value of engaging in meaningful activities to improve mood and reduce fatigue. These findings have implications for the timing of short-term interventions, such as just-in-time adaptive intervention approaches. For example, one direction that the findings from this study may serve to contribute to future interventions is through personalized communication which informs individuals about the types of occupations that may have improved their emotional well-being in the past based on their data. For healthcare practitioners who work directly with patients, it is suggested that physical activity recommendations consider the unique needs and characteristics of the patient, including which activities are important and meaningful for them. The promotion of physical activity should be part of an integrated approach to enhance meaningful occupations and healthy lifestyle behaviors. ## References 1. Azevedo MR, Araújo CLP, Reichert FF, Siqueira FV, da Silva MC, Hallal PC. Gender differences in leisure-time physical activity. International Journal of Public Health (2007.0) 52 8-15. DOI: 10.1007/s00038-006-5062-1 2. Bates D, Mächler M, Bolker B, Walker S. Fitting Linear Mixed-Effects Models Using lme4. Journal of Statistical Software (2015.0) 67 1-48. DOI: 10.18637/jss.v067.i01 3. Bell, B. (2013). 433–2013: A multilevel model primer using SAS® PROC MIXED, 19. http://www.cdc.gov/nchs/fastats/exercise.htm 4. Bennett BL, Whisenhunt BL, Hudson DL, Wagner AF, Latner JD, Stefano EC, Beauchamp MT. Examining the impact of social media on mood and body dissatisfaction using ecological momentary assessment. Journal of American College Health (2020.0) 68 502-508. DOI: 10.1080/07448481.2019.1583236 5. Berkman ET, Giuliani NR, Pruitt AK. Comparison of text messaging and paper-and-pencil for ecological momentary assessment of food craving and intake. Appetite (2014.0) 81 131-137. DOI: 10.1016/j.appet.2014.06.010 6. Bowden Davies KA, Sprung VS, Norman JA, Thompson A, Mitchell KL, Halford JCG, Harrold JA, Wilding JPH, Kemp GJ, Cuthbertson DJ. Short-term decreased physical activity with increased sedentary behaviour causes metabolic derangements and altered body composition: Effects in individuals with and without a first-degree relative with type 2 diabetes. Diabetologia (2018.0) 61 1282-1294. DOI: 10.1007/s00125-018-4603-5 7. Broderick JE, Schwartz JE, Schneider S, Stone AA. Can end-of-day reports replace momentary assessment of pain and fatigue?. The Journal of Pain (2009.0) 10 274-281. DOI: 10.1016/j.jpain.2008.09.003 8. Brønd JC, Arvidsson D. Sampling frequency affects the processing of Actigraph raw acceleration data to activity counts. Journal of Applied Physiology (2016.0) 120 362-369. DOI: 10.1152/japplphysiol.00628.2015 9. CDC/National Center for Health Statistics. (2021, June 11). Exercise or physical activity. centers for disease control and prevention. http://www.cdc.gov/nchs/fastats/exercise.htm 10. Craft BB, Carroll HA, Lustyk MK. Gender differences in exercise habits and quality of life reports: Assessing the moderating effects of reasons for exercise. International Journal of Liberal Arts and Social Science (2014.0) 2 65-76. PMID: 27668243 11. Crawford JR, Henry JD. The positive and negative affect schedule (PANAS): Construct validity, measurement properties and normative data in a large non-clinical sample. British Journal of Clinical Psychology (2004.0) 43 245-265. DOI: 10.1348/0144665031752934 12. Citko A, Górski S, Marcinowicz L, Górska A. Sedentary lifestyle and nonspecific low back pain in medical personnel in north-east Poland. BioMed Research International (2018.0) 2018 e1965807. DOI: 10.1155/2018/1965807 13. Curran PJ, Bauer DJ. The disaggregation of within-person and between-person effects in longitudinal models of change. Annual Review of Psychology (2011.0) 62 583-619. DOI: 10.1146/annurev.psych.093008.100356 14. Chomistek AK, Manson JE, Stefanick ML, Lu B, Sands LM, Going SB, Garcia L, Allison MA, Sims ST, LaMonte MJ, Johnson KC, Eaton CB. Relationship of Sedentary Behavior and Physical Activity to Incident Cardiovascular Disease. Journal of the American College of Cardiology (2013.0) 61 2346-2354. DOI: 10.1016/j.jacc.2013.03.031 15. Cong YJ, Gan Y, Sun HL, Deng J, Cao SY, Xu X, Lu ZX. Association of sedentary behaviour with colon and rectal cancer: A meta-analysis of observational studies. British Journal of Cancer (2014.0) 110 817-826. DOI: 10.1038/bjc.2013.709 16. Dunton GF, Berrigan D, Ballard-Barbash R, Graubard BI, Atienza AA. Social and physical environments of sports and exercise reported among adults in the American time use survey. Preventive Medicine (2008.0) 47 519-524. DOI: 10.1016/j.ypmed.2008.07.001 17. Dunton GF, Huh J, Leventhal AM, Riggs N, Hedeker D, Spruijt-Metz D, Pentz MA. Momentary assessment of affect, physical feeling states, and physical activity in children. Health Psychology: Official Journal of the Division of Health Psychology, American Psychological Association (2014.0) 33 255-263. DOI: 10.1037/a0032640 18. Gabriel KP, McClain JJ, Schmid KK, Storti KL, High RR, Underwood DA, Kuller LH, Kriska AM. Issues in accelerometer methodology: The role of epoch length on estimates of physical activity and relationships with health outcomes in overweight, post-menopausal women. International Journal of Behavioral Nutrition and Physical Activity (2010.0) 7 1. DOI: 10.1186/1479-5868-7-53 19. Hamrik Z, Sigmundová D, Kalman M, Pavelka J, Sigmund E. Physical activity and sedentary behaviour in Czech adults: Results from the GPAQ study. European Journal of Sport Science (2014.0) 14 193-198. DOI: 10.1080/17461391.2013.822565 20. Harris PA, Taylor R, Thielke R, Payne J, Gonzalez N, Conde JG. Research electronic data capture (REDCap)—A metadata-driven methodology and workflow process for providing translational research informatics support. Journal of Biomedical Informatics (2009.0) 42 377-381. DOI: 10.1016/j.jbi.2008.08.010 21. Healy GN, Dunstan DW, Salmon J, Cerin E, Shaw JE, Zimmet PZ, Owen N. Breaks in sedentary time: Beneficial associations with metabolic risk. Diabetes Care (2008.0) 31 661-666. DOI: 10.2337/dc07-2046 22. Jones M, Taylor A, Liao Y, Intille SS, Dunton GF. Real-time subjective assessment of psychological stress: Associations with objectively-measured physical activity levels. Psychology of Sport and Exercise. (2017.0) 31 79-87. DOI: 10.1016/j.psychsport.2017.03.013 23. Juvet LK, Thune I, Elvsaas IKØ, Fors EA, Lundgren S, Bertheussen G, Leivseth G, Oldervoll LM. The effect of exercise on fatigue and physical functioning in breast cancer patients during and after treatment and at 6 months follow-up: A meta-analysis. Breast Edinburgh Scotland (2017.0) 33 166-177. DOI: 10.1016/j.breast.2017.04.003 24. Kanning MK, Ebner-Priemer UW, Schlicht WM. How to investigate within-subject associations between physical activity and momentary affective states in everyday life: A position statement based on a literature overview. Frontiers in Psychology (2013.0) 4 187. DOI: 10.3389/fpsyg.2013.00187 25. Kincaid, C. D. (2005). 198–30: Guidelines for selecting the covariance structure in mixed model analysis, 8. 26. Laurenceau, J.-P. (2013). Using mobile devices to study daily diabetes management in marital context. NIDDK/NIH R21 DK098679. 27. Liao Y, Intille SS, Dunton GF. Using ecological momentary assessment to understand where and with whom adults’ physical and sedentary activity occur. International Journal of Behavioral Medicine (2015.0) 22 51-61. DOI: 10.1007/s12529-014-9400-z 28. Marquez DX, Aguiñaga S, Vásquez PM, Conroy DE, Erickson KI, Hillman C, Stillman CM, Ballard RM, Sheppard BB, Petruzzello SJ, King AC, Powell KE. A systematic review of physical activity and quality of life and well-being. Translational Behavioral Medicine (2020.0) 10 1098-1109. DOI: 10.1093/tbm/ibz198 29. Merwin RM, Dmitrieva NO, Honeycutt LK, Moskovich AA, Lane JD, Zucker NL, Surwit RS, Feinglos M, Kuo J. Momentary predictors of insulin restriction among adults with type 1 diabetes and eating disorder symptomatology. Diabetes Care (2015.0) 38 2025-2032. DOI: 10.2337/dc15-0753 30. Michie, S., West, R., Campbell, R., Brown, J., & Gainforth, H. (2014). ABC of behaviour change theories. Silverpack Publishing. 31. Oberoi S, Robinson PD, Cataudella D, Culos-Reed SN, Davis H, Duong N, Gibson F, Götte M, Hinds P, Nijhof SL, Tomlinson D, van der Torre P, Cabral S, Dupuis LL, Sung L. Physical activity reduces fatigue in patients with cancer and hematopoietic stem cell transplant recipients: A systematic review and meta-analysis of randomized trials. Critical Reviews in Oncology/hematology (2018.0) 122 52-59. DOI: 10.1016/j.critrevonc.2017.12.011 32. Paing AC, McMillan KA, Kirk AF, Collier A, Hewitt A, Chastin SFM. The associations of sedentary time and breaks in sedentary time with 24-hour glycaemic control in type 2 diabetes. Preventive Medicine Reports (2018.0) 12 94-100. DOI: 10.1016/j.pmedr.2018.09.002 33. Patterson R, McNamara E, Tainio M, de Sá TH, Smith AD, Sharp SJ, Edwards P, Woodcock J, Brage S, Wijndaele K. Sedentary behaviour and risk of all-cause, cardiovascular and cancer mortality, and incident type 2 diabetes: A systematic review and dose response meta-analysis. European Journal of Epidemiology (2018.0) 33 811-829. DOI: 10.1007/s10654-018-0380-1 34. Poole L, Steptoe A, Wawrzyniak AJ, Bostock S, Mitchell ES, Hamer M. Associations of objectively measured physical activity with daily mood ratings and psychophysiological stress responses in women. Psychophysiology (2011.0) 48 1165-1172. DOI: 10.1111/j.1469-8986.2011.01184.x 35. Pyatak EA, Hernandez R, Pham L, Mehdiyeva K, Schneider S, Peters A, Ruelas V, Crandall J, Lee P-J, Jin H, Hoogendoorn CJ, Crespo-Ramos G, Mendez-Rodriguez H, Harmel M, Walker M, Serafin-Dokhan S, Gonzalez JS, Spruijt-Metz D. Function and Emotion in Everyday Life with Type 1 Diabetes (FEEL-T1D): A fully remote intensive longitudinal study of blood glucose, function, and emotional well-being in adults with type 1 diabetes. JMIR Research Protocols (2021.0). DOI: 10.2196/30901 36. Pyatak EA, Sequeira PA, Whittemore R, Vigen CP, Peters AL, Weigensberg MJ. Challenges contributing to disrupted transition from paediatric to adult diabetes care in young adults with Type 1 diabetes. Diabetic Medicine (2014.0) 31 1615-1624. DOI: 10.1111/dme.12485 37. Razazian N, Kazeminia M, Moayedi H, Daneshkhah A, Shohaimi S, Mohammadi M, Jalali R, Salari N. The impact of physical exercise on the fatigue symptoms in patients with multiple sclerosis: A systematic review and meta-analysis. BMC Neurology (2020.0) 20 93. DOI: 10.1186/s12883-020-01654-y 38. Revelle, W. (2021). Procedures for psychological, psychometric, and personality research. Software. https://cran.rstudio.org/web/packages/psych/psych.pdf 39. Riddell MC, Gallen IW, Smart CE, Taplin CE, Adolfsson P, Lumb AN, Kowalski A, Rabasa-Lhoret R, McCrimmon RJ, Hume C, Annan F, Fournier PA, Graham C, Bode B, Galassetti P, Jones TW, Millán IS, Heise T, Peters AL, Laffel LM. Exercise management in type 1 diabetes: A consensus statement. The Lancet Diabetes & Endocrinology (2017.0) 5 377-390. DOI: 10.1016/S2213-8587(17)30014-1 40. Rothman, A. J., Baldwin, A. S., Hertel, A. W., & Fuglestad, P. T. (2011). Self-regulation and behavior change: Disentangling behavioral initiation and behavioral maintenance. In Handbook of self-regulation: Research, theory, and applications, 2nd ed (pp. 106–122). Guilford Press. 41. Saffer H, Dave D, Grossman M, Ann Leung L. Racial, ethnic, and gender differences in physical activity. Journal of Human Capital (2013.0) 7 378-410. DOI: 10.1086/671200 42. Sardinha LB, Magalhães JP, Santos DA, Júdice PB. Sedentary patterns, physical activity, and cardiorespiratory fitness in association to glycemic control in type 2 diabetes patients. Frontiers in Physiology (2017.0) 8 262. DOI: 10.3389/fphys.2017.00262 43. Schmid, D., & Leitzmann, M. F. (2014). Television viewing and time spent sedentary in relation to cancer risk: A meta-analysis. Journal of the National Cancer Institute,106(7). 10.1093/jnci/dju098 44. Scott SB, Ram N, Smyth JM, Almeida DM, Sliwinski MJ. Age differences in negative emotional responses to daily stressors depend on time since event. Developmental Psychology (2017.0) 53 177-190. DOI: 10.1037/dev0000257 45. Shiffman S, Stone AA, Hufford MR. Ecological momentary assessment. Annual Review of Clinical Psychology (2008.0) 4 1-32. DOI: 10.1146/annurev.clinpsy.3.022806.091415 46. Shiota MN, Papies EK, Preston SD, Sauter DA. Positive affect and behavior change. Current Opinion in Behavioral Sciences (2021.0) 39 222-228. DOI: 10.1016/j.cobeha.2021.04.022 47. Smith KE, Haedt-Matt A, Mason TB, Wang S, Yang C-H, Unick JL, Bond D, Goldschmidt AB. Associations between naturalistically assessed physical activity patterns, affect, and eating in youth with overweight and obesity. Journal of Behavioral Medicine (2020.0) 43 916-931. DOI: 10.1007/s10865-020-00152-3 48. Stone AA, Shiffman S, Schwartz JE, Broderick JE, Hufford MR. Patient non-compliance with paper diaries. BMJ (2002.0) 324 1193-1194. DOI: 10.1136/bmj.324.7347.1193 49. Stults-Kolehmainen MA, Sinha R. The effects of stress on physical activity and exercise. Sports Medicine (auckland, NZ) (2014.0) 44 81-121. DOI: 10.1007/s40279-013-0090-5 50. Subramanian, S., & Arun, B. (2017). Risk factor analysis in sedentary office workers with low back pain. 51. Troiano RP, Berrigan D, Dodd KW, Mâsse LC, Tilert T, Mcdowell M. Physical activity in the united states measured by accelerometer. Medicine & Science in Sports & Exercise (2008.0) 40 181-188. DOI: 10.1249/mss.0b013e31815a51b3 52. Van Cappellen P, Rice EL, Catalino LI, Fredrickson BL. Positive affective processes underlie positive health behaviour change. Psychology & Health (2018.0) 33 77-97. DOI: 10.1080/08870446.2017.1320798 53. Warburton DER, Bredin SSD. Health benefits of physical activity: A systematic review of current systematic reviews. Current Opinion in Cardiology (2017.0) 32 541-556. DOI: 10.1097/HCO.0000000000000437 54. Warburton DER, Nicol CW, Bredin SSD. Health benefits of physical activity: The evidence. CMAJ (2006.0) 174 801-809. DOI: 10.1503/cmaj.051351 55. Welk GJ, Kim Y. Context of physical activity in a representative sample of adults. Medicine and Science in Sports and Exercise (2015.0) 47 2102-2110. DOI: 10.1249/MSS.0000000000000641 56. Wen CKF, Liao Y, Maher JP, Huh J, Belcher BR, Dzubur E, Dunton GF. Relationships among affective states, physical activity, and sedentary behavior in children: Moderation by perceived stress. Health Psychology: Official Journal of the Division of Health Psychology, American Psychological Association (2018.0) 37 904-914. DOI: 10.1037/hea0000639 57. Wiese CW, Kuykendall L, Tay L. Get active? A meta-analysis of leisure-time physical activity and subjective well-being. The Journal of Positive Psychology (2018.0) 13 57-66. DOI: 10.1080/17439760.2017.1374436 58. Whitfield GP, Hyde ET, Carlson SA. Participation in leisure-time aerobic physical activity among adults, national health interview survey, 1998–2018. Journal of Physical Activity and Health (2021.0) 18 S25-S36. DOI: 10.1123/jpah.2021-0014 59. Yang C-H, Zink J, Belcher BR, Kanning M, Dunton GF. Age-varying bi-directional associations between momentary affect and movement behaviors in children: Evidence from a multi-wave ecological momentary assessment study. Annals of Behavioral Medicine: A Publication of the Society of Behavioral Medicine (2021.0) 55 918-931. DOI: 10.1093/abm/kaaa124
--- title: 'Seeking lifestyle counselling at primary health care centres: a cross-sectional study in the Swedish population' authors: - Frida Lundin Gurné - Per-Arne Svensson - Ida Björkman - Eva Lidén - Sofie Jakobsson journal: BMC Primary Care year: 2023 pmcid: PMC10026786 doi: 10.1186/s12875-023-02035-3 license: CC BY 4.0 --- # Seeking lifestyle counselling at primary health care centres: a cross-sectional study in the Swedish population ## Abstract ### Background Millions of people follow an unhealthy lifestyle in terms of tobacco consumption, hazardous use of alcohol, poor eating habits, and insufficient physical activity. Healthy lifestyles can to a large extent prevent and/or delay progression of non-communicable diseases. Factors influencing persons health-seeking behaviour regarding unhealthy lifestyles are of importance for sustainable health-promotive and disease-preventive work in primary health care. Generally, lifestyle interventions within primary health care are seen as feasible, but rarely reach all members of the general population. Few studies have been conducted about the likelihood among the general population to voluntarily contact a primary health care centre for support regarding lifestyle changes. The present study therefore aimed to investigate the general population’s likelihood of contacting a primary health care centre regarding their lifestyles, and factors associated with a lower such likelihood. ### Methods A probability sample of adults living in Sweden ($$n = 3$$ 750) were invited to participate in a cross-sectional survey regarding how societal developments affect attitudes and behaviours of the adult Swedish population. Data were collected between September and December 2020. Participants completed a questionnaire about lifestyle changes, and the data were analysed using descriptive statistics, Chi-square test and logistic regression analysis. ### Results The response rate was $52.0\%$ ($$n = 1$$ 896). Few persons responded that they would be likely to contact a primary health care centre for support regarding their lifestyles. Factors predicting a lower likelihood of contacting primary health care included few yearly visits to a primary health care centre, male sex, and living in a rural area. ### Conclusions Primary health care centres are not the first choice for lifestyle counselling for the majority of adults living in Sweden. We have identified factors predicting low likelihood of using the support available at these centres. In order to work with sustainable and visible health-promotive and disease-preventive strategies at primary health care centres, these settings need to find valid methods to involve and collaborate with the members of the general community, to meet the needs of a population struggling with unhealthy lifestyles. ## Background Lifestyle risk factors such as tobacco consumption, high consumption of alcohol, unhealthy eating habits, and insufficient physical activity lead to major global health problems. These risk factors make a large contribution to non-communicable diseases (NCDs) such as cardiovascular diseases, cancer, chronic respiratory diseases, and type 2 diabetes [1]. Healthy lifestyles can to a large extent prevent NCDs and/or delay progression of NCDs [1, 2], promote self-assessed general health [1], and reduce the burden on the health care budget [3]. Sweden, like many other Western countries, is facing increasing rates of NCDs [4], in which approximately one-fifth of the overall care burden can be attributed to unhealthy lifestyles [5]. Offering specific lifestyle interventions for members of the general community can contribute to improved lifestyles in a population [6, 7]. In a Swedish study Brobeck et al. [ 8] reported that approximately $40\%$ of the patients changed their lifestyles when health care professionals working in primary health care addressed lifestyle changes with supplementary advice in patient encounters, however with an uncertainty about the long-lasting effects. Furthermore, the EUROPREVIEW showed that most healthy patients without any diagnosis of lifestyle diseases such as cardiovascular diseases or type 2 diabetes wanted to visit health care professionals once a year or more often to check for example their blood pressure [9]. Primary health care has been proposed to be well placed to provide lifestyle counselling aimed at supporting the general population to adopt healthier lifestyles [10, 11]. However, such interventions may not reach the people in most need if they have not yet developed chronic diseases, or if they do not attend such clinical care settings [12]. Health-seeking behaviour within primary health care is influenced by several factors. For example, persons with low health literacy can experience greater difficulties in finding the ‘right’ care pathway [13, 14]. Smoking, regular alcohol consumption and physical inactivity are other factors that can decrease the probability of seeking primary health care [12]. Additionally, persons with risk factors for, for example cardiovascular diseases and/or obesity are more incline to seek care for weight loss in comparison with persons having overweight without risk factors for cardiovascular diseases [15]. Furthermore, a study published in 2018 showed that only a minority of persons with obesity sought help for weight loss the past year. Instead, they preferred self-guided efforts to manage ill health [16]. Another study reported that approximately $10\%$ of overweight/obese adults sought care from either dieticians/nutritionist, personal trainer or doctors to lose weight [17]. However, in contrast, persons with obesity or overweight have reported a need of more help with weight management from physicians in primary health care [18]. Two population-based studies among British citizens showed that being a smoker was associated with reduced likelihood of seeking help for symptoms linked to their unhealthy lifestyle habit such as cough [19, 20], also the complex phenomena of lung cancer stigma delayed the intension to seek help [20]. In Sweden, several initiatives have been launched towards health-promotive and disease-preventive practices such as national target goals for public health in 2003 [21]. Additional plans for action were in 2011 and 2018 when national clinical guidelines for unhealthy lifestyles were released. These guidelines particularly addressed that health care professionals should support the general population to change tobacco- and alcohol consumption, to change eating habits and to increase levels of physical activity [5, 22]. There are more than 1100 primary health care centres available where different health care professionals work such as general practitioners, registered nurses, district nurses, physiotherapists, and psychologists. Since 2010, each member in the general population is free to select a primary health care provider of their choice [23]. It is of importance to address to which extent members of the general population seek community care for their lifestyles in order to support lifestyle changes among members of the general population. This article presents data from a Swedish survey conducted among persons aged between 16 and 85 years. The aim was to investigate the general population’s likelihood of contacting a primary health care centre regarding their lifestyles, and factors associated with a lower such likelihood. ## Study design and participants A cross-sectional survey was conducted with a nationally representative sample of persons living in Sweden. This was part of a large national survey conducted by the SOM Institute (society, opinion and media) at the University of Gothenburg, which aimed to investigate how societal developments affect the attitudes and behaviours of the adult Swedish population [24]. The total sample consisted of 3 750 invited participants aged between 16 and 85 years. Potential participants were identified through a probability sample provided by the Swedish Tax Authority from a pool which included all persons residing in Sweden at the end of August 2020. ## Data collection The survey was sent to the participants’ home addresses by standard mail between September and December 2020, and reminders were sent by mail and text messages. The survey could be completed either by answering a paper-based questionnaire or by logging on to a digital platform using a code. The data collection period was 98 days, and the selected respondents could withdraw their participation at any time during this period. A note of thanks containing a lottery ticket to the value of 50 SEK (≈ 5 EUR) was sent to the participants who returned a completed questionnaire. ## Questionnaire Likelihood of contacting a primary health care centre for support for lifestyle changes was assessed using a 4-point Likert scale regarding each of the four lifestyles investigated in this study: smoking habits, alcohol consumption, eating habits, and physical activity. The question was: “How likely is it that you would contact a primary health care centre if you needed support to change any of the following lifestyles?” with response options; “very likely”, “fairly likely”, “not likely”, and “not at all likely”. Number of visits to a primary health care centre during the last year was measured on a 5-point scale, with answer alternatives “none”, “one or two”, “three or four”, “five or six” and “seven or eight visits or more”. The responses to this question were stratified into “none”, “one or two”, “three or four”, and “five or more”. Participants were asked to self-report both their health and their perceived need to change any of the four lifestyles. Self-reported health was measured using a scale ranging from 0 to 10, with higher scores indicating better health, and stratified in the analysis into 0–4 = poor, 5–6 = fairly good, 7–8 = good, and 9–10 = very good health. To report need to change any of the four lifestyles, the participants could answer “yes” or “no” for smoking habits, alcohol consumption, eating habits, and physical activity, respectively. Participants’ self-reported sex, age, education level, living area, and total household income were also used. Responses regarding age were stratified into four groups: 16–29 years, 30–49 years, 50–64 years, and 65–85 years. Participants reporting total household income of < 300 000 SEK per year were classified as low-income, those reporting 300 000–700 000 SEK were classified as middle-income, and those reporting > 700 000 SEK were classified as high-income. ## Statistical analysis Descriptive statistics were used to describe sociodemographic characteristics and health-related factors. Likelihood of contacting a primary health care centre was stratified into “likely” (comprising the response alternatives “very likely” and “fairly likely”) and “not likely” (comprising the response alternatives “not likely” and “not at all likely”). The probability of answering likely to seek support from primary health care for lifestyle counselling was analysed in subgroups of perceived need to change that specific lifestyle habit using Chi-square test. Logistic regression was used to calculate the odds ratio (OR) for having lower likelihood of contacting a primary health care centre regarding lifestyle changes, with certain sociodemographic characteristics and health-related factors as predictors. P-values < 0.05 were considered statistically significant, and no correction for multiple testing was performed. All data analyses were performed using version 27.0 of IBM SPSS Statistics (IBM Corp., Armonk, NY, USA). No missing values were imputed for the analysis. For the main question about likelihood of contacting a primary health care centre regarding lifestyle changes, some respondents did not answer that question, thus some data were missing, namely 99, 93, 69, and 80 persons respectively, for the four lifestyles (smoking habits, alcohol consumption, eating habits, and physical activity). ## Results Of the 3 750 questionnaires that were distributed, 1 896 valid responses were returned, giving a response rate of $52.0\%$. Sociodemographic characteristics and health-related factors are presented in Table 1. Among the 1 896 participants, 1 837 answered the question about sex; $48.0\%$ were men and $52.0\%$ were women. The most common educational level ($34.2\%$) was a post-secondary education of more than 3 years, the most common place of residence ($48.9\%$) was a city (but not a “major city”; see Table 1), and the most common total household income bracket was middle-income ($42.4\%$). Nearly half of the participants rated their self-reported health as good ($46.5\%$). The most common reported frequency of visits to a primary health care centre in the past years was one or two ($41.5\%$). Of the four lifestyles, physical activity was the one most commonly identified by the participants as in need of change ($54.7\%$).Table 1Sociodemographic characteristics and health-related factors among the sampleVariableTotal ($$n = 1$$ 896)n%Sex Men88248.0 Women95552.0 Missinga59Age (years) 16–2928915.2 30–4954828.9 50–6446624.6 65–8559331.3 Missinga0Education Primary education/school26514.5 Upper secondary education53029.0 Post-secondary education of < 3 years40722.3 Post-secondary education of > 3 years62634.2 Missinga68Living area Major cityb33818.4 City89848.9 Town34418.7 Rural area25714.0 Missinga59Total household incomec Low39022.4 Middle73742.4 High61335.2 Missinga156Self-reported health Poor1518.2 Fairly good26714.5 Good85646.5 Very good56830.8 Missinga54Number of visits to a primary health care centre in the past year None37920.5 One or two76941.5 Three or four39621.4 Five or more30916.7 Missinga43Need to change lifestylesd Smoking habits1397.7 Alcohol consumption21511.9 Eating habits80043.7 Physical activity100254.7a No / unclear answersbStockholm, Gothenburg, Malmöc Low income = up to 300 000 SEK; middle income = 300 000–700 000 SEK; high income = > 700 000 SEK (1 SEK ≈ 0.10 EUR on 22 November 2021). d Participants may have given several response alternatives Figure 1 shows the reported likelihood of contacting a primary health care centre regarding lifestyle changes. For all four lifestyles, “not at all likely” was the most common answer, ranging from $52.5\%$ to $63.9\%$. When the responses were stratified as described earlier, the proportions of participants stating that they were “not likely” to contact a primary health care centre were $82.4\%$ for smoking habits, $80.7\%$ for alcohol consumption, $82.1\%$ for eating habits and $84.9\%$ for physical activity. In comparison between all four lifestyles, alcohol consumption had the highest percentage of participants who reported that they were “very likely” to seek support at a primary health care centre ($6.6\%$).Fig. 1Likelihood of contacting a primary health care centre regarding support for lifestyle changes among all persons included. Values are given in % Figure 2 shows participants who reported a need to change respective lifestyle habit in relation to the 4-point Likert scale of likelihood. For all four lifestyles, “not at all likely” was the most common answer, ranging from $43.0\%$ to $51.8\%$. When the responses were stratified, smoking habits hade the highest percentages of all four lifestyles of participants who reported that they were “likely” to seek support at a primary health care centre ($25.9\%$). Furthermore, Table 2 shows the stratified response alternatives divided by sociodemographic characteristics and health-related factors. Fig. 2Likelihood of contacting a primary health care centre regarding support for lifestyle changes among participants that perceived a need to change each lifestyle habit. Values are given in %Table 2Likelihood of contacting a primary health care centre for lifestyle changesSmoking habitsAlcohol consumptionEating habitsPhysical activityParticipants (n)$$n = 1$$ 797n = 1 803n = 1 827n = 1 816Likelihood (%)LikelyNot likelyLikelyNot likelyLikelyNot likelyLikelyNot likelySex Men12.987.115.484.615.384.712.687.4 Women21.878.222.777.320.479.617.382.7Age, yrs 16–2920.779.326.673.418.981.114.885.2 30–4918.181.920.879.214.685.410.189.9 50–6415.384.716.183.917.182.913.886.2 65–8517.382.717.083.021.278.821.079.0Education Primary14.785.315.085.019.680.419.980.1 Upper secondary16.183.916.483.617.083.014.086.0 Post-secondary < 3 yrs20.479.621.978.119.580.516.583.5 Post-secondary > 3yrs18.381.721.978.116.783.312.787.3Living area Major citya17.882.221.178.918.082.014.185.9 City17.582.520.379.719.180.915.784.3 Town19.280.819.680.419.280.817.882.2 Rural area15.684.414.485.613.087.012.487.6Total household income Low17.682.417.482.624.175.922.078.0 Middle17.882.219.880.218.082.015.284.8 High17.682.420.779.313.686.410.189.9Self-reported health Poor15.584.515.584.525.374.727.472.6 Fairly good16.783.316.883.218.481.617.682.4 Good16.883.219.180.917.182.913.686.4 Very good19.580.521.778.316.683.412.787.3PHCC visits last year None14.985.117.582.512.687.49.190.9 1 or 218.481.619.680.415.884.212.287.8 3 or 415.085.017.182.919.280.817.282.8 5 or more21.478.623.077.027.972.126.773.3Need to change lifestyleb Smoking habits25.974.1------ Alcohol consumption--16.483.6---- Eating habits----18.381.7-- Physical activity------16.583.5aStockholm, Gothenburg, MalmöbParticipants who answered “yes” to the question about the “need to change lifestyles regarding …”. PHCC = Primary health care centre. Values are given in %. Likely refers to the response alternatives “very likely” and “fairly likely”, not likely refers to the response alternatives “not likely” and “not at all likely” The likelihood of seeking support regarding lifestyle counselling from primary health care was significantly higher among the group with a perceived need versus no perceived need, for smoking habits ($$P \leq 0.008$$) and physical activity ($$P \leq 0.044$$). This was not the case however for alcohol consumption ($$P \leq 0.284$$) and eating habits ($$P \leq 0.704$$). The logistic regression analysis presented in Fig. 3 showed that male sex predicted lower likelihood of contacting a primary health care centre, irrespective of the lifestyles investigated (OR range: 1.50–1.84). With living in a major city as reference, living in a rural area was a significant predictor of being less likely to contact a primary health care centre regarding alcohol consumption, eating habits, and physical activity (OR: 1.96, 2.07, and 1.68, respectively). Compared with high consumption of primary health care (five or more visits per year), lower consumption predicted lower likelihood of contacting a primary health care centre for all of the lifestyles investigated (OR range: 1.64–4.30), and this finding was particularly strong for both eating habits and physical activity (OR 2.01–4.30). Perceived need to change lifestyle was a predictor for only smoking habits and not the other lifestyles. Fig. 3Logistic regression analysis for predicting who would be “not likely” to contact a primary health care centre regarding support for lifestyle changes, giving odds ratios (ORs) and $95\%$ confidence intervals (CIs). Bold letters indicate significant ORs ($p \leq 0.05$). * “ High” total household income: > 700 000 SEK per year. PHCC = primary health care centre ## Discussion Our main finding that only a minority of the participants considered seeking support from a primary health care centre for lifestyle changes does not correspond to the overall trust in health care services in Sweden. In a survey from 2020, $69\%$ of respondents reported that they had high or very high trust in healthcare in general. Regarding primary health care centres, $66\%$ reported high or very high trust, but the trust in care at hospitals was even higher, with $76\%$ of respondents reporting high or very high trust [25]. Two studies from Sweden found that trust in health care professionals made it easier to change lifestyles [26, 27]. Additionally, a population-based survey from 2016 showed that $97\%$ of the respondents were positive to discuss their lifestyles with health care professionals in order to receive adequate care and treatment [28]. In our study, smoking habits and alcohol consumption were the lifestyles that were ranked as most likely to seek support from a primary health care centre for, and particularly evident in persons reporting a need to change smoking habits. This could be interpreted as meaning that people consider these lifestyles as more of risk factors to cause long-term ill health, and therefore important to seek support for and discuss with health care professionals. Few yearly visits to a primary health care centre predicted a lower likelihood of seeking support from a primary health care centre for lifestyle changes. Reasons for this finding might include low visibility of primary health care centres in the community together with lack of awareness that these centres offer lifestyle counselling. Another possible reason could be that members of the general population contacts other actors in society for support regarding this issue. For example, they might contact a personal trainer or a health coach, or join a slimming club. Feng et al. [ 12] highlight alternative providers of lifestyle counselling for support in lifestyle changes. Altogether, it is likely that healthy lifestyles need to be approached from several angles in society, to help members of the general population to adopt healthier lifestyles. Earlier studies have shown that men are less likely to seek care [29, 30], and our study extends this finding by showing that it also applies to seeking lifestyle support in a primary health care setting. Novak et al. [ 31] suggest that men’s overall avoidance of health care might be related to their own perceptions of the male gender role as being tough and being able to push through pain, and that men’s health-seeking behaviour is affected by their perceptions of how helpful or knowledgeable a physician is. However, a Swedish study found that men were reported to be more likely than women to be asked about lifestyle issues in encounters with health care [8]. Further research is needed to explore gender disparities in health-seeking behaviour with regard to lifestyle counselling. Furthermore, we also found that living in a rural area predicted a lower likelihood of seeking support for lifestyle changes. The reason for this finding might be connected to accessibility in primary health care, which has been found to be essential in relation to patients’ needs of support in care [32]. A recent study from Sweden reported that rural areas and smaller cities have higher proportion of obese people in comparison to large cities [33], which suggests that accessibility of health care might play some role. The rising use of digital technology in society and health care implies a need to explore digital tools in lifestyle counselling, which may be more relevant for persons living in rural areas. A recent systematic review reported that digital self-monitoring behavioural interventions regarding physical activity and eating habits were effective in regard to supporting persons in weight loss, in comparison to interventions excluding digital self-monitoring [34]. Chatterjee et al. [ 35] found that successful digital interventions to promote healthier lifestyles was built upon several factors such as participants holding digital literacy and personalised feedback. Although persons in our study perceived a need to change their lifestyles, few persons reported that they would be likely to seek such support from a primary health care centre. This finding brings the reasoning to stigmatization which is a well-known barrier, for example regarding smoking habits in which smokers are less likely than non-smokers to seek primary health care [36]. This raises the question of whether primary health care centres are able to reach the persons who are in most need of changing their lifestyle in this respect [12]. There is a fine and time-consuming balance between removing the blame associated with smoking and encouraging persons to actively seek help when needed [37]. In addition, obesity is a characteristic that is often stigmatized in relation to unhealthy lifestyles in which physicians and nurses are known to hold negative attitudes about obese persons [38]. Research indicates that the attitudes from health care providers can cause feelings of disrespect or of not being welcomed in obese patients, thus negatively affecting both the specific encounter and the individual’s willingness to seek care [39]. The impact of stigma in health care may explain the low likelihood of seeking lifestyle counselling from primary health care even among respondents who reported a perceived need to change their lifestyles. For this reason, attention needs to be given to acknowledging stigmatization as a barrier, in order both to support lifestyle changes and to reach those in most need. However, from a broader societal perspective, healthy diet and regular physical activity should be seen as valuable factors for health and well-being for all members in society, not only for persons with obesity. In Sweden, structural efforts for increased health-promotive and disease-preventive work have been discussed, for example better routines in practice for lifestyle counselling along with an equality perspective for such tasks across the country, i.e. persons that seek support from a primary health care centre should be able to receive support to change lifestyles to the same extent regardless of which county council they live in [40]. A Swedish report from 2013 (two years after the implementation of the national clinical guidelines for unhealthy lifestyles), showed that reimbursement to caregivers can be roughly divided into three different categories: fixed, variable and special economic compensation. Fixed compensation refers to economic compensation within the framework of the basic mission of offering health-promotive and disease-preventive services for listed patients, for example annual compensation of 100 SEK per listed resident. Variable compensation is linked to specific clinical tasks that aims to change lifestyles, such as supporting patients to change eating habits, or prescribing physical activity on prescription. Also, variable economic compensation can be given when health care professionals have identified risk factors in patients, for example high blood pressure. Finally, some county councils can seek special economic compensation with the aim to develop the health-promotive and disease-preventive work, along with establishment of collaborations with other actors in society, to take greater responsibility for health preventive services in the surrounding area [41]. It is also important to highlight the individual expenses associated with a lifestyle counselling as the costs might be much higher if a person has initiated the lifestyle counselling themselves rather than if a health care professional have taken the initiative. ## Limitations and strengths Some limitations of our study need to be addressed. There were no significant differences in geographical composition of respondents compared to the general population as a whole [42]. But notably age distribution and education level differed from the general population as the age group 16–29 years were underrepresented and persons with high level of education were overrepresented. Another limitation was that the questions were not validity and/or reliability tested before the study was conducted, which can be seen as a limitation. Furthermore, we had no data from the participants regarding outcome measures such as weight or body mass index, and no information about their current lifestyles, in terms of, for example eating habits or smoking; this prohibited us from conducting more in-depth analyses. Further, the data collection took place during the COVID-19 pandemic, which may have influenced the response rate and the answers given. However, the response rate of approximately $50\%$ could be seen as a strength, as this is relatively high both for a postal survey and in comparison with international surveys [24]. Our study was conducted within the Swedish primary health care context, and so our findings may have limited generalizability to other countries due to differences between health care systems. ## Conclusion This study provides knowledge about the likelihood of members of the general community contacting a primary health care centre regarding support for lifestyle changes. We can conclude that primary health care centres are not the main choice for lifestyle counselling, and we have identified factors predicting low likelihood of using this support. In light of the growing need for lifestyle changes, primary health care centres need to find valid methods for engaging with and meeting the needs of a population struggling with unhealthy lifestyles. ## References 1. 1.World Health Organization: Global status report on non-comminicable diseases 2014. http://apps.who.int/iris/bitstream/handle/10665/148114/9789241564854_eng.pdf;jsessionid=6B5D1C629215F65F5ACB19033044E13F?sequence=1. (2014). Accessed 28 Mar 2022. 2. Lee IMD, Shiroma EJM, Lobelo FMD, Puska PMD, Blair SNPED, Katzmarzyk PTP. Effect of physical inactivity on major non-communicable diseases worldwide: an analysis of burden of disease and life expectancy. Lancet (2012.0) 380 219-29. DOI: 10.1016/S0140-6736(12)61031-9 3. Mattli R, Farcher R, Syleouni M-E, Wieser S, Probst-Hensch N, Schmidt-Trucksäss A. Physical activity interventions for primary prevention in adults: a systematic review of randomized controlled trial-based economic evaluations. Sports Med (2020.0) 50 731-750. DOI: 10.1007/s40279-019-01233-3 4. 4.World Health Organization: Noncommunicable diseases country profiles 2018. https://apps.who.int/iris/handle/10665/274512 (2018). Accessed Feb 17 2022. 5. 5.The national Board of Health and Welfare. National guidelines for prevention and treatment of unhealthy life habits: support for governance and management. (In Swedish) https://www.socialstyrelsen.se/globalassets/sharepoint-dokument/artikelkatalog/nationella-riktlinjer/2018-6-24.pdf 2018. Accessed 20 April 2022. 6. Waller M, Blomstrand A, Högberg T, Ariai N, Thorn J, Hange D. A primary care lifestyle programme suitable for socioeconomically vulnerable groups - an observational study. Scand J Prim Health Care (2016.0) 34 352-359. DOI: 10.1080/02813432.2016.1248628 7. Blomstrand A, Lindqvist P, Carlsson IE, Pedersen N, Bengtsson C. Low-budget method for lifestyle improvement in primary care. experiences from the göteborg health profile project. Scand J Prim Health Care (2005.0) 23 82-7. DOI: 10.1080/02813430510018482 8. Brobeck E, Bergh H, Odencrants S, Hildingh C. Lifestyle advice and lifestyle change: to what degree does lifestyle advice of healthcare professionals reach the population, focusing on gender, age and education?. Scand J Caring Sci (2015.0) 29 118-125. DOI: 10.1111/scs.12139 9. Brotons C, Bulc M, Sammut MR, Sheehan M, da Manuel Silva Martins C, Björkelund C. Attitudes toward preventive services and lifestyle: the views of primary care patients in Europe. the EUROPREVIEW patient study. Fam Pract (2012.0) 29 i168-i76. DOI: 10.1093/fampra/cmr102 10. Starfield B, Shi L, Macinko J. Contribution of primary care to health systems and health. Milbank Q (2005.0) 83 457-502. DOI: 10.1111/j.1468-0009.2005.00409.x 11. Wise M, Nutbeam D. Enabling health systems transformation: what progress has been made in re-orienting health services?. Promot Educ (2007.0) Suppl 2 23-7. DOI: 10.1177/10253823070140020801x 12. Feng X, Girosi F, McRae IS. People with multiple unhealthy lifestyles are less likely to consult primary healthcare. BMC Fam Pract (2014.0). DOI: 10.1186/1471-2296-15-126 13. von Wagner C, Knight K, Steptoe A, Wardle J. Functional health literacy and health-promoting behaviour in a national sample of British adults. J Epidemiol Community Health (2007.0) 61 1086-1090. DOI: 10.1136/jech.2006.053967 14. Berkman ND, Sheridan SL, Donahue KE, Halpern DJ, Crotty K. Low health literacy and health outcomes: an updated systematic review. Ann Intern Med (2011.0) 155 97-107. DOI: 10.7326/0003-4819-155-2-201107190-00005 15. Tol J, Swinkels IC, De Bakker DH, Veenhof C, Seidell JC. Overweight and obese adults have low intentions of seeking weight-related care: a cross-sectional survey. BMC Public Health (2014.0) 14 582. DOI: 10.1186/1471-2458-14-582 16. Evans EH, Sainsbury K, Kwasnicka D, Bolster A, Araujo-Soares V, Sniehotta FF. Support needs of patients with obesity in primary care: a practice-list survey. BMC Fam Pract (2018.0) 19 6. DOI: 10.1186/s12875-017-0703-4 17. de Heer HD, Kinslow B, Lane T, Tuckman R, Warren M. Only 1 in 10 patients told to lose weight seek help from a health professional: a nationally representative sample. Am J Health Promot (2019.0) 33 1049-1052. DOI: 10.1177/0890117119839904 18. Potter MB, Vu JD, Croughan-Minihane M. Weight management: what patients want from their primary care physicians. J Fam Pract (2001.0) 50 513-518. PMID: 11401737 19. Friedemann Smith C, Whitaker KL, Winstanley K, Wardle J. Smokers are less likely than non-smokers to seek help for a lung cancer 'alarm' symptom. Thorax (2016.0) 71 659-661. DOI: 10.1136/thoraxjnl-2015-208063 20. Carter-Harris L. Lung cancer stigma as a barrier to medical help-seeking behavior: Practice implications. J Am Assoc Nurse Pract (2015.0) 27 240-245. DOI: 10.1002/2327-6924.12227 21. 21.Ministry for Health and Social Affairs. Objectives for the publics’ health. (In Swedish) https://data.riksdagen.se/fil/4324E0A2-E80C-4D2A-B1A4-6951B487E7FE 2002. Accessed 10 Nov 2022. 22. 22.The National Board of Health and Welfare. National guidelines for disease preventive methods 2011. Tobbacco use, risky use of alcohol, insufficient physical activity and unhealthy eating habits. Support for governance and management 2011. (In Swedish). https://untobaccocontrol.org/impldb/wp-content/uploads/reports/sweden_annex5_national_guidelines_for_methods_of_preventing_disease_2011.pdf 2011. Accessed 11 Nov 2022. 23. Anell A, Glenngard AH, Merkur S. Sweden health system review. Health Syst Transit (2012.0) 14 1-159. PMID: 22894859 24. 24.SOM-Institute. Methods. (In Swedish) https://www.gu.se/som-institutet/som-undersokningarna/metod-0 2022. Accessed 13 May 2022. 25. 25.Swedish Association of Local Authorities and Regions: Health Care barometer 2020- Population attitudes to, confidence of and perception of health and health care. (In Swedish) https://skr.se/download/18.5627773817e39e979ef386db/1642164568072/7585-916-3.pdf 2020. Accessed 10 Apr 2022. 26. Nymberg P, Drevenhorn E. Patients' experience of a nurse-led lifestyle clinic at a Swedish health centre. Scand J Caring Sci (2016.0) 30 349-355. DOI: 10.1111/scs.12254 27. Brobeck E, Odencrants S, Bergh H, Hildingh C. Patients' experiences of lifestyle discussions based on motivational interviewing: a qualitative study. BMC Nurs (2014.0) 13 13. DOI: 10.1186/1472-6955-13-13 28. 28.The National Board of Health and Welfare. This is how patients want to tell health care about their lifestyles – results of a population survey in 2016. (In Swedish) https://www.socialstyrelsen.se/globalassets/sharepoint-dokument/artikelkatalog/ovrigt/2016-12-13.pdf (2016). Accessed 15 Nov 2022. 29. Thompson AE, Anisimowicz Y, Miedema B, Hogg W, Wodchis WP, Aubrey-Bassler K. The influence of gender and other patient characteristics on health care-seeking behaviour: a QUALICOPC study. BMC Fam Pract (2016.0). DOI: 10.1186/s12875-016-0440-0 30. Adamson J, Ben-Shlomo Y, Chaturvedi N, Donovan J. Ethnicity, socio-economic position and gender—do they affect reported health—care seeking behaviour?. Soc Sci Med (2003.0) 57 895-904. DOI: 10.1016/S0277-9536(02)00458-6 31. Novak JR, Peak T, Gast J, Arnell M. Associations Between Masculine Norms and Health-Care Utilization in Highly Religious, Heterosexual Men. Am J Mens Health (2019.0) 13 1557988319856739-. DOI: 10.1177/1557988319856739 32. Papp R, Borbas I, Dobos E, Bredehorst M, Jaruseviciene L, Vehko T. Perceptions of quality in primary health care: perspectives of patients and professionals based on focus group discussions. BMC Fam Pract (2014.0) 15 128. DOI: 10.1186/1471-2296-15-128 33. 33.The Public Health Agency. Open comparison public health 2019. (In Swedish). https://www.folkhalsomyndigheten.se/contentassets/ec714fca0b0145eab3d7924511550a74/oppna-jamforelser-folkhalsa-2019-18076.pdf (2019). Accessed 2 Nov 2022. 34. Berry R, Kassavou A, Sutton S. Does self-monitoring diet and physical activity behaviors using digital technology support adults with obesity or overweight to lose weight? A systematic literature review with meta-analysis. Obes Rev (2021.0) 22 e13306. DOI: 10.1111/obr.13306 35. Chatterjee A, Prinz A, Gerdes M, Martinez S. Digital Interventions on healthy lifestyle management: systematic review. J Med Internet Res (2021.0) 23 e26931. DOI: 10.2196/26931 36. Jorm LR, Shepherd LC, Rogers KD, Blyth FM. Smoking and use of primary care services: findings from a population-based cohort study linked with administrative claims data. BMC Health Serv Res (2012.0). DOI: 10.1186/1472-6963-12-263 37. Scott N, Crane M, Lafontaine M, Seale H, Currow D. Stigma as a barrier to diagnosis of lung cancer: patient and general practitioner perspectives. Prim Health Care Res Dev (2015.0) 16 618-622. DOI: 10.1017/S1463423615000043 38. Puhl RM, Heuer CA. The stigma of obesity: a review and update. Obesity (Silver Spring) (2009.0) 17 941-964. DOI: 10.1038/oby.2008.636 39. Phelan SM, Burgess DJ, Yeazel MW, Hellerstedt WL, Griffin JM, van Ryn M. Impact of weight bias and stigma on quality of care and outcomes for patients with obesity. Obes Rev (2015.0) 16 319-326. DOI: 10.1111/obr.12266 40. 40.The National Board of Health and Welfare. National guidelines – evaluation 2014. Disease prevention methods. Indicators and basis for assessments in 2014. (In Swedish). https://www.socialstyrelsen.se/globalassets/sharepoint-dokument/artikelkatalog/nationella-riktlinjer/2015-1-11.pdf (2014). Accessed 15 Jan 2023. 41. 41.Helene Johansson. Public Health Report Series 2014:1. Economic control systems and disease prevention efforts in primary health care. A summary of the county councils/regions’ regulations for the year 2013. (In Swedish). https://umu.diva-portal.org/smash/get/diva2:714549/FULLTEXT01.pdf 42. 42.SOM Institute. The national SOM survey 2020 – a method overview 2021. (In Swedish). https://www.gu.se/sites/default/files/2021-03/2.%20Metodrapport%20Riks%202020.pdf. (2021). Accessed 12 Nov 2022.
--- title: 'Diabetes type 2: relationships between lysosomal exocytosis of circulating normal-sized platelets and in vitro ɑ-thrombin-evoked platelet responses' authors: - Maria Edvardsson - Magnus Oweling - Petter Järemo journal: Annals of Medicine year: 2023 pmcid: PMC10026805 doi: 10.1080/07853890.2023.2171108 license: CC BY 4.0 --- # Diabetes type 2: relationships between lysosomal exocytosis of circulating normal-sized platelets and in vitro ɑ-thrombin-evoked platelet responses ## Abstract ### Background/objective Type 2 diabetes is a major risk factor for atherosclerotic disease. It is well agreed that the reactivity of diabetic platelets is increased but how platelet reactivity regulates is unknown. In our laboratory, density separated platelets have been investigated extensively and high- and low-density platelets circulate in an activated state. The density distribution of circulating platelets is altered in diabetes type 2 as well. We hypothesize that such platelets modify whole blood (WB) in vitro α-thrombin-evoked (10 μM/mL) activity in type 2 diabetes. Thus, the study aims to identify features of circulating normal-sized density subpopulations affecting whole blood (WB) platelet reactivity in type 2 diabetes. ### Patients and methods Patients with type 2 diabetes ($$n = 16$$) were enrolled. Their normal-sized platelets were divided into density subfractions ($$n = 16$$) using continuous polyvinylpyrrolidone-coated silica (Percoll™) gradients (density span, 1.090–1.040 kg/L) containing prostaglandin E1. The proportions (%) of such density-separated platelets expressing lysosomal-associated membrane protein 1 (LAMP-1) were analyzed using a flow cytometer. Further, determinations of WB ɑ-thrombin-evoked (10 U/mL) surface LAMP-1 (an assessment of lysosomal release), the fibrinogen (αIIbβ3) receptor activity, annexin V (binds to exposed membrane phosphatidylserine), and mitochondrial transmembrane potentials (an estimate of organelle integrity) were performed. Surface LAMP-1 expressions of individual normal-sized platelet density subpopulations were stratified into equal-sized groups ($$n = 2$$) depending on reactivity, as judged from the ɑ-thrombin-induced WB activity markers. ### Results With some exceptions, the proportion of normal-sized circulating platelets showing spontaneous LAMP-1 was strongly associated with WB ɑ-thrombin-evoked (10 U/mL) surface LAMP-1 and αIIbβ3 receptor activity. LAMP-1-expressing normal-sized platelets also displayed inverse associations with WB ɑ-thrombin-induced surface annexin V and mitochondrial damage, which are features of procoagulant platelets. ### Conclusions From the current descriptive work only involving type 2 diabetes, it is impossible to judge whether the findings are features of the disease or if they occur in healthy individuals as well. However, the study describes LAMP-1 expressing subpopulations of circulating normal-sized platelets that associate with WB α-thrombin (10 U/mL) responses in vitro. Increased proportions of such platelets induced lysosomal release and αIIbβ3 receptor activity, whereas lower proportions promoted WB agonist-induced procoagulant platelet creation. It is to hypothesize that the new described regulatory mechanism could in the future offer a possibility to influence platelet behavior in type 2 diabetes. Key messagesLysosomal exocytosis of circulating platelets influences reactivity, as determined by agonist-induced platelet reactions in vitroThus, the low release of lysosomes by normal-sized platelets in vivo increases agonist-evoked procoagulant platelet production. Higher lysosomal exocytosis of circulating normal-sized platelets promotes platelet aggregation and secretion. ## Introduction Platelets were deemed to be either resting or activated, and when activated, they were assumed to perform similar tasks. Recently, the binary view has been abandoned, and the concept that platelet functions are unevenly distributed has emerged [1,2]. Thus, if activated, some platelets discharge lysosomal constituencies, as evidenced by lysosomal-associated membrane protein 1 (LAMP-1), whereas others activate fibrinogen receptors (αIIbβ3), thereby accomplishing aggregation. Furthermore, during activation, other platelets generate procoagulant corpuscles exposing phosphatidylserine (PS), as estimated from elevated surface annexin V levels, and increased mitochondrial injury [3–5]. PS enables the accumulation of surface prothrombinase complexes and accelerates platelet thrombin generation [6]. The mechanisms that steer the activation routes remain unknown. Recently, interest in platelet heterogeneity has increased, and the diversity of non-activated in vitro density-separated platelets has been studied by us [7–9]. Platelet density follows a Gaussian dispersion and varies within the range of 1.090–1.040 kg/L [10–12]. High-density platelets contain more granules (both ɑ- and dense) and mitochondria [10–12] and judging from fibrinogen expression, high- and low-density corpuscles circulate, showing a higher activation than the bulk of platelets [8,9]. Procoagulant platelets are clinically important [13,14] and many human diseases alter platelet diversity. Examples include tiny subfractions of larger platelets that forecast prostatic cancer relapse [15] and subpopulations of circulating Alzheimer platelets expressing less membrane-bound fibrinogen [16]. Furthermore, platelet density increases in conjunction with acute myocardial infarctions [17] and is then inversely linked to inflammatory reactions [18]. Moreover, high, and low platelet densities reflect inflammatory bowel disease activity [19] and preeclampsia severity [20], respectively. Type 2 diabetes enhances atherothrombotic risk and diabetic platelets demonstrate a higher baseline activity [21,22]. They also show increased activity both at baseline and in response to agonists such as ADP and α- thrombin [21–24]. Thus, impaired aspirin efficacy has been reported in conjunction with type 2 diabetes [25]. Multicolor flow cytometry is a powerful tool in platelet research. It can be used for low-platelet-count samples and is thus suitable for analyzing platelet subpopulations. The device separates corpuscles according to their relative size [4,26] and permits concurrent analysis of many activation markers. The current basic science study used this technique [4] and examined type 2 diabetes, i.e. a prothrombotic condition. The relationship between the activity of density-separated platelet subpopulations and whole blood (WB) ɑ-thrombin induced (10 U/mL) lysosomal release, αIIbβ3 receptor activity, and procoagulant platelet creation. Therefore, this study aimed to elucidate whether lysosomal exocytosis and fibrinogen receptor activities of circulating normal-sized platelet subpopulations influence WB ɑ-thrombin-induced platelet activity in type 2 diabetes. ## Ethics statements, patients, and blood sampling After ethical approval (Regionala Etikprövningsnämnden, Medicinska Fakultetens Kansli, Linköpings Universitet, SE-581 83 Linköping (registration number $\frac{2018}{54}$-31)) patients with type 2 diabetes ($$n = 16$$) were recruited (Table 1). The study adhered to the principles of Helsinki and written informed consent was obtained from all participants. A well-founded diagnosis of type 2 diabetes was the only inclusion criterion. There were no exclusion criteria. This disease was studied because it is a risk factor for thrombotic events. The patients joined a clinical control in primary care and were enrolled in 2020 when laboratory resources were available. Ethylenediamine tetra acetic acid (EDTA) anticoagulated WB (3 mL) was sampled for routine laboratory analysis (Table 1). For functional flow cytometry studies, WB (9 mL) was collected from the antecubital vein into $3.2\%$ sodium citrate tubes and immediately transferred to an inhibitory cocktail [7,26,27] containing equal volumes of the following stock solutions:0.13 M of Na2EDTA and 0.15 M of Na2Citrate (pH 7.4 at 25 °C),2.7 mM of theophylline dissolved in 150 mM of TRIS chloride buffer (pH 7.4 at 25 °C), and1 mg/L of prostaglandin E1 dissolved in $95\%$ (w/v) ethanol. Table 1. \| Male/female (n) \| 9/7 \| \| --- \| --- \| \| Age (years) \| 70 ± 11 (SD) \| \| Body weight (kg) \| 84 ± 11 (SD) \| \| Duration of diabetes (years) \| 10 ± 5 (SD) \| \| Previous myocardial disease (%) \| 25 \| \| Previous cerebral disease (%) \| 12 \| \| Insulin (%) \| 33 \| \| Metformin (%) \| 73 \| \| Other oral antidiabetic drugs (%) \| 40 \| \| β-Blockers (%) \| 53 \| \| Diuretics (%) \| 47 \| \| Ca2+-blockers (%) \| 53 \| \| Aspirin (%) \| 40 \| \| ADP receptor-blockers (%) \| 7 \| \| ACE or A2-inhibitors (%) \| 53 \| \| Statins (%) \| 75 \| \| Platelet count (×109/L) \| 242 ± 64 (SD) \| \| Neutrophil count (×109/L) \| 4.1 ± 0.8 (SD) \| \| Hemoglobin A1c level (mmol/L) \| 57 ± 15 (SD) \| \| Creatinine level (μmol/L) \| 78 ± 25 (SD) \| Then, this mixture was used for platelet density separation [7,26,27]. Platelet fractionation and flow cytometer studies commenced approximately 120 min after sampling. ## Platelet density fractionation Previous studies have described how platelet density varies between 1.090 and 1.040 kg/L [10–12]. Thus, linear polyvinylpyrrolidone-coated silica gradients covering that span were used to separate the platelets [7]. To avoid activation in the test tubes, the gradients contained EDTA, prostaglandin E1, and theophylline. After centrifugation, the platelet population (i.e. the gradient) was split by gravidity into density subpopulations ($$n = 16$$) [7]. Thus, high-density corpuscles were in low-digit populations, so higher numbers denoted less-dense platelets. ## Flow cytometry A Gallios flow cytometer (Beckman Coulter, Brea, CA) equipped with a multicolor design and three lasers (405, 488, and 638 nm) was used. The following platelet features were determined: (a) platelet size, normal-sized platelets, small platelets, and vesicles; (b) lysosomal exocytosis, surface-attached LAMP-1; (c) fibrinogen receptor (αIIbβ3) activity, surface αIIbβ3 activation specific antibody (PAC-1); (d) platelet surface phosphatidylserine surface-bound annexin V; and (e) mitochondrial transmembrane potentials, retention of 1,1′,3,3,3′,3′-hexamethyl-indodicarbo-cyanine iodide (DiIC1[5]). A flow cytometry protocol, first described by Ramström’s research group, was used without substantial alterations [4]. The antibodies and probes used are listed in Table 2. Platelets were recognized by forward scatter (size) and GPIIb receptor fluorescence, and normal-sized populations, as determined by flow cytometry gating, were studied. The probes and antibodies were added to the density subpopulations (Table 2) as described previously [4,26] and we did not extract platelets from the polyvinylpyrrolidone-coated silica. Gating has been extensively explained elsewhere [26] but is also shown in Figure 1. It demonstrates an individual sample after α-thrombin (10 U/mL) provocation and shows the gating of the GPIIb positive populations (normal-sized platelets, small platelets, and vesicles). In all samples, the proportions of positive corpuscles (%) were evaluated, and mean fluorescence intensities were not determined. The management of different controls has been described in detail [4]. After 10 min at room temperature the reactions were stopped by dilution in HEPES-Ca2+. Flow cytometry particle acquisition ended either after counting >5000 corpuscles or after 2 min. Thus, the number of assessed particles differed depending on the subpopulation corpuscle count. Figure 1.: Shows for patient no. 16 whole blood size-dependent platelet populations after ɑ-thrombin (10 U/mL) provocation as determined by flow-cytometry. The illustration demonstrates the size-dependent platelet subfractions (normal-sized, small platelets, and vesicles) together with manual the gating of the device. TABLE_PLACEHOLDER:Table 2. ## Data assessment and analysis During the assessment, surface LAMP-1 was used as a sign of platelet lysosomal release, and PAC-1 indicated aggregatory corpuscles. Membrane-attached annexin V and mitochondrial damage identified procoagulant populations [4,26]. The flow cytometer recognized normal-sized platelets in density subpopulations ($$n = 16$$) [4,26]. For all patients ($$n = 16$$) and in all fractions ($$n = 16$$), the ratios (%) of normal-sized corpuscles that were positive for LAMP-1 were analyzed. This procedure was repeated for the activated αIIbβ3 receptor (PAC-1). Subsequently, WB activity markers LAMP-1 (Figure 2), PAC-1 (Figure 3), annexin V (Figure 4), and DiILC1[5] (Figure 5) were determined after α-thrombin (10 U/mL) provocation in a separate sample. For all participants ($$n = 16$$), means and standard deviations of the ratio (%) (LAMP-1 or PAC-1) of neighboring ($$n = 4$$) density populations were calculated. Then, means were arranged into groups ($$n = 2$$) consisting of patients (each $$n = 8$$) with lower and higher in vitro ɑ-thrombin-induced WB activity as aforementioned. The unpaired Student t-test (Excel, Microsoft, Redmond, WA,) was used to perform statistical analysis. A p-value <0.05 was considered statistically significant. Figure 2.: Surface LAMP-1 (left) and PAC-1 (right) expressions (mean ± SD) of circulating density-separated normal-sized platelets divided into low- and high-responders according to WB LAMP-1 expression after ɑ-thrombin provocation (10 U/mL). The colors indicate statistical significance (red, p < 0.01; yellow, p < 0.05; black, not significant) between the pairs. For each pair, the left bar summarizes WB ɑ-thrombin low-responders (L), and the right bar shows high-responding participants (H). A–D denote the density intervals. A: density subpopulations nos. 1–4; density span 1.090–1.079 kg/L. B: density subpopulations nos. 5–8; density span 1.079–1.067 kg/L. C: density subpopulations nos. 9–12; density span 1.067–1.054 kg/L. D: density subpopulations nos. 13–16; density span 1.054–1.040 kg/L. LAMP-1: lysosomal-associated membrane protein 1; nos.: numbers; PAC-1: fibrinogen receptor (αIIbβ3) activity; SD: standard deviation; WB: whole blood. The number (mean ± SD) of normal-sized platelets evaluated using flow cytometry for the density fractions (nos. 1–16): (1) 87 ± 95; (2) 320 ± 311; (3) 594 ± 688; (4) 1244 ± 1812; (5) 2578 ± 1812; (6) 4631 ± 904; (7) 4814 ± 844; (8) 5138 ± 118; (9) 5261 ± 158; (10) 5305 ± 212; (11) 5139 ± 871; (12) 3831 ± 1738; (13) 2718 ± 1602; (14) 2053 ± 1813; (15) 1585 ± 1296; (16) 1012 ± 605. Figure 3.: Membrane LAMP-1 (left) and PAC-1 (right) (mean ± SD) of density-separated normal-sized platelets split according to WB PAC-1 responses after ɑ-thrombin provocation (10 U/mL). The colors denote significance (red, p < 0.01; yellow, p < 0.05; black, not significant). For each pair, the left and right bars display WB ɑ-thrombin low- (L) and high- (H) responders, respectively. A–D denote the density intervals. A: density subpopulations nos. 1–4; density span 1.090–1.079 kg/L. B: density subpopulations nos. 5–8; density span 1.079–1.067 kg/L. C: density subpopulations nos. 9–12; density span 1.067–1.054 kg/L. D: density subpopulations nos. 13–16; density span 1.054–1.040 kg/L. LAMP-1: lysosomal-associated membrane protein 1; nos.: numbers; PAC-1: fibrinogen receptor (αIIbβ3) activity; SD: standard deviation; WB: whole blood. The quantities (mean ± SD) of normal-sized platelets assessed by the flow cytometer apparatus for the density fractions (nos. 1–16) are given in the legend of Figure 2. Figure 4.: Surface LAMP-1 (left) and PAC-1 (right) (mean ± SD) expressions of density-separated ordinary platelets divided according to WB annexin V after ɑ-thrombin stimulation (10 U/mL). The colors illustrate statistical significance (red, p < 0.01; yellow, p < 0.05; black, not significant). For each couple, the left bar demonstrates WB ɑ-thrombin low-responders (L), and the right one shows their high-responding counterparts (H). A–D indicate the density intervals. A: density subpopulations nos. 1–4; density span 1.090–1.079 kg/L. B: density subpopulations nos. 5–8; density span 1.079–1.067 kg/L. C: density subpopulations nos. 9–12; density span 1.067–1.054 kg/L. D: density subpopulations nos. 13–16; density span 1.054–1.040 kg/L. LAMP-1: lysosomal-associated membrane protein 1; nos.: numbers; PAC-1: fibrinogen receptor (αIIbβ3) activity; SD: standard deviation; WB, whole blood. The numbers (mean ± SD) of normal-sized corpuscles evaluated by the flow cytometer for the density fractions (nos. 1-16) are shown in Figure 2. Figure 5.: Surface LAMP-1 (left) and PAC-1 (right) (mean ± SD) of density separated normal-sized platelets divided according to WB DiIC1(5) responses after ɑ-thrombin provocation (10 U/mL). In this setting, a high DiIC1(5) denotes less disintegrated mitochondria. The colors designate the levels of significance (red, p < 0.01; yellow, p < 0.05; black, not significant). For each pair, the left bar (L) summarizes patients with more disintegrated WB platelet mitochondria after ɑ-thrombin provocation. The right bar (H) displays patients showing more retained organelles after stimulation. A–D denote the density intervals. A: density subpopulations nos. 1–4; density span 1.090–1.079 kg/L. B: density subpopulations nos. 5–8; density span 1.079–1.067 kg/L. C: density subpopulations nos. 9–12; density span 1.067–1.054 kg/L. D: density subpopulations nos. 13–16; density span 1.054–1.040 kg/L. DiIC1(5): mitochondrial transmembrane potential, that is, retention of 1,1′,3,3,3′,3′-hexamethylindodicarbocyanine iodide; LAMP-1: lysosomal-associated membrane protein 1; nos.: numbers; PAC-1: fibrinogen receptor (αIIbβ3) activity; SD: standard deviation; WB: whole blood. The numbers (mean ± SD) of normal-sized corpuscles evaluated by the flow cytometer for the density fractions (nos. 1–16) are shown in Figure 2. ## Results Figure 2 shows that LAMP-1 expression of very dense platelets (A: subpopulations numbers nos. 1–4; density span 1.090–1.079 kg/L) ($p \leq 0.01$), semi-dense populations (B: subpopulations nos. 5–8; density span 1.079–1.067 kg/L) ($p \leq 0.05$), and light platelets (D: subpopulations nos. 13–16; density span 1.054–1.040 kg/L) ($p \leq 0.05$) was closely related to WB ɑ-thrombin-evoked (10 U/mL) lysosomal release (LAMP-1). In contrast, surface PAC-1 of density-separated circulating platelets did not affect WB LAMP-1 expression after provocation (Figure 2). The ratio (%) of normal-sized platelets expressing LAMP-1 in the density subpopulations and WB ɑ-thrombin-evoked (10 U/mL) fibrinogen receptor activity were closely related (Figure 3). For subpopulations (A, B: nos. 1–8; density span 1.090–1.067 kg/L) and (D: subpopulations nos. 13–16; density span 1.054–1.040 kg/L), the levels of significance were <0.05. Figure 3 also shows the close relationship between normal-sized platelets expressing the activated fibrinogen receptor (PAC-1) and WB agonist-evoked αIIbβ3 receptor activity. The p-values reached <0.01 for very dense populations (A: subpopulations nos. 1–4; density span 1.090–1.079 kg/L) and the cohorts (B–C: subpopulations nos. 5–12; density span 1.079–1.054 kg/L) displayed corresponding higher p-values ($p \leq 0.05$). Figure 4 shows the proportions (%) of LAMP-1-expressing density-separated normal-sized platelets and inverse relationships with ɑ-thrombin-evoked (10 U/mL) WB annexin V. For A, B: subpopulations nos. 1–8 (density span 1.090–1.067 kg/L), the levels of significance were <0.01. Lighter platelets (C, D: subpopulations nos. 9–16; density span 1.067–1.040 kg/L) showed lower but still significant differences ($p \leq 0.05$). It is also apparent that the percentage of PAC-1-positive semi-dense platelets (B: subpopulation: nos. 5–8; density span 1.079–1.067 kg/L) and WB ɑ-thrombin-evoked annexin V were inversely associated ($p \leq 0.05$). Figure 5 compares the ratios (%) of density-separated LAMP-1 positive platelets and WB ɑ-thrombin-induced (10 U/mL) DilC1[5]. For A, B: subpopulations nos. 1–8 (density span 1.090–1.067 kg/L), the levels of significance were <0.01; the lower subpopulation surface LAMP-1 was associated with more disintegrated mitochondria after WB provocation. The corresponding p-values for lighter, normal-sized platelets (C, D: subpopulations nos. 9–16; density span 1.067–1.040 kg/L) were <0.05. The ratio (%) of PAC-1 expressing, medium dense normal-sized platelets (B: subpopulation nos. 5–8; density span 1.079–1.067 kg/L) also showed an inverse relationship with WB mitochondrial integrity after ɑ-thrombin provocation ($p \leq 0.05$) (Figure 4). Platelet lysosomal release (LAMP-1) was lower in unstimulated WB than in semi-dense normal-sized subpopulations, indicating that centrifugation causes lysosomal release. In contrast, sample processing did not provoke the αIIbβ3 receptor (PAC-1), surface annexin V, or mitochondrial disintegration (Table 3). Table 3. \| Unnamed: 0 \| LAMP-1 (% of positive platelets) \| PAC-1 (% of positive platelets) \| Annexin V (% of positive platelets) \| DiIC1(5) (% of positive platelets) \| \| --- \| --- \| --- \| --- \| --- \| \| Whole blooda \| 4 ± 4 \| 4 ± 3 \| 2 ± 1 \| 99 ± 1 \| \| Density subpopulation Ba,b \| 19 ± 11 \| 5 ± 3 \| 2 ± 1 \| 99 ± 1 \| \| Density subpopulation Ca,c \| 20 ± 15 \| 4 ± 3 \| 3 ± 4 \| 99 ± 2 \| ## Discussion This descriptive study showed that platelet activity, that is, proportions of LAMP-1 expressing circulating normal-sized populations, although with some exceptions, are closely related to WB in vitro ɑ-thrombin-evoked (10 U/mL) platelet lysosomal release (LAMP-1) (Figure 2). Proportions were also associated with WB α-thrombin-induced fibrinogen receptor activity (PAC-1) (Figure 3) and inversely associated with procoagulant platelet reactions (i.e. surface annexin V (Figure 4) and mitochondrial injury (Figure 5) after provocation. It is to hypothesize that such previously unknown mechanisms affecting platelet reactivity offer new options for influencing platelets in type 2 diabetes. Lysosomal exocytosis (surface LAMP-1) of normal-sized, semi-light platelets (C: subpopulations nos. 9–12; density span 1.067–1.054 kg/L) was not associated with ɑ-thrombin-provoked LAMP-1 and the αIIbβ3 receptor PAC-1 activity (Figures 2 and 3). This finding suggests that platelet heterogeneity affects ɑ-thrombin-induced lysosomal release and fibrinogen receptor activity in vitro. The percentage of circulating normal-sized platelets displaying activated fibrinogen receptors promoted ɑ-thrombin-evoked αIIbβ3 reactions, i.e. aggregation (Figure 3), but failed to modify the WB lysosomal discharge (Figure 2). Such platelets, although to a lesser degree, affected the generation of procoagulant platelets (Figures 4 and 5). To avoid platelet activation during density separation, citrate-anticoagulated WB was added instantaneously to an inhibitory mixture containing EDTA and prostaglandin E1. The gradients also contained such compounds [7,26,27]. Sample handling neither activated αIIbβ3 receptors nor the creation of procoagulant platelets. In contrast, the surface LAMP-1 levels were higher after laboratory processing (Table 3). The blocking solution also makes it impossible to use subpopulations for functional studies. For normal-sized platelets, statistical evaluation was performed for the mean values of multiple density intervals ($$n = 4$$). LAMP-1 expression in these platelets was closely related to ɑ-thrombin-induced (10 U/mL) lysosomal release and αIIbβ3 receptor activation (Figure 2). The ratios of LAMP-1 expressing normal-sized density subpopulations were also inversely linked to procoagulant platelet creation after agonist provocation (Figures 4 and 5). It is possible that such multiple associations make statistical errors unlikely. Subpopulations showed substantial individual variations with respect to surface LAMP-1 (Figures 2–5), but WB platelet counts (Table 1) and platelet activity in response to ɑ-thrombin also displayed large standard deviations. Consequently, heterogeneity at the baseline may partly explain the individual divergence after sample processing. Community-dwelling seniors who visited an outpatient clinic for type 2 diabetes were enrolled. We wanted to study a procoagulant condition with respect to differences within the diseased group so we did not include healthy individuals as controls; therefore, our findings could be features of disease or they may occur in health as well. The participants were heterogeneous with respect to comorbidities and medications (Table 1), which may have influenced the results. As judged from the medications (Table 1), elevated cholesterol and hypertension were common. When providing consent, some patients were taking either ADP receptor blockers ($$n = 1$$) or aspirin ($$n = 6$$). Platelet ɑ-thrombin responses were unrelated to aspirin (data not shown), but clopidogrel affects agonist-evoked platelet WB activity [28]. The current study investigates normal-sized platelets as defined by flow cytometry gating (Figure 1) [26]. This made it possible to study a homogeneous population with respect to size, as the device excluded small platelets and extracellular vesicles. Flow cytometry did not quantify the individual platelets with respect to the degree of activation. It is thus possible that assessment of the fluorescence strength of activated platelets would provide a better understanding of platelet reactions. Type 2 diabetes is characterized by an increased mean platelet volume but in this study WB platelet size was measured in EDTA anticoagulated blood by an automatic cell counter. The time between sampling and analysis was not standardized. Thus, we did not evaluate WB mean platelet volumes scientifically. This study verifies findings of previous work [26,27] by showing that circulating normal-sized high-density (A: subpopulations nos. 1–4; density span 1.090–1.079 kg/L) and low-density (D: subpopulations nos. 12–16; density span 1.054–1.040 kg/L) platelets display more enhanced αIIbβ3 activity in vivo than intermediate-density cohorts (Figure 2). This work broadens the observations by demonstrating that surface LAMP-1 (Figure 2) and annexin V (Figure 4) of resting normal-sized platelets behaved similarly with high- and low-density subpopulations, showing increased in vivo activities. Possible platelet agonist responses involve lysosomal release, enhanced fibrinogen receptor activity, and an increased procoagulant platelet count. This study did not explain the mechanisms that bifurcate WB platelets in performing different tasks when activated. It is not to confuse association with causality, as the latter requires more in-depth experimentation also including animal studies. It makes it obligatory to appraise the practical consequences of the current findings in forthcoming research. However, in type 2 diabetes, lysosomal release, as estimated from the surface LAMP-1 of circulating ordinary platelets, forecasts in vitro WB reactions (lysosomal discharge and αIIbβ3 receptor activity) after ɑ-thrombin provocation (10 U/mL) (Figures 2 and 3). The marker also showed inverse relationships with agonist-induced procoagulant platelet formation (Figures 4 and 5). Thus, lysosomal discharge of circulating normal-sized platelets associated with ɑ-thrombin-induced WB platelet response in vitro. ## Conclusions Lysosomal exocytosis of circulating density-separated normal-sized platelets forecasts in vitro WB ɑ-thrombin-induced (10 U/mL) responses in type 2 diabetes. Higher proportions of circulating platelets expressing LAMP-1 promoted WB lysosomal release and αIIbβ3 receptor activity, and lower ratios enhanced WB procoagulant platelet production. It is to hypothesize that the new described regulatory mechanism offers new ways to influence platelet behavior in type 2 diabetes. ## Disclosure statement The authors declare that the study was conducted in the absence of financial links that may be construed as conflicts of interest. ## Data Availability statement Raw data are available from the corresponding author upon reasonable request. ## References 1. Heemskerk JWM, Mattheij NJA, Cosemans JMEM.. Platelet-based coagulation: different populations, different functions. J Thromb Haemost (2013) 11 2-16 2. Mazepa M, Hoffman M, Monroe D.. Superactivated platelets: thrombus regulators, thrombin generators, and potential clinical targets. Arterioscler Thromb Vasc Biol (2013) 33 1747-1752. PMID: 23864724 3. Dale GL.. Procoagulant platelets: further details but many more questions. Arterioscler Thromb Vasc Biol (2017) 37 1596-1597. PMID: 28835486 4. Södergren AL, Ramström S.. Platelet subpopulations remain despite strong dual agonist stimulation and can be characterised using a novel six-colour flow cytometry protocol. Sci Rep (2018) 8 1441. PMID: 29362366 5. Panteleev MA, Ananyeva NM, Greco NJ. Two subpopulations of thrombin-activated platelets differ in their binding of the components of the intrinsic factor X-activating complex. J Thromb Haemost (2005) 3 2545-2553. PMID: 16241952 6. Bevers EM, Comfurius P, van Rijn JL. Generation of prothrombin-converting activity and the exposure of phosphatidylserine at the outer surface of platelets. Eur J Biochem (1982) 122 429-436. PMID: 7060583 7. Järemo P.. Computerised method for recording platelet density distribution. Eur J Haematol (1995) 54 304-309. PMID: 7781754 8. Milovanovic M, Lysen J, Ramström S. Identification of low-density platelet populations with increased reactivity and elevated alpha-granule content. Thromb Res (2003) 111 75-80. PMID: 14644083 9. Milovanovic M, Lotfi K, Lindahl TL. Platelet density subpopulations in essential thrombocythemia and healthy volunteers. Pathophysiol Haemost Thromb (2010) 37 35-42. PMID: 20484885 10. Corash L, Costa JL, Shafer B. Heterogeneity of human whole blood platelet subpopulations. III. Density-dependent differences in subcellular constituents. Blood (1984) 64 185-193. PMID: 6234035 11. Chamberlain KG, Penington DG.. Monoamine oxidase and other mitochondrial enzymes in density subpopulations of human platelets. Thromb Haemost (1988) 59 29-33. PMID: 3363531 12. Chamberlain KG, Seth P, Jones MK. Subcellular composition of platelet density subpopulations prepared using continuous Percoll gradients. Br J Haematol (1989) 72 199-207. PMID: 2547422 13. Prodan CI, Stoner JA, Cowan LD. Higher coated-platelet levels are associated with stroke recurrence following non-lacunar brain infarction. J Cereb Blood Flow Metab (2013) 33 287-292. PMID: 23149559 14. Prodan CI, Ross ED, Stoner JA. Coated-platelet levels and progression from mild cognitive impairment to Alzheimer disease. Neurology (2011) 76 247-252. PMID: 21242492 15. Wong CK, Namdarian B, Chua J. Levels of a subpopulation of platelets, but not circulating endothelial cells, predict early treatment failure in prostate cancer patients after prostatectomy. Br J Cancer (2012) 107 1564-1573. PMID: 23047552 16. Järemo P, Milovanovic M, Nilsson S. Platelet heterogeneity and Alzheimer´s disease: low density platelets populations demonstrate low. Platelets (2012) 23 116-120. PMID: 22150375 17. Martin JF, Plumb J, Kilbey RS. Changes in volume and density of platelets in myocardial infarction. Br Med J (Clin Res Ed) (1983) 287 456-459 18. Järemo P, Hansson G, Nilsson O.. Elevated inflammatory parameters are associated with lower platelet density in acute myocardial infarctions with ST-elevation. Thromb Res (2000) 100 471-478. PMID: 11152926 19. Järemo P, Sandberg-Gertzen H.. Platelet density and size in inflammatory bowel disease. Thromb Haemost (1996) 5 560-561 20. Järemo P, Lindahl TL, Lennmarken C. The use of platelet density and volume measurements to estimate the severity of pre-eclampsia. Eur J Clin Invest (2000) 30 1113-1118. PMID: 11122327 21. Santilli F, Simeone P, Liani R. Platelets and diabetes mellitus. Prostaglandins Other Lipid Mediat (2015) 120 28-39. PMID: 25986598 22. Ferroni P, Basili S, Falco A. Platelet activation in type 2 diabetes mellitus. J Thromb Haemost (2004) 2 1282-1291. PMID: 15304032 23. Järemo P, Milovanovic M, Lindahl TL. Elevated platelet density and enhanced platelet reactivity in stable angina pectoris complicated by diabetes mellitus type II. Thromb Res (2009) 124 373-374. PMID: 19230957 24. Cohen Z, Gonzales RF, Davis-Gorman GF. Thrombin activity and platelet microparticle formation are increased in type 2 diabetic platelets: a potential correlation with caspase activation. Thromb Res (2002) 107 217-221. PMID: 12479881 25. Bethel MA, Harrison P, Sourij H. Randomized controlled trial comparing impact on platelet reactivity of twice-daily with once-daily aspirin in people with type 2 diabetes. Diabet Med (2016) 33 224-230. PMID: 26043186 26. Edvardsson M, Oweling M, Järemo P.. Small procoagulant platelets in diabetes type 2. Thromb Res (2020) 195 1-7. PMID: 32629151 27. Milovanovic M, Nilsson S, Winblad B. Inverse relationship between erythrocyte size and platelet reactivity in elderly. Platelets (2016) 24 1-5 28. Järemo P, Lindahl TL, Fransson SG. Individual variations of platelet inhibition after loading doses of clopidogrel. J Int Med (2002) 52 233-238
--- title: 'Safety of hepatitis E vaccination for pregnancy: a post-hoc analysis of a randomized, double-blind, controlled phase 3 clinical trial' authors: - Guohua Zhong - Chunlan Zhuang - Xiaowen Hu - Qi Chen - Zhaofeng Bi - Xinhua Jia - Siying Peng - Yufei Li - Yue Huang - Qiufen Zhang - Ying Hong - Youlin Qiao - Yingying Su - Huirong Pan - Ting Wu - Lihui Wei - Shoujie Huang - Jun Zhang - Ningshao Xia journal: Emerging Microbes & Infections year: 2023 pmcid: PMC10026809 doi: 10.1080/22221751.2023.2185456 license: CC BY 4.0 --- # Safety of hepatitis E vaccination for pregnancy: a post-hoc analysis of a randomized, double-blind, controlled phase 3 clinical trial ## ABSTRACT Special attention has been paid to Hepatitis E (HE) prophylaxis for pregnant women due to poor prognosis of HE in this population. We conducted a post-hoc analysis based on the randomized, double-blind, HE vaccine (Hecolin)-controlled phase 3 clinical trial of human papillomavirus (HPV) vaccine (Cecolin) conducted in China. Eligible healthy women aged 18–45 years were randomly assigned to receive three doses of Cecolin or Hecolin and were followed up for 66 months. All the pregnancy-related events throughout the study period were closely followed up. The incidences of adverse events, pregnancy complications, and adverse pregnancy outcomes were analysed based on the vaccine group, maternal age, and interval between vaccination and pregnancy onset. During the study period, 1263 Hecolin receivers and 1260 Cecolin receivers reported 1684 and 1660 pregnancies, respectively. The participants in the two vaccine groups showed similar maternal and neonatal safety profiles, regardless of maternal age. Among the 140 women who were inadvertently vaccinated during pregnancy, the incidences of adverse reactions had no statistical difference between the two groups ($31.8\%$ vs $35.1\%$, $$p \leq 0.6782$$). The proximal exposure to HE vaccination was not associated with a significantly higher risk of abnormal foetal loss (OR 0.80, $95\%$ CI 0.38–1.70) or neonatal abnormality (OR 2.46, $95\%$ CI 0.74–8.18) than that to HPV vaccination, as did distal exposure. Significant difference was not noted between pregnancies with proximal and distal exposure to HE vaccination. Conclusively, HE vaccination during or shortly before pregnancy is not associated with increased risks for both the pregnant women and pregnancy outcomes. ## Introduction Hepatitis E (HE) is a common viral hepatitis caused by the hepatitis E virus (HEV) infection. There are an estimated 20 million HEV infections worldwide every year, leading to approximately 3.3 million symptomatic cases and 44,000 deaths [1]. Although HEV infection is an acute and self-limited disease, progressions to fulminant hepatic failure and other severe consequences have been recorded in pregnant women [2,3]. In some high-prevalence areas, HEV infection has become a major contributor to maternal mortality and abnormal foetal loss [4]. A substantially higher maternal case fatality rate in pregnant women with HEV infection, varied between $3.2\%$ and $70\%$, was shown in a meta-analysis [5]. Other adverse pregnancy outcomes, including low birth weight, stillbirth, preterm birth (<32 weeks), and intrauterine deaths, were also reported to be highly associated with HEV infection [6]. A recombinant HE vaccine (Hecolin; Xiamen Innovax, Xiamen, China), which has been licensed in 2011 and 2020 in China and Pakistan, separately, is the world’s first and only vaccine to prevent HEV infection. It has been demonstrated with good safety and an efficacy of $100\%$ within 12 months after the third dose in a large-scale, randomized, double-blind, placebo-controlled, phase 3 trial [7]. The efficacy of $93.3\%$ ($95\%$ confidence interval [CI] 78.6–97.9) in the per-protocol set was confirmed within 4.5 years of follow-up [8]. The highly efficacious HE vaccine may hold great promise for reducing the high mortality and multiple adverse outcomes associated with HEV infection in pregnant women. However, concerns on whether HE vaccination has a causal effect on adverse pregnancy outcomes warrant further investigation in the absence of rigorous data in clinical trials currently. A published post-hoc analysis with a limited sample size of inadvertent HE vaccinees during pregnancy in Hecolin’s phase 3 trial did not show any potential risk for foetus [9]. Therefore, with a view to further investigation of the potential risk posed by HE vaccination during or shortly before pregnancy, we conducted a post-hoc analysis of pregnancy outcomes based on a large-scale phase 3 clinical trial of an Escherichia coli (E. coli)-produced human papillomavirus (HPV) bivalent vaccine, in which the control group received Hecolin. Given that the current safety data for the HPV vaccine is reassuring [10], we compared the safety profile between HE vaccination and HPV vaccination for pregnancy in the absence of the unvaccinated control population. ## Study population This post-hoc analysis was based on a multi-centre, double-blind, randomized, and controlled phase 3 trial (ClinicalTrials.gov: NCT01735006) of an E. coli-produced HPV bivalent (type 16 and 18) vaccine (Cecolin, Xiamen Innovax, Xiamen, China) in China as described previously [11]. In brief, eligible healthy women aged 18–45 years from five study sites were randomly assigned in a 1:1 ratio with stratification of age (18–26 and 27–45 years) to receive three doses of Cecolin or Hecolin at months 0, 1, and 6. The main exclusion criteria included immunodeficiency disorder, previous severe allergic reactions after vaccination, severe internal disease, pregnancy, and previous HPV vaccination. All the participants were followed up at least for 66 months according to the protocol. The trial was approved by the Independent Ethics Committee. Written informed consent was obtained from each participant, and the study was done in accordance with the principles of the Declaration of Helsinki, the standards of Good Clinical Practice, and Chinese regulatory requirements. ## Vaccines The control HE vaccine (Hecolin, Xiamen Innovax, Xiamen, China) contains 30 μg of the purified HEV239 particulate antigen adsorbed to 0.8 mg aluminium hydroxide suspended in 0.5 mL buffered saline [7]. Hecolin, which has been licensed in China in 2011 and Pakistan in 2020, is the world’s first and only vaccine to prevent HEV infection. The HPV vaccine (Cecolin, Xiamen Innovax, Xiamen, China) contains 40 μg of HPV-16 and 20 μg of HPV-18 L1 VLPs, suspended in 0.5 mL of buffered saline and 208 μg of aluminium adjuvant [12]. Cecolin has been licensed in China, Morocco, Nepal, Thailand, and Congo (DRC) during 2019–2022, and has been included in World Health Organization (WHO) prequalification list in 2021 [13]. The three-dose-regimen for both Hecolin and Cecolin was given intramuscularly at 0, 1, and 6 months. ## Safety monitoring Participants were observed for 30 minutes after each vaccination for the occurrence of any immediate reactions. All the participants were trained to record any solicited and unsolicited adverse events, medication administration, and other vaccinations occurred within 30 days after vaccination on the diary card distributed. Serious adverse events (SAEs) occurring within the 66-month study period were collected regularly by safety assessors. Adverse reactions, defined as adverse events related to vaccination, were judged by investigators with physician qualification according to the implementation rules, as described previously [14]. All the participants in the trial should have negative urine pregnancy test results before each vaccination and were requested to avoid pregnancy within eight months after the first dose. Women with a positive test result should be suspended the vaccination and may continue two weeks after the end of pregnancy (delivery or miscarriage). All pregnancy events during the study period were recorded in detail through periodic telephone or in-person investigations. For pregnant women planning to deliver, regular telephone follow-up should be conducted at least once every three months. The last telephone survey should be conducted one month after the end of delivery. ## Pregnancy outcomes and complications Gestational age was measured from the first day of the last menstrual period (LMP) according to the international consensus. If LMP was unknown or unclear, gestational age would be reported based on ultrasound examination and an estimated LMP was also calculated. Pregnancy has three trimesters: first trimester (0–13 weeks), second trimester (14–26 weeks), and third trimester (27–40 weeks). The due date was the date of the first day of LMP plus 40 weeks (280 days), and maternal age was defined as the age at the time of the due date. Advanced maternal age was defined as 35 years or older. Pregnancy outcomes included termination and delivery. Elective termination was defined as an intentional abortion by an artificial procedure and was not considered as an adverse pregnancy outcome [15]. Adverse pregnancy outcomes referred to abnormal foetal loss and neonatal abnormalities. The former included spontaneous abortion, stillbirth, and termination of pregnancy caused by maternal complications, while the latter included abnormal weight, preterm birth, low Apgar scores, congenital anomaly, and other neonatal complications. Spontaneous abortion was defined as the termination of pregnancy without human interference prior to 28 weeks of gestation [16]. Stillbirth was defined as an abnormal foetal loss at 28 weeks of gestation and beyond. Terminations of pregnancy due to complications with a clear diagnosis were classified as a separate section of abnormal foetal loss. As for neonatal abnormalities, neonates weighing less than 2.5 kg or more than 4.0 kg were considered as abnormal weight, and Apgar score of less than 8 was considered low score. Preterm birth is defined as a live baby born before the completion of a 37-week gestation [17]. Congenital anomalies comprise a wide range of abnormalities of body structure or function that are present at birth [18]. Pregnancy complications were any health problems that occurred during pregnancy, which might involve the mother’s health, the baby’s health, or both [19]. We focused on the occurrence of ectopic pregnancy, gestational hypertension, gestational diabetes, eclampsia, and preeclampsia because of being common and important markers of maternal health. ## Statistical analyses All statistical comparisons were made between the HE vaccine group and the HPV vaccine group using Student’s T-test, Chi-Squared test, Fisher’s exact test, Wilcoxon test, or Cochran-Mantel-Haenszel test. Characteristics of women who became pregnant and their pregnancy outcomes and complications were displayed. Women who inadvertently received HE vaccine during pregnancy were matched in a 1:2 ratio to those HE vaccinees who were nonpregnant throughout the study period. The matched factors included age at enrolment (no more than two years of difference), receiving the same HE vaccine doses and study site. The incidences of adverse events in participants vaccinated during pregnancy, as well as in matched nonpregnant women, were summarized. Pregnancy-related adverse events were divided into three parts: abnormal foetal loss outcomes, neonatal abnormality outcomes, and pregnancy complications in focus. To measure the impact of exposure time for pregnancy on adverse pregnancy outcomes and pregnancy complications, exposure types were classified as proximal and distal based on the interval between exposure and pregnancy. Proximal exposure was defined as vaccination during pregnancy or the onset of pregnancy within 90 days post any dose. Odds ratios with $95\%$ confidence intervals were estimated with the use of a logistic regression model, with presence/absence of adverse pregnancy events considered as the dependent variable and the type of vaccine received and maternal age used as independent variables. The primary comparisons included: [1] pregnant women with any HE vaccine exposure vs those with any HPV vaccine exposure; [2] pregnant women with proximal HE vaccine exposure vs those with proximal HPV vaccine exposure; [3] pregnant women with distal HE vaccine exposure vs those with distal HPV vaccine exposure. In addition, comparison of pregnant women having proximal exposure with those having distal exposure was drawn specifically in the HE vaccine group, using the exposure types and maternal age as independent variables. To avoid masking the acute effects of vaccine, we conducted a sensitivity analysis based on defining proximal exposure as vaccination during pregnancy or the onset of pregnancy within 30 days post any dose. All analyses were performed using the generalized estimation equation (GEE) method to account for possible correlations between pregnancies of the same mother. The dependent variable was binomial (presence/absence of adverse pregnancy events), and hence the logit link function was used in the GEE model. Pregnancy complications and adverse events were coded using the Medical Dictionary for Regulatory Activities (MedDRA, version 20.0) developed by ICH. SAS version 9.4 was used for all the analyses. All reported p-values are two-sided with an α value of 0.05. ## Results During 22 November 2012 to 1 April 2013, 7372 healthy women aged 18–45 years old were enrolled and randomly assigned to receive Hecolin ($$n = 3683$$) or the HPV vaccine ($$n = 3689$$) and were followed up until 1 August 2019 (Figure 1). During the follow-up period, 2523 ($34.2\%$, $\frac{2523}{7372}$) reported having at least one pregnancy, of which 1263 were in the HE vaccine group, and 1260 in the HPV vaccine group (Figure 1). One woman with one pregnancy event in the HPV vaccine group was excluded from the analysis because of incomplete information on pregnancy outcome. Baseline demographic characteristics of the two groups were generally comparable. The median follow-up time for the two groups was 68.5 and 68.2 months ($$p \leq 0.5947$$), respectively. $71.7\%$ ($\frac{905}{1263}$) and $72.9\%$ ($\frac{918}{1259}$) of the pregnant women had only one pregnancy in the two vaccine groups, respectively (Table 1). A total of 3344 pregnancy events occurred during the study, of which 1684 were in the HE vaccine group, and 1660 in the HPV vaccine group (Table 1). The median ($$p \leq 0.1094$$) and distribution ($$p \leq 0.0608$$) of maternal age were similar in two groups. The majority of pregnancies occurred after completing the full vaccination course, and the median interval between the onset of pregnancy and the last vaccination was 23.8 months and 22.7 months in the HE vaccine and HPV vaccine group ($$p \leq 0.2164$$). However, a total of 143 pregnancy events occurred in 140 pregnant women who were inadvertently vaccinated during pregnancy (66 in HE vaccine group and 77 in HPV vaccine group), possibly because early undetected pregnancy or the false negative test results. Only one woman in the HE vaccine group received more than one vaccination (two doses, first and second) during the same pregnancy event. The gestational stage of vaccine exposure was the first trimester for all pregnancy events with vaccination during pregnancy. The proportion of proximal exposure was in the minority in both groups ($12.6\%$ and $13.6\%$, $$p \leq 0.4376$$). Figure 1.Study profile. HE: Hepatitis E. HPV: Human papillomavirus. All the women who received at least one dose of vaccine were followed up to 66 months. * By mistake, one participant in the HPV vaccine group was given the HE vaccine for dose 1, and two participants in the HE vaccine group were given the HPV vaccine for dose 3. These three participants were included in the HPV vaccine group for safety analysis, according to the protocol. †One woman with one pregnancy event in the HPV vaccine group was excluded from the analysis due to loss of follow-up and incomplete information on pregnancy outcome. ‡Proximal exposure was defined as vaccination during pregnancy or the onset of pregnancy within 90 days post any dose. Abbreviations: HE, hepatitis E; HPV, human papillomavirus. Table 1.Characteristics of HE or HPV vaccinees who became pregnant throughout the study. CharacteristicHE vaccine groupHPV vaccine groupp-valuePregnant women No.12631259– Median follow-up months (IQR)68.5 (66.6,69.4)68.2 (66.6,69.3)0.5947 No. of pregnancy (%) 0.8479 1905 (71.7)918 (72.9) 2298 (23.6)283 (22.5) 353 (4.2)53 (4.2) ≥47 (0.6)5 (0.4) Pregnancy events No.16841660– Median maternal age (IQR)28.0 (26.0, 30.0)28.0 (26.0, 31.0)0.1094 Distribution of maternal age-No. (%) 0.0608 <351505 (89.4)1449 (87.3) ≥35179 (10.6)211 (12.7) Site distribution-No. (%) 0.1087 Liuzhou City753 (44.7)719 (43.3) Funing County388 (23.0)446 (26.9) Xinmi County305 (18.1)267 (16.1) Fengning County131 (7.8)130 (7.8) Yangcheng County107 (6.4)98 (5.9) Vaccination during pregnancy-No. (%) 0.3040 Yes66 (3.9)77 (4.6) No1618 (96.1)1583 (95.4) Median interval months between the onset of pregnancy and the last vaccination (IQR)†23.8 (10.8, 37.8)22.7 (9.1 37.9)0.2164 Classification of vaccination exposure‡ -No. (%) 0.4376 Proximal213 (12.6)225 (13.6) Distal1471 (87.4)1435 (86.4) Only one woman in HE vaccine group received more than one vaccination (two doses, first and second) during the same pregnancy event, and her stage of gestational vaccine exposure was defined as the stage at the time of the first dose. The gestational stage of vaccine exposure was the first trimester for all pregnancy events with vaccination during pregnancy.†Data were calculated only for those events without vaccine exposure during pregnancy.‡Proximal exposure was defined as vaccination during pregnancy or the onset of pregnancy within 90 days post any dose. Abbreviations: HE, hepatitis E; HPV, human papillomavirus; No., number; IQR, interquartile range. $11.7\%$ ($\frac{390}{3344}$) of pregnancy events were observed in women of advanced maternal age. As presented, the participants in the two vaccine groups showed similar maternal and neonatal safety profiles in both maternal age strata (Table 2). Elective terminations accounted for $86.6\%$ ($\frac{1397}{1614}$) of termination events, followed by spontaneous abortions ($10.0\%$, $\frac{161}{1614}$). In total, there were 1730 delivery events with 1738 newborns, almost all of which were singleton pregnancies, with only four twin pregnancies occurred in each vaccine group. No differences were observed in the weights, gestational weeks, Apgar scores, and sex distribution of the newborns between the two vaccine groups in both maternal age strata ($p \leq 0.05$). The most common neonatal abnormality in both groups was abnormal weight, with the proportion being $6.1\%$ ($\frac{51}{836}$)/$7.8\%$ ($\frac{59}{761}$) and $16.1\%$ ($\frac{10}{62}$)/$12.7\%$ ($\frac{10}{79}$), respectively, among women with non-advanced or advanced pregnancy. Only three newborns were documented with congenital anomaly and other neonatal complications, of which two infants in the HE vaccine group were diagnosed with congenital ureter stenosis of the right kidney and heritable thalassaemia, respectively, and one male newborn in the HPV vaccine group presented with neonatal asphyxia at birth. These events were not considered to be related to vaccination by the data and safety monitoring board. A small number of pregnancy complications in focus were observed, and the differences were not statistically significant between two vaccine groups. Table 2.Overall summary of pregnancy outcomes and complications. Non-advanced maternal ageAdvanced maternal ageHE vaccine groupHPV vaccine groupp-valueHE vaccine groupHPV vaccine groupp-valueTermination No. (%) $\frac{672}{1505}$ (44.7)$\frac{691}{1449}$ (47.7)$\frac{0.0979118}{179}$ (65.9)$\frac{133}{211}$ (63.0)0.5528 Causes of termination-No. (%) Elective termination$\frac{583}{1505}$(38.7)$\frac{597}{1449}$(41.2)$\frac{0.171899}{179}$(55.3)$\frac{118}{211}$(55.9)0.9027 Spontaneous abortion$\frac{66}{1505}$(4.4)$\frac{69}{1449}$(4.8)$\frac{0.624213}{179}$(7.3)$\frac{13}{211}$(6.2)0.6639 Stillbirth$\frac{13}{1505}$(0.9)$\frac{12}{1449}$(0.8)$\frac{0.91583}{179}$(1.7)$\frac{1}{211}$(0.5)0.3371 Maternal complications$\frac{10}{1505}$(0.7)$\frac{13}{1449}$(0.9)$\frac{0.47193}{179}$(1.7)$\frac{1}{211}$(0.5)0.3371Delivery No. (%) $\frac{833}{1505}$ (55.3)$\frac{758}{1449}$ (52.3)$\frac{0.097961}{179}$ (34.1)$\frac{78}{211}$ (37.0)0.5528 Number of newborns836761 6279 Gender of newborns-No. (%) 0.8669 0.7456 Male459 (54.9)421 (55.3) 37 (59.7)45 (57.0) Female377 (45.1)340 (44.7) 25 (40.3)34 (43.0) Median gestational week of newborns (IQR)39.4 (38.9,40.0)39.6 (39.0,40.0)0.868939.0 (38.0,40.0)39.1 (38.4,40.0)0.1328 Median weight of newborns (IQR)3.3 (3.0,3.6)3.3 (3.1,3.7)0.45563.5 (3.1,3.7)3.4 (3.1,3.7)0.5286 Median Apgar score of newborns (IQR)10 [9,10]10 [9,10]0.560010 [10,10]10 [9,10]0.2422 Neonatal abnormality-No. (%) Abnormal weight$\frac{51}{836}$ (6.1)$\frac{59}{761}$ (7.8)$\frac{0.192810}{62}$ (16.1)$\frac{10}{79}$ (12.7)0.5577 Preterm birth$\frac{26}{836}$ (3.1)$\frac{30}{761}$ (3.9)$\frac{0.36666}{62}$ (9.7)$\frac{5}{79}$ (6.3)0.5535 Low Apgar score$\frac{10}{836}$ (1.2)$\frac{6}{761}$ (0.8)$\frac{0.41381}{62}$ (1.6)$\frac{2}{79}$ (2.5)1.0000 Congenital anomaly and other neonatal complications†$\frac{2}{836}$ (0.2)$\frac{1}{761}$ (0.1)1.000000–Pregnancy complications in focus$\frac{17}{1505}$(1.1)$\frac{24}{1449}$ (1.7)$\frac{0.22126}{179}$ (3.4)$\frac{4}{211}$ (1.9)0.5529Four twin pregnancies occurred in both groups, resulting in a greater number of newborns than delivery events.†In the HE vaccine group, a female infant was diagnosed with congenital ureter stenosis of the right kidney and a male infant was diagnosed with heritable thalassaemia. In the HPV vaccine group, a male newborn presented with neonatal asphyxia at birth. These events were not considered to be related to vaccination by the data and safety monitoring board. Abbreviations: HE, hepatitis E; HPV, human papillomavirus; No., number; IQR, interquartile range. Table 3 showed the reported adverse events of the 140 women (66 in HE vaccine group and 74 in HPV vaccine group) who were vaccinated inadvertently during pregnancy. The overall reactogenicity profiles of the two vaccines were similar. The incidence of serious adverse events (SAEs) after vaccinating during pregnancy in the two groups was comparable ($$p \leq 0.1080$$), and none of them were judged as related to vaccination by the investigators. All the adverse reactions were mild and with severity no more than grade 2 (data not shown). The incidence of adverse reactions between the two groups showed no statistical difference ($31.8\%$ vs $35.1\%$, $$p \leq 0.6782$$). Local symptoms, such as pain, redness, induration, and pruritus at the injection site, occurred in both groups. Moreover, the most common systemic adverse reaction in both groups was fever ($18.2\%$ vs $20.3\%$, $$p \leq 0.7545$$), with all events of fever being grade 1 (axillary temperature between 37.1°C and 37.5°C). There were also no statistically significant differences in the incidences of SAE, systemic symptom, and local symptom among women who received HE vaccine during pregnancy compared with matched nonpregnant vaccinees (Table 3). Table 3.Adverse events in women who were inadvertently vaccinated during pregnancy. Classification of adverse eventsHE vaccine group ($$n = 66$$)HPV vaccine group ($$n = 74$$)p-value‡Matched nonpregnant HE vaccinees ($$n = 132$$)p-value§Serious adverse events after vaccinating during pregnancy -No. (%) 3 (4.5)9 (12.2)0.108010 (7.6)0.5499Adverse reaction†-No. (%) 21 (31.8)26 (35.1)0.678246 (34.8)0.6710 Systemic symptom17 (25.8)19 (25.7)0.991235 (26.5)0.9091 Fever12 (18.2)15 (20.3)0.754523 (17.4)0.8952 Myalgia1 (1.5)2 (2.7)1.00005 (3.8)0.6657 Headache2 (3.0)1 (1.4)0.60173 (2.3)1.0000 Fatigue3 (4.5)0 (0.0)0.10224 (3.0)0.6880 Allergic reaction1 (1.5)0 (0.0)0.47142 (1.5)1.0000 Diarrhoea1 (1.5)1 (1.4)1.00002 (1.5)1.0000 Cough0 (0.0)0 (0.0)-4 (3.0)0.3034 Nausea0 (0.0)1 (1.4)1.00001 (0.8)1.0000 Dizziness0 (0.0)0 (0.0)-1 (0.8)1.0000 Abdominal pain0 (0.0)0 (0.0)-1 (0.8)1.0000 Local symptom9 (13.6)12 (16.2)0.669618 (13.6)1.0000 Pain4 (6.1)10 (13.5)0.142315 (11.4)0.2324 Redness1 (1.5)2 (2.7)1.00000 (0.0)0.3333 Pruritus4 (6.1)1 (1.4)0.18804 (3.0)0.4446 Swelling1 (1.5)0 (0.0)0.47142 (1.5)1.0000 Induration1 (1.5)4 (5.4)0.37012 (1.5)1.0000Serious adverse events included only those that occurred after vaccinating during pregnancy. None of severe adverse events reported was related to the vaccine.†Adverse reaction: Adverse events related to vaccine. For the HE vaccine group and the HPV vaccine group, adverse reactions included only those that occurred within 30 days after the vaccine administered during pregnancy. For matched nonpregnant HE vaccinees, adverse events were included and analysed only for the matched doses with the pregnant women group.‡p-value for the comparison between the HE vaccine group and the HPV vaccine group.§p-value for the comparison between the HE vaccine group and the matched nonpregnant HE vaccinees. Abbreviations: HE, hepatitis E; HPV, human papillomavirus; No., number. The incidence of each adverse pregnancy outcome was similar between the HE vaccine group and the HPV vaccine group (Supplementary Table S1). Adverse pregnancy outcome of spontaneous abortion has the highest incidence, with $5.9\%$ and $6.4\%$ in two groups, respectively ($$p \leq 0.6056$$). The incidence of stillbirth was lower, at $1.3\%$ and $1.0\%$, respectively ($$p \leq 0.5811$$). The incidence of pregnancy complications in focus was also comparable ($1.7\%$ vs $2.2\%$, $$p \leq 0.3852$$). As presented in Table 4, in both strata of pregnancy events with proximal or distal exposure to vaccination, the risks of each type of adverse pregnancy outcome in the HE vaccine group were comparable to that in the HPV vaccine group. The proximal exposure to HE vaccination was not associated with a significantly higher risk of abnormal foetal loss than that of HPV vaccination (OR 0.80, $95\%$ CI 0.38–1.70), as did distal exposure (OR 1.02, $95\%$ CI 0.76–1.37). HE vaccination exposure at either time window (proximal exposure: OR 2.46, $95\%$ CI 0.74–8.18; distal exposure: OR 0.91, $95\%$ CI 0.66–1.24) was also not associated with neonatal abnormality, compared with HPV vaccination. Within the HE vaccine group, significant differences were not noted between proximal and distal exposure in terms of abnormal foetal loss (OR 1.17, $95\%$ CI 0.64–2.15), neonatal abnormality (OR 0.82, $95\%$ CI 0.40–1.68) and pregnancy complications in focus (OR 1.34, $95\%$ CI 0.37–4.86). And there were no differences in the occurrences of pregnancy complications among the different strata of vaccination exposure. The percentage of each pregnancy complication in focus was presented comprehensively in Supplementary Table S2. A sensitivity analysis for exposure time, which changed the definition of proximal exposure from the onset of pregnancy within 90 days post vaccination to 30 days post vaccination, was conducted (Supplementary Table S3). Similarly, increased risk of any adverse pregnancy event was not observed in women who received the HE vaccine during pregnancy or started the pregnancy within 30 days post vaccination. Table 4.Association between exposure to vaccination and adverse pregnancy outcomes or pregnancy complications. Proximal exposureDistal exposureExposure at any timeOR† ($95\%$ CI)p-value†HE vaccine group ($$n = 213$$)HPV vaccine group ($$n = 225$$)OR ($95\%$ CI)p-valueHE vaccine group ($$n = 1471$$)HPV vaccine group ($$n = 1435$$)OR ($95\%$ CI)p-valueHE vaccine group ($$n = 1684$$)HPV vaccine group ($$n = 1660$$)OR ($95\%$ CI)p-valueAbnormal fetal loss14 (6.6)18 (8.0)0.80 (0.38,1.70)0.566894 (6.4)91 (6.3)1.02 (0.76,1.37)0.9037108 (6.4)109 (6.6)0.98 (0.74,1.29)0.89351.17 (0.64,2.15)0.6092Spontaneous abortion14 (6.6)13 (5.8)1.13 (0.51,2.53)0.760465 (4.4)69 (4.8)0.92 (0.65,1.31)0.656079 (4.7)82 (4.9)0.95 (0.69,1.31)0.75851.67 (0.91,3.08)0.0989Stillbirth0 (0.0)1 (0.4)NANA16 (1.1)12 (0.8)1.31 (0.62,2.78)0.485316 (1.0)13 (0.8)1.22 (0.58,2.55)0.5948NANAMaternal complications0 (0.0)4 (1.8)NANA13 (0.9)10 (0.7)1.29 (0.56,2.98)0.547813 (0.8)14 (0.8)0.93 (0.43,2.00)0.8557NANANeonatal abnormality9 (4.2)4 (1.8)2.46 (0.74,8.18)0.142980 (5.4)86 (6.0)0.91 (0.66,1.24)0.536389 (5.3)90 (5.4)0.98 (0.72,1.32)0.88800.82 (0.40,1.68)0.5876Abnormal weight6 (2.8)2 (0.9)3.28 (0.64,16.82)0.154254 (3.7)66 (4.6)0.79 (0.55,1.15)0.217460 (3.6)68 (4.1)0.87 (0.61,1.24)0.43940.81 (0.34,1.95)0.6396Preterm birth1 (0.5)2 (0.9)0.52 (0.05,5.97)0.600430 (2.0)32 (2.2)0.92 (0.55,1.55)0.763731 (1.8)34 (2.0)0.91 (0.55,1.51)0.70870.26 (0.03,1.95)0.1905Low Apgar score1 (0.5)2 (0.9)0.52 (0.05,5.75)0.59648 (0.5)6 (0.4)1.31 (0.45,3.77)0.61969 (0.5)8 (0.5)1.11 (0.43,2.88)0.82980.84 (0.11,6.58)0.8711Congenital anomaly and other neonatal complications1 (0.5)0 (0.0)NANA1 (0.1)1 (0.1)1.00 (0.06,16.70)1.00002 (0.1)1 (0.1)1.97 (0.18,21.71)0.58047.58 (0.78,73.66)0.0807Pregnancy complications in focus3 (1.4)5 (2.2)0.65 (0.15,2.82)0.568420 (1.4)23 (1.6)0.85 (0.46,1.59)0.619123 (1.4)28 (1.7)0.82 (0.46,1.45)0.49271.34 (0.37,4.86)0.6592The reported measures of association are odds ratios estimated with the use of logistic regression model.†Comparison of the risk between proximal and distal exposures was specifically drawn in the HE vaccine group. Abbreviations: HE, hepatitis E; HPV, human papillomavirus; OR, odds ratio; NA, no answer; CI, confidence interval. ## Discussion This post-hoc analysis showed that the HE vaccine did not have a negative effect on pregnancy relative to risks observed in HPV-vaccinated subjects. The incidences of adverse pregnancy outcomes were low and comparable between the HE vaccine receivers and HPV vaccine receivers. No potential risk was observed in vaccinated women of both the advanced or non-advanced maternal age. Additionally, there was no increased risk of abnormal foetal loss, neonatal abnormality, and pregnancy complication in women who received the HE vaccine during pregnancy or within 90 days pre-pregnancy compared with those who became pregnant at more than 90 days after vaccination. The vaccine was well tolerated during pregnancy and no vaccine-related SAE was reported. Since licensure in 2006, over 500 million doses of HPV vaccines have been distributed and 126 countries have introduced the HPV vaccine into their national immunization programmes [10,20]. As of January 2023, Cecolin has been licensed in China, Morocco, Nepal, Thailand, and Congo (DRC) while Hecolin has been licensed only in China and Pakistan. Although HPV vaccines have not yet been recommended for use in pregnant women, accumulating evidences from clinical trials and post-marketing surveillance revealed no adverse signals after maternal HPV vaccine exposure [21–23]. Also, the Global Advisory Committee for Vaccine Safety (GACVS) has not identified any safety concerns through regular review [10]. Therefore, receiving HPV vaccine during pregnancy does not cause for alarm. As stated in the updated WHO position paper, termination of pregnancy is not indicated if vaccination was carried out inadvertently during pregnancy, and data on the safety of HPV vaccination during pregnancy are thought as reassuring [10]. Therefore, we compared the safety profile of Hecolin with that of Cecolin in the absence of the unvaccinated control population. A large proportion of terminations were caused by elective terminations in our study, which may be closely associated with China’s birth control policy. During the first year of our study, only couples who were both from one-child families were allowed to have two children. Since November 2013, a selective two-child policy was available if either spouse was an only child, and a universal two-child policy was implemented by Chinese government after 2016 [24]. In line with the changes in fertility policy, we found that elective terminations that occurred prior to 2016 accounted for approximately $\frac{2}{3}$ (data not shown) of all the elective terminations occurring throughout the study. Among the pregnant women, the incidence of spontaneous abortion was higher than that of stillbirth and termination caused by maternal complications. The incidence of spontaneous abortion was $5.9\%$ in the HE vaccine group, which was similar as the prevalence of spontaneous abortion in Chinese pregnant women described in other studies [25–27]. In a retrospective study containing around 100,000 live births in southern China, the estimated percentages of low birth weight and macrosomia were about $8.7\%$ and $3.1\%$ between 2005 and 2017 [28]. Our study showed a resemblance, where the proportion of newborns with abnormal weight was $6.1\%$ and $16.1\%$ among those whose mothers were of non-advanced or advanced maternal age in the HE vaccine group, respectively. Our study showed the preterm birth rate in the HE vaccine group was $2.4\%$, which was lower than data from a large-scale observational study of more than 9 million Chinese pregnant women ($5.9\%$ in 2012 and $6.4\%$ in 2018) [29]. The incidence rates of each complication during pregnancy ranged from $0.9\%$ to $4.2\%$ according to previous reports [30], which were similar to the overall incidence of pregnancy complications in focus ($1.7\%$) in the HE vaccine group in our study. There is increasing momentum to develop and implement maternal immunization to prevent specific infections to which pregnancy and newborns are susceptible. Theoretically, it is safe to receive an inactivated vaccine during pregnancy, whereas live vaccines should be contraindicated owing to the potential risk of infection to the foetus [31]. Furthermore, pregnancy is not a contraindication to the use of some recombinant sub-unit vaccines, such as the hepatitis B vaccine [32]. Several recombinant HBV vaccines approved for decades are currently in use worldwide, and no evidence of an increased risk of adverse pregnancy outcomes was observed in more than 20 years of post-marketing surveillance [33,34]. Influenza vaccine and Tdap vaccine have also been currently recommended internationally for pregnant women. HAV vaccine, HBV vaccine, serogroup B meningococcal vaccine, and pneumococcal vaccine have been recommended for specific high-risk groups of pregnancy [35,36]. No events of adverse pregnancy outcomes associated with the HE vaccine were reported by post-marketing surveillance. As the only one commercialized HE vaccine around the world, Hecolin is of high priority for susceptible populations, including women of child-bearing age, and vaccinating women of child-bearing age has been indicated to be cost-effective from the societal perspective especially in epidemic regions [37,38]. In 2022, Médecins Sans Frontières (MSF) and South Sudan’s Ministry of Health jointly carried out two rounds of the hepatitis E vaccination campaign in Bentiu internally displaced persons camp in South Sudan’s Unity state. Around 25,000 people, including pregnant women, have received the vaccine, which was the first rolled-out vaccination in HE epidemic region [39]. To our knowledge, our study is the first one to systematically describe adverse pregnancy outcomes after exposure to HE vaccine during or around the time of pregnancy. These results also confirmed and considerably expanded on results from previous studies of the vaccine, providing additional support for the safety of the vaccine in pregnant women. Certainly, it is of great public health significance to promote clinical research and post-marketing surveillance about maternal vaccination. This post-hoc analysis was accompanied by a number of limitations. Firstly, the long-term effects of the vaccine on infants could not be analysed because the pregnancy events were followed only up to one month after delivery. Another limitation was the small number of women vaccinated during pregnancy. It is challenging to observe rare vaccine-related adverse outcomes based on such a limited sample size. A randomized, controlled clinical trial designed specifically for pregnant women would be the “golden rule” to rule out a meaningful difference in the percentage of adverse pregnancy outcomes. In addition, this phase 3 clinical trial was designed to evaluate the efficacy against HPV infection, therefore the antibody response specific to HEV after vaccination during pregnancy was not assessed and remained to be investigated. In conclusion, HE vaccination during or shortly before pregnancy is not associated with increased risks for both the pregnant women and pregnancy outcomes, but caution is needed until more adequate evidence is available. These data may be helpful to guide clinical practice in reducing undue concerns about the safety of vaccination during or shortly before pregnancy. ## Disclosure statement Qiufen Zhang and Huirong Pan report being either current employees of Xiamen Innovax. No other potential conflicts of interest relevant to this article were reported. ## References 1. 1WHO. Hepatitis E 2020. Available from: https://www.who.int/news-room/fact-sheets/detail/hepatitis-e 2. Kamar N, Izopet J, Pavio N. Hepatitis E virus infection. Nat Rev Dis Primers (2017.0) 3 17086. PMID: 29154369 3. Desai AN.. Hepatitis E. JAMA (2020.0) 323 1862. PMID: 32396182 4. Kourtis AP, Read JS, Jamieson DJ.. Pregnancy and infection. N Engl J Med (2014.0) 370 2211-2218. PMID: 24897084 5. Bergløv A, Hallager S, Weis N.. Hepatitis E during pregnancy: maternal and foetal case-fatality rates and adverse outcomes – a systematic review. J Viral Hepat (2019.0) 26 1240-1248. PMID: 31095813 6. Bigna JJ, Modiyinji AF, Nansseu JR. Burden of hepatitis E virus infection in pregnancy and maternofoetal outcomes: a systematic review and meta-analysis. BMC Pregnancy Childbirth (2020.0) 20 426. PMID: 32723309 7. Zhu FC, Zhang J, Zhang XF. Efficacy and safety of a recombinant hepatitis E vaccine in healthy adults: a large-scale, randomised, double-blind placebo-controlled, phase 3 trial. Lancet (2010.0) 376 895-902. PMID: 20728932 8. Zhang J, Shih JW, Xia NS.. Long-term efficacy of a hepatitis E vaccine. N Engl J Med (2015.0) 372 2265-2266 9. Wu T, Zhu F-C, Huang S-J. Safety of the hepatitis E vaccine for pregnant women: a preliminary analysis. Hepatology (2012.0) 55 2038. PMID: 22161542 10. 10WHO. Human papillomavirus vaccines: WHO position paper; 2022. Available from: https://apps.who.int/iris/handle/10665/365351 11. Zhao F-H, Wu T, Hu Y-M. Efficacy, safety, and immunogenicity of an. Lancet Infect Dis (2022.0) 22 1756-1768. PMID: 36037823 12. Qiao YL, Wu T, Li RC. Efficacy, safety, and immunogenicity of an. J Natl Cancer Inst (2020.0) 112 145-153. PMID: 31086947 13. 13PATH. New HPV vaccine from Innovax receives WHO prequalification 2021. Available from: https://www.path.org/media-center/new-hpv-vaccine-innovax-receives-who-prequalification/ 14. Zhuang CL, Lin ZJ, Bi ZF. Inflammation-related adverse reactions following vaccination potentially indicate a stronger immune response. Emerg Microbes Infect (2021.0) 10 365-375. PMID: 33583360 15. Skipetrova A, Wartel TA, Gailhardou S.. Dengue vaccination during pregnancy - An overview of clinical trials data. Vaccine (2018.0) 36 3345-3350. PMID: 29716774 16. Chinese experts consensus on the diagnosis and treatment of spontaneous abortion. Chin J Practical Gynecol Obstet (2020.0) 36 1082-1090 17. 17WHO. Preterm birth 2018. Available from: https://www.who.int/news-room/fact-sheets/detail/preterm-birth 18. 18CDC. Birth defects surveillance toolkit. Available from: https://www.cdc.gov/ncbddd/birthdefects/surveillancemanual/chapters/chapter-1/chapter1-4.html 19. 19CDC. Pregnancy complications 2020. Available from: https://www.cdc.gov/reproductivehealth/maternalinfanthealth/pregnancy-complications.html 20. 20WHO. HPV vaccination dashboard. Available from: https://app.powerbi.com/view?r=eyJrIjoiNDIxZTFkZGUtMDQ1Ny00MDZkLThiZDktYWFlYTdkOGU2NDcwIiwidCI6ImY2MTBjMGI3LWJkMjQtNGIzOS04MTBiLTNkYzI4MGFmYjU5MCIsImMiOjh9 21. Kharbanda EO, Vazquez-Benitez G, DeSilva MB. Association of inadvertent 9-valent human papillomavirus vaccine in pregnancy with spontaneous abortion and adverse birth outcomes. JAMA Netw Open (2021.0) 4 e214340. PMID: 33818618 22. Scheller NM, Pasternak B, Molgaard-Nielsen D. Quadrivalent HPV vaccination and the risk of adverse pregnancy outcomes. N Engl J Med (2017.0) 376 1223-1233. PMID: 28355499 23. Panagiotou OA, Befano BL, Gonzalez P. Effect of bivalent human papillomavirus vaccination on pregnancy outcomes: long term observational follow-up in the Costa Rica HPV vaccine trial. Br Med J (2015.0) 351 h4358. PMID: 26346155 24. Zeng Y, Hesketh T.. The effects of China’s universal two-child policy. Lancet (London, England) (2016.0) 388 1930-1938. PMID: 27751400 25. Hu CY, Yang XJ, Hua XG. Risk factors for spontaneous abortion from a prevention perspective in rural China: a population-based follow-up study. J Matern Fetal Neonatal Med (2021.0) 34 2583-2591. PMID: 31533498 26. Zhao Y, Zhao Y, Fan K. Association of history of spontaneous or induced abortion with subsequent risk of gestational diabetes. JAMA Netw Open (2022.0) 5 e220944. PMID: 35238932 27. Zhou LY, Zhang HX, Lan YL. Epidemiological investigation of risk factors of the pregnant women with early spontaneous abortion in Beijing. Chin J Integr Med (2017.0) 23 345-349. PMID: 25877464 28. Rao J, Fan D, Wu S. Trend and risk factors of low birth weight and macrosomia in south China, 2005-2017: a retrospective observational study. Sci Rep. (2018.0) 8 3393. PMID: 29467433 29. Deng K, Liang J, Mu Y. Preterm births in China between 2012 and 2018: an observational study of more than 9 million women. Lancet Glob Health (2021.0) 9 e1226-e1241. PMID: 34416213 30. Luo J, Fan C, Luo M. Pregnancy complications among nulliparous and multiparous women with advanced maternal age: a community-based prospective cohort study in China. BMC Pregnancy Childbirth (2020.0) 20 581. PMID: 33008331 31. Munoz FM, Jamieson DJ.. Maternal immunization. Obstet Gynecol (2019.0) 133 739-753. PMID: 30913173 32. 32CDC. Guidelines for vaccinating pregnant women. Available from: https://www.cdc.gov/vaccines/pregnancy/hcp-toolkit/guidelines.html#hepb 33. Groom HC, Irving SA, Koppolu P. Uptake and safety of hepatitis B vaccination during pregnancy: a vaccine safety datalink study. Vaccine (2018.0) 36 6111-6116. PMID: 30194002 34. Moro PL, Zheteyeva Y, Barash F. Assessing the safety of hepatitis B vaccination during pregnancy in the vaccine adverse event reporting system (VAERS), 1990-2016. Vaccine. (2018.0) 36 50-54. PMID: 29174107 35. Nassar AH, Visser GHA, Nicholson WK. FIGO statement: vaccination in pregnancy. Int J Gynaecol Obstet (2021.0) 152 139-143. PMID: 33128249 36. Lin C, Franceschi S, Clifford GM.. Human papillomavirus types from infection to cancer in the anus, according to sex and HIV status: a systematic review and meta-analysis. Lancet Infect Dis (2018.0) 18 198-206. PMID: 29158102 37. Xia R, Sun S, Shen M. Targeted hepatitis E vaccination for women of childbearing age is cost-effective in China. Vaccine (2019.0) 37 5868-5876. PMID: 31443991 38. Zhao Y, Zhang X, Zhu F. A preliminary cost-effectiveness analysis of hepatitis E vaccination among pregnant women in epidemic regions. Hum Vaccin Immunother (2016.0) 12 2003-2009. PMID: 26900799 39. Ciglenecki I, Rumunu J, Wamala JF. The first reactive vaccination campaign against hepatitis E. Lancet Infect Dis (2022.0) 22 1110-1111. PMID: 35870452
--- title: Metabolic Pathway of Monounsaturated Lipids Revealed by In-Depth Structural Lipidomics by Mass Spectrometry authors: - Simin Cheng - Donghui Zhang - Jiaxin Feng - Qingyuan Hu - Aolei Tan - Zhuoning Xie - Qinhua Chen - Huimin Huang - Ying Wei - Zheng Ouyang - Xiaoxiao Ma journal: Research year: 2023 pmcid: PMC10026824 doi: 10.34133/research.0087 license: CC BY 4.0 --- # Metabolic Pathway of Monounsaturated Lipids Revealed by In-Depth Structural Lipidomics by Mass Spectrometry ## Abstract The study of lipid metabolism relies on the characterization of the lipidome, which is quite complex due to the structure variations of the lipid species. New analytical tools have been developed recently for characterizing fine structures of lipids, with C=C location identification as one of the major improvements. In this study, we studied the lipid metabolism reprograming by analyzing glycerol phospholipid compositions in breast cancer cell lines with structural specification extended to the C=C location level. Inhibition of the lipid desaturase, stearoyl-CoA desaturase 1, increased the proportion of n-10 isomers that are produced via an alternative fatty acid desaturase 2 pathway. However, there were different variations of the ratio of n-9/n-7 isomers in C18:1-containing glycerol phospholipids after stearoyl-CoA desaturase 1 inhibition, showing increased tendency in MCF-7 cells, MDA-MB-468 cells, and BT-474 cells, but decreased tendency in MDA-MB-231 cells. No consistent change of the ratio of n-9/n-7 isomers was observed in SK-BR-3 cells. This type of heterogeneity in reprogrammed lipid metabolism can be rationalized by considering both lipid desaturation and fatty acid oxidation, highlighting the critical roles of comprehensive lipid analysis in both fundamental and biomedical applications. ## Introduction Lipids, as an important class of biomolecules, perform critical functions including membrane formation, energy storage, and signal transduction [1–4]. In recent years, our understandings of lipid biofunctions are rapidly increasing in-depth, providing more potential associations between lipids and many diseases, such as cardiovascular disease, COVID-19, and cancer [5–14]. In particular, lipid metabolism reprogramming has been shown to be closely related to cancer cell proliferation, tumor formation, and invasion [12,15,16]. For example, the uptake and de novo synthesis of fatty acid (FA) have been found to be more active in a variety of cancers [17,18]. It has also been reported that a higher degree of lipid unsaturation, reflected by the ratio of stearic acid to oleic acid, is linked to cancer malignancy [19,20]. In addition, phospholipids with longer acyl chains are found to be of higher amounts in squamous cell carcinomas than in normal tissues, and downstream metabolites of FAs such as sphingolipids play an important role in focal adhesion signaling [21,22]. With the rapid development of advanced lipid analysis tools, lipid metabolism reprogramming can be studied with assistance by the information acquired for lipids at more detailed structure levels. In particular, lipid analysis at the C=C location level has become mature with some important demonstrations on its applicability in biological studies [23,24]. The methods developed toward this purpose include ozone-induced dissociation, electron impact excitation of ions from organics, ultraviolet photodissociation, and chemical derivatizations (Paternò–Büchi [PB] reaction, epoxidation reaction, and photooxidation) [25–36]. These tools have not only advanced the lipid structure analysis to an unprecedented level, but also provided opportunities for biomedical analysis, such as discovery of new biomarkers for disease diagnosis [37]. For instance, these methods contributed to the important finding of significant changes in the relative amounts of lipid C=C location isomers in blood or tissue samples for cancer diseases, such as human prostate cancer, colorectal cancer, breast cancer, lung cancer, and lymph nodes with thyroid cancer metastasis [8,32,38–41]. These new findings enabled by the new analytical capability led to further investigation on the underlying biological mechanisms leading to such changes in lipid composition. For instance, the fatty acid desaturase 2 (FADS2) pathway, an alternative FA desaturation pathway bypassing stearoyl-CoA desaturase 1 (SCD1), can cause an abnormal and elevated amount of n-10 isomers, adding to cancer plasticity [7,23,24]. Such kind of studies can enhance our understanding of the altered lipid metabolisms related to cancers and is important for developing more effective cancer therapies. In this work, we aim to study the underlying mechanism of altered lipid compositions, particularly at the C=C location level, after modulating the lipid enzymes’ bioactivity (Fig. 1). The newly established PB derivatization coupled with liquid chromatography–tandem mass spectrometry (LC-PB-MS/MS) workflow was used to monitor omics-level lipid alterations. Upon SCD1 inhibition, lipid unsaturation in human breast cancer cells decreased significantly, as monitored by lipid profiling. Using the new analytical tools, we quantified the relative amounts of the C=C location isomers, including n-10, n-9, and n-7 series isomers. A comprehensive analysis of C=C location isomers toward different breast cancer cell lines revealed that both lipid dehydrogenation and lipid oxidation coregulated lipid fine structure metabolic reprogramming. Interestingly, we have found that the C16:1 n-9/n-7 isomer ratio, which is highly correlated with fatty acid oxidation (FAO) activity, was closely related to the invasiveness of breast cancer cells. Our research demonstrates that lipidomics can be a powerful tool to explore the underlying mechanisms of lipid reprogramming in diseased states. Structural lipidomics will therefore not only serve as a more cost-effective analytical method to acquire detailed lipid composition but also contribute to the study of lipid metabolism. Fig. 1.: Deep GP structure analysis workflow by LC-MS coupling with online PB reaction to reveal C=C location isomer metabolic pathway variations and FAO activity. ## Monitoring lipid unsaturation after SCD1 inhibition The de novo biosynthesis of monounsaturated lipids has been well studied, as shown in Fig. 2A. The C=C location isomers of C18:1- and C16:1-containing glycerol phospholipids (GPs) consist of n-10, n-7, and n-9 isomers. In this lipid metabolic network, SCD1 serves as a central role and is responsible to produce monounsaturated lipids. The composition of unsaturated lipids frequently undergoes significant metabolic changes in various diseased states, due to altered activity of SCD1 and related enzymes. We show in this section the changes in the lipidome of MCF-7 human breast cancer cells after SCD1 inhibition (via CAY10566). Overall, the degree of unsaturation in the lipidome of MCF-7 cells decreased after SCD1 inhibition, consistent with previous studies [42–44]. Specifically, the relative intensity of phosphatidylcholine (PC) 32:0 increased while the intensity ratio of PC 32:1(m/z 732) to PC 32:0 (m/z 734) decreased as SCD1 inhibitor concentration increased (Fig. 2B, C). A similar trend was observed for phosphatidylethanolamine (PE) (Fig. S1), and the intensity ratio of PE 34:1(m/z 718) to PE 34:0 (m/z 720) decreased to reach a plateau at 100 nM CAY10566 (Fig. 2D). The inhibition of SCD1 blocked the dehydrogenation of C16:0 and C18:0, resulting in the accumulation of C16:0 and C18:0. As a result, the abundance of PC 32:0 and PE 34:0 increased. Such higher levels of saturated lipids indicate a more rigid cell membrane and low membrane fluidity [45,46]. Other than lipid desaturation, we also observed altered cell morphology and a reduced number of growing cells. The pseudopod extension was clearly observed, possibly due to reduced cell viability and proliferation after SCD1 inhibition (Fig. S2A). The cell viability assay showed a significant inhibition effect on cell proliferation by CAY10566 in vitro (Fig. S2B), suggesting the essential role of SCD1 for cell growth and survival. Fig. 2.: Lipid unsaturation decreased upon SCD1 inhibition. (A) Biosynthetic pathways of monounsaturated FA. SCD1 and SCD5, stearoyl-CoA desaturase 1 and 5; FADS2, fatty acid desaturase 2; FAO, fatty acid oxidation; Elovl, fatty acid elongase. (B) Mass spectra of PC profile before and after SCD1 inhibition in MCF-7 cells. (C and D) The relative ratios of PC 32:1/PC 32:0 and PE 34:1/PE 34:0. Differences between CAY10566-treated group samples and control group samples were evaluated for statistical significance using the 2-tailed Student’s t-test (P < 0.05, P < 0.01, P < 0.001). Error bar represents the standard deviation, n = 3. ## Increase of n-10 isomers in C16:1- and C18:1-containing GPs after SCD1 inhibition Recently, scientists have taken a new look at FADS2, also known as D6D, which was exploited to produce n-10 isomers in carcinoma, illustrating heterogeneity in FA desaturation and making it more plastic in signaling networks [7,23]. Herein, we studied the relative amounts of n-10 isomers after SCD1 inhibition in breast cancer cells. As expected, the relative amounts of n-10 isomers in both C16:1- and C18:1-containing GPs increased after SCD1 inhibition (Fig. 3A to C). For example, in BT-474 cells, the n-10 isomer for PC 18:0_18:1 increased significantly (Fig. 3D), as evidenced by the increased abundance of diagnostic ions (m/z 664 and m/z 690) for C=C location isomers. The above observations can be rationalized by suppressed n-9 and n-7 isomers synthesis and possibly increased FADS2 activity upon SCD1 inhibition [7]. The increased amounts of polyunsaturated GPs after SCD1 inhibition, whose biosynthesis is also closely related to FADS2, provide additional evidence to support increased FADS2 activity (Fig. 2B and Fig. S1). Fig. 3.: The relative amount of n-10 C=C location isomers increased significantly after SCD1 inhibition. (A to C) Relative increase of n-10 isomers in C16:1- and C18:1-containing GPs after SCD1 inhibition. (D) PB-MS/MS spectra of PC 36:1 before and after SCD1 inhibition in BT-474 cells and the corresponding structures. (E) Compositional increase of n-10 C=C location isomers in BT-474 cells after SCD1 inhibition. Differences between the 2 groups were evaluated for statistical significance using the 2-tailed Student’s t test (P < 0.05, P < 0.01, P < 0.001). Error bar represents the standard deviation, n = 3. It is worth noting that BT-474 cells contain much higher levels of n-10 isomers than other breast cancer cell types and the lipid isomeric changes are also more significant after SCD1 inhibition. We performed LC-PB-(−MS3) in negative ion mode for the direct characterization of unsaturated fatty acyl isomers (Fig. S3), confirming the changes of n-10 isomers in BT-474 cells observed by LC-PB-(+MS2). Global profiling of GPs in BT-474 cells following CAY10566 inhibition shows consistent increases of n-10 isomers for all GP types, including PCs, PEs, or phosphatidylserines (PSs) (Fig. 3E). The relative expression levels of both SCD1 and FADS2 were analyzed by using basal genome-wide expression data previously collected from shared stocks on the Cancer Cell Line Encyclopedia (CCLE) database (http://www.broadinstitute.org/ccle/home, Fig. S4). Compared with SCD1, FADS2 expression generally agrees well with our C=C specific lipid analysis, except for SK-BR-3 cells. For BT-474 cells, in particular, it has the highest level of FADS2 expression and the highest relative amounts of n-10 isomers, as expected. ## Global profiling of n-7 and n-9 isomers in C18:1-containing GPs after SCD1 inhibition in MCF-7 cells The most common and abundant GPs with C=C location isomers are n-7 and n-9 isomers in C18:1-containing GPs. The isomeric compositions of C18:1 GPs showed significant changes after SCD1 inhibition, along with decreased lipid desaturation as revealed by systematic LC-PB-MS/MS analysis. The typical MS/MS spectra of PC 16:0_18:1 and PC 18:0_18:1 in MCF-7 cells before and after SCD1 inhibition are presented in Fig. 4A. Clearly, the intensity ratios of n-9/n-7 isomers for both PCs increased after SCD1 inhibition. The increased intensity ratios of n-9/n-7 isomers for other C18:1-containing PCs, PEs, and PSs are shown in Fig. 4B to D, consistent with previous studies [32,47]. Similar experimental results were observed after siRNA-mediated gene silencing of SCD1, in which the ratios of n-9/n-7 isomers increased in both C18:1-containing PCs, PEs, and PSs (Fig. 4E and Fig. S5A and B). Additionally, gene silencing of SCD1 by siRNA reduced SCD1 mRNA expression levels in MCF-7 cells (Fig. S5c). This set of experiments verifies the known effects of lipid desaturases to modulate lipid desaturation and lipid C=C location isomer composition. From the perspective of biosynthetic pathways in human, there are 2 isoforms of stearoyl-CoA desaturases, i.e., SCD1 and SCD5 [48], both of which catalyze the desaturation of C18:0-CoA to produce C18:1 (n-9)-CoA. However, only SCD1 can convert C16:0-CoA to C16:1 (n-9)-CoA, which can be further elongated to C18:1 (n-7)-CoA. Therefore, once SCD1 is inhibited, the pathway leading to the generation of C18:1 (n-7)-CoA is more severely blocked while C18:1 (n-9)-CoA can still be produced via SCD5, leading to increased n-9/n-7 isomer ratios for C18:1 GPs. Moreover, changes in the chain compositions of GPs were observed by LC-MS/MS in negative ion mode. For instance, upon increased CAY10566 treatment, the relative amount of PC 18:1_18:2 decreased and that of PC 16:0_20:3 increased, due to suppressed C18:1-CoA synthesis by the inhibition of lipid desaturation (Fig. S6). Fig. 4.: Isomer-resolved lipidomics analysis of C18:1-containing GPs in MCF-7 cells. (A) PB-MS/MS analysis of PC 34:1 and PC 36:1 before and after SCD1 inhibition. (B to D) Compositional variations of C=C location isomers in C18:1 fatty acyl in PCs, PEs, and PSs after SCD1 inhibition. (E) Compositional variations of C=C location isomers in C18:1-containing PCs after siRNA-mediated gene silencing of SCD1. Differences between the 2 groups were evaluated for statistical significance using the 2-tailed Student’s t test (P < 0.05, P < 0.01, P < 0.001). Error bar represents the standard deviation, n = 3. ## Differential lipidomics response to SCD1 inhibition in human breast cancer cells It has been reported that lipid metabolism among different breast cancer cell lines is quite different, but few in-depth studies are performed at the C=C location level with large-scale lipidomics [32,49–51]. We conducted a comprehensive lipidomics mapping of C18:1-containing GPs and distinct differences among different subtypes of human breast cancer cell lines. The observed compositions of GPs C=C location isomers can enable the discrimination of different breast cancer cell subtypes via hierarchical cluster analysis (Fig. 5A), in which the cell lines of SK-BR-3 and MDA-MB-231 show a higher degree of similarity with a much high intensity ratio of n-9/n-7 isomers in C18:1-containing GPs. Fig. 5.: Human breast cancer cell lines responded differently to SCD1 inhibition. (A) Hierarchical cluster analysis discriminated the 5 subtypes of human breast cancer cells by quantitative analysis of n-9/n-7 isomers in C18:1-containing GPs. (B) Fold change of n-9/n-7 isomeric ratio in C18:1-containing GPs after SCD1 inhibition. (C) Compositional variations of C=C location isomers in FA 18:1 after SCD1 inhibition. Differences between the 2 groups were evaluated for statistical significance using the 2-tailed Student’s t test (P < 0.05, P < 0.01, P < 0.001). Error bar represents the standard deviation, n = 3. Different from consistent increases in n-10 GP isomers, breast cancer cells of different subtypes show entirely different sensitivity to SCD1 inhibition with diverse tendencies of the ratios of n-9/n-7 isomers in C18:1-containing GPs. The fold changes of C=C location isomers in C18:1-containing GPs after SCD1 inhibition are shown in Fig. 5B. As expected, the intensity ratios of n-9/n-7 isomers of most C18:1-containing GPs in MDA-MB-468 and BT-474 cells increased, similar to MCF-7 cells. However, SK-BR-3 cells and MDA-MB-231 cells showed distinct changes in terms of n-9/n-7 isomer ratios. For SK-BR-3 cells, the n-9/n-7 isomer ratios increased in monounsaturated GPs but decreased in polyunsaturated GPs. For MDA-MB-231 cells, the n-9/n-7 isomer ratios decreased in both monounsaturated and polyunsaturated GPs, except for few GP species, e.g., PE 18:1_18:2 and PE 18:1_19:0. We divide all C18:1-containing GPs into 4 groups based on their fatty acyl compositions that include an odd-number fatty acyl, a saturated fatty acyl, a monounsaturated fatty acyl, or a polyunsaturated fatty acyl. In each group, the changes in n-9/n-7 isomer ratios are overall consistent, possibly suggesting a role of the other fatty acyl on the selective incorporation of C18:1 n-9/n-7 into the GP [8]. The different changes of n-9/n-7 isomers of C18:1-containing GPs in breast cancer cell lines aroused our interest in discerning C=C locations in FA building blocks. As displayed in Fig. 5C, the ratio of n-9/n-7 isomers in free FA 18:1 in MCF-7 cells, MDA-MB-468 cells, and BT-474 cells showed increased variation trends, and that in MDA-MB-231 cells showed decreased variation trends, corresponding to its GP tendency. Because of 2 trends in the change of phospholipids, the change of FA 18:1 in SK-BR-3 cells is the most unpredictable. The final result showed a declining process. Combined together, these variations in n-9/n-7 isomers of C18:1-containing GPs suggested that there must be some other factors that influence the C=C location isomeric ratio, and the universality of drugs targeting SCD1 treatment should be well considered since different cancer cell subtypes respond differently even within the same cancer type. ## FAO also modulates the composition of GP C=C location isomers Based on the biosynthetic pathway of different lipid C=C location isomers (Fig. 2A), downstream of the SCD1-catalyzed lipid desaturation, C16:1 (n-9)-CoA is synthesized from C18:1 (n-9)-CoA via FAO. Therefore, we hypothesize that FAO may contribute to the regulation of C18:1 (n-9) and C18:1 (n-7). Then, we analyzed PC 16:0_16:1, which is the most abundant GPs with C16:1 (Fig. S7). Coupled with PB reaction, 2 pairs of C=C location specific diagnostic ions were detected for PC 16:0_16:1, indicating the presence of n-9 (m/z 622 and 648) to n-7 (m/z 650 and 676) isomers (Fig. 6A). Fig. 6.: The measurement of FAO reveals its potential relationship with breast cancer cell line invasiveness. (A) PB-MS/MS spectra of PC 16:0_16:1 in MCF-7 cells before and after SCD1 inhibition. (B to D) Compositional variations of C=C location isomers in C16:1-containing GPs after small-molecule inhibitor or siRNA-mediated gene silencing of SCD1. (E) The inhibition of FAO by etomoxir compensates for the decrease of n-9/n-7 isomers in C18:1-containing GPs caused by SCD1 inhibition in MDA-MB-231 cells. (F) Schematic illustration of the indicators to measure FAO activity, and the relationship between FAO and metastasis. The indicators of C16:1 n-9/n-7 ratio can be used to infer FAO activity and cancer cell invasiveness. Differences between the group samples were evaluated for statistical significance using the 2-tailed Student’s t test (P < 0.05, P < 0.01, P < 0.001). Error bar represents the standard deviation, n = 3. For MCF-7 cells, the ratio of n-9/n-7 isomers in PC 16:0_16:1 increased after SCD1 inhibition along with increased CAY10566 concentration (Fig. 6B). To provide additional evidence, SCD1 gene silencing was performed, leading to an increased level of n-9/n-7 isomer ratio in PC 16:0_16:1, consistent with the results observed after CAY10566 treatment (Fig. 6C). Interestingly, MDA-MB-231 cells have a higher proportion of n-9 isomers in C16:1-containing GPs compared to the other cell lines (Fig. 6D), suggesting that FAO is more active, which is consistent with previous research [52]. Furthermore, the increases of n-9 isomer in C16:1-containing GPs were consistently observed in all human breast cancer cell lines (Fig. 6D). Previous studies showed that the inhibition of SCD1 activates carnitine palmitoyltransferase 1 (CPT1), which is a solute carrier transporter at the outer membrane responsible for transporting FAs into mitochondria to support FAO [53–56], thus leading to increased amounts of C16:1 (n-9). To acquire additional evidence to understand the role of FAO, we analyzed MDA-MB-231 cells, which showed unexpected changes in C18:1 C=C location isomers of GPs upon SCD1 inhibition, to test whether FAO is correlated with n-9/n-7 composition in C18:1-containing GPs. We inhibited the important rate-limiting enzyme of FAO, CPT1 with etomoxir, and then analyzed C18:1 C=C location isomers of GPs. Results showed that the simultaneous inhibition of CPT1 and SCD1 can successfully increase n-9/n-7 isomer ratios in FA 18:1 and GPs with C18:1, which is consistent with others when inhibited with SCD1 only (Fig. 6E and Fig. S8). This set of experiments suggests that both lipid desaturation and β-oxidation are involved in the dynamic regulation of C18:1 C=C location isomers. It is generally believed that triple-negative breast cancer is highly aggressive and the prognosis is worse. Both MDA-MB-468 and MDA-MB-231 cells were triple-negative breast cancer cells from the pleural effusion of women with metastatic breast cancer, but their invasiveness is very different. MDA-MB-231 cells are commonly regarded as highly invasive, as evidenced by the high levels of invasion markers including CXCR4 and stem cell markers (CD44/CD24 and ALDH1+), secreted exosomes with high levels of miRNA-122 or 105, interleukin-8 (IL-8), or AKR1B10 [57–62]. However, the levels of these markers were low in MDA-MB-468 cells, which showed a low metastatic potential both in vivo and in vitro [57–60,63]. Therefore, though MDA-MB-468 and MDA-MB-231 cells are both triple-negative, the latter are more flexible in metabolic reprogramming (e.g., FAO) to promote uncontrolled growth and metastasis of cancer cells, and a clear connection of FAO with the metastatic potentials of cancer cells has been established [64–66]. Same as MDA-MB-231 cells that have higher FAO activities revealed by PB-MS/MS, SK-BR-3 cells also have higher metastasis capability, further validating the correlation between FAO and metastasis [60,61,67]. In summary, the in-depth lipidomics analysis has enabled comprehensive analysis of lipids in biological samples at a large scale. Detailed lipid structure characterization is indispensable for accurate lipid pathway mapping, and in this work, we have demonstrated an example of monitoring FAO activity using lipid analysis by quantifying lipid C=C location isomers, which implies information about the invasiveness of cancer cells (Fig. 6F). This workflow is simple, rapid, and straightforward, as it does not require direct protein detection or addition of any internal standard for the relative quantitation of lipid isomers. ## Discussion SCD1 expression is frequently found to be high in cancer development, accompanied by extensive lipid metabolism reprogramming [9,20,68]. Previous studies have explored the critical role of lipids in the occurrence and development of cancers, including tumor formation, tumor growth, and metastasis [20,69]. From the perspective of lipid unsaturation, saturated FAs were found to trigger endoplasmic reticulum stress response and slow tumor growth [9,70,71], while unsaturated FAs were correlated with cancer presence, poor prognosis, and greater death rates [72,73]. Such findings suggest that both saturated and unsaturated lipids are related to cancer and their specific functions need to be further studied. The recent progress in lipid analysis tools has importantly improved the coverage of the fine structures that can be resolved through a streamlined experiment workflow. Structural lipidomics is likely to become an essential technique to support the studies of tumor biology, biomarker discovery, and diagnostic applications. It was demonstrated that higher n-7/n-9 ratios in C18:1-containing lipids were found in colon cancer cell lines, breast cancer tissues, and lung cancer tissue [32,38,47,74,75]. Lower n-7/n-9 ratios in C18:1-containing lipids were found in the lymph node tissues with thyroid cancer metastasis [39]. Clearly, as shown by this work and others, lipid remodeling, which involves recombination and reconstruction of different carbon chain FAs, is closely correlated with these cancers and a variety of enzymes are involved in lipid metabolism. By comprehensive lipid analysis via PB-MS/MS, we have shown that SCD1 and FAO activities are high in specific subtypes of human breast cancer cell lines, both of which are correlated with cancer metastasis and invasiveness (high metabolic rate). Certainly, the detailed functional outcome of many structurally distinct lipids or a group of remodeled lipids remains to be elucidated, with the emergence of more powerful lipid analysis tools. In this work, by using LC-PB-MS/MS, we have monitored the lipidome alterations of human breast cancer cells of different subtypes, with key lipid metabolic enzymes including SCD1 inhibited. We were able to reveal how lipid desaturation is involved in regulating the lipidome at the C=C level. By exploring the different responses of different cell lines to SCD1 inhibition, we have demonstrated that lipid β-oxidation also has an important impact on lipid metabolism reprogramming. While the metabolic pathways of lipid metabolism studied here are well-known from a biological point of view, we showed that quantitative lipid C=C location isomer analysis can be conveniently used to probe the activity of key enzymes involved in the metabolism pathways. For instance, the relative amount of n-9 isomer of C16:1 is highly correlated with FAO activity and the invasiveness of human breast cancer cells, providing a new perspective for metabolic phenotyping and cancer diagnostics. This study may also contribute to the development of more effective cancer therapies through the inhibition of multiple enzyme targets, e.g., SCD1, FADS2, and CPT1, to minimize cancer recurrence and metastasis. ## Materials and chemicals Ammonium acetate and dimethyl sulfoxide (DMSO) were purchased from Sigma-Aldrich (St Louis, MO, USA). HCl was purchased from Beijing Chemical Works (Beijing, China). Ammonium hydroxide was purchased from Modern Oriental Fine Chemistry (Beijing, China). All other HPLC-grade organic solvents such as acetone, acetonitrile, water, and formic acid were purchased from Fisher Scientific (NJ, USA) and used without further purification. The CCK-8 assay kit was purchased from BioDee Biotechnology (Beijing, China). DharmaFECT 1 transfection reagent was obtained from Dharmacon Research, Inc. (Cambridge, UK). MCF-7 (catalog number 3111C0001CCC000013) was purchased from the National Infrastructure of Cell Line Resource (Beijing, China). BT-474 (catalog number 3111C0001CCC000129), SK-BR-3 (catalog number 3111C0001CCC000085), MDA-MB-231 (catalog number 3111C0001CCC000014), and MDA-MB-468 (catalog number 3111C0001CCC000249) cells were purchased from Shanghai Enzyme Research Biotechnology Co. Ltd (Shanghai, China). Dulbecco’s modified Eagle’s medium (DMEM), Roswell Park Memorial Institute-1640 medium (RPMI-1640, w/o Hepes), Leibovitz’s L-15 Medium (L-15), DPBS, fetal bovine serum (FBS), and penicillin−streptomycin (100 U·ml−1) were purchased from Gibco (Life Technologies, Carlsbad, CA). Small-molecule inhibitors, CAY10566, and etomoxir were purchased from MedChem Express (Monmouth Junction, USA), dissolved in DMSO, and stored at −80 °C. ## Cell culture MCF-7 and SK-BR-3 cells were cultured in DMEM supplemented with $10\%$ FBS and $1\%$ penicillin–streptomycin. BT-474 cells were cultured in RPMI-1640 medium supplemented with $10\%$ FBS and $1\%$ penicillin–streptomycin. These 3 types of cells were cultured with a breathable dish in a humidified atmosphere containing $5\%$ CO2. MDA-MB-468 and MDA-MB-231 cells were cultured in L-15 medium with the same supplements and in sealed culturing dishes. All cells were cultured at 37 °C and passaged every 2 or 3 days. After reaching $90\%$ confluence, cells were detached using $0.1\%$ trypsin solution and collected by centrifugation. Then, 1 ml of methanol was immediately added to quench cellular activity. As for SCD1 and CPT1 inhibition, cells were treated with 0, 10, 100, 300, 500, 800, and 1,000 nM CAY10566, or 30 μM etomoxir in DMSO, while the control group was treated with an equal volume of DMSO. ## Lipid extraction A modified Folch method was employed for phospholipid extraction from cultured cells. The cell suspension in methanol (1 ml) was sonicated in a water bath for 10 min and then transferred to a 10-ml centrifuge tube diluted with 1 ml of deionized water and 2 ml of chloroform. After 30 s vortex, the mixture was centrifuged at 10,000 g for 10 min. The bottom chloroform layer was collected in a borosilicate glass tube. Another 2 ml of chloroform was added to the residual top layer to repeat the extraction process. The bottom layer of chloroform was collected and combined with the previously collected fraction and dried under nitrogen flow. Finally, the dried lipid extract was redissolved with 1 ml of methanol and stored at −20 °C before MS analysis. For free FA extraction, the cell suspension in methanol (1 ml) was sonicated in a water bath for 10 min and then transferred to a 10-ml centrifuge tube. The sample was diluted with 300 μl of deionized water and acidified with HCl to 25 mM final concentration. After adding 1 ml of isooctane, the mixture was vortexed and centrifuged at 3,000 g for 1 min. The top layer was transferred to a glass tube and dried under nitrogen flow. Finally, extracted free FAs were redissolved with 100 μl of methanol and stored at −20 °C before MS analysis. ## Lipidomics analysis The analysis of phospholipids was performed on an LC–PB–MS/MS system, including a 4500 QTRAP triple quadrupole/linear ion trap hybrid mass spectrometer (Applied Biosystems/Sciex, Toronto, Canada), an ExionLC AC system (Sciex, Toronto, CA), and a home-built flow microreactor. The flow microreactor used for post-column online PB derivation is made from fluorinated ethylene propylene tubing (0.03-in. internal diameter, $\frac{1}{16}$ outer diameter) and a low-pressure mercury lamp with emission centered around 254 nm (Model No.: 80-1057-01, BHK, Inc., CA, USA). Lipid separation was performed on a hydrophilic interaction chromatography (HILIC) column (150 mm × 2.1 mm, silica spheres, 2.7 μm) from Sigma-Aldrich (St. Louis, MO, USA). The column temperature was set at 30 °C. The mobile phase consisted of (A) acetone/ACN/HAc ($\frac{50}{50}$/0.2, v/v/v) and (B) ammonium acetate aqueous solution (10 mM). Gradient elution was applied for separation (A started from $90\%$, decreased to $85\%$ at 5 min, then decreased to $80\%$ at 8 min, kept at $80\%$ within 8 to 15 min, and decreased to $70\%$ in 16 min and kept this percentage to 20 min) at a flow rate of 0.2 ml min−1. For MS analysis, each sample was analyzed by 3 steps. Based on the first step, LC-MS/MS, i.e., neutral loss scan (NLS) 141 for PEs, NLS 185 for PSs, and precursor ion scan (PIS) 184 for PCs, a list of potential PB precursors was provided to guide the second online LC-PB-MS/MS step. For n-10 isomer validation in BT-474 cells, LC-PB-MS/MS/MS was applied. The third step was GP chain information analysis, which was conducted by LC-MS/MS in negative ion mode. For the preliminary screen of C16:1- and C18:1-containing GPs, PIS or NLS was applied in negative ion mode or positive ion mode in the presence of 1 mM LiOH, which was added to eliminate counterparts. The analysis of free FA was conducted by online PB reaction coupled with nano-electrospray ionization (nanoESI). The nanoESI tip was aligned with the sampling orifice of the mass spectrometer. To initiate PB reaction, a low-pressure mercury lamp was set 1.0 cm over the tip. Before PB reaction, the FA extract was dried and redissolved in acetone:water (7:3, v/v) added with $0.1\%$ ammonium hydroxide. ## siRNA-mediated gene knockdown strategy MCF-7 cells were suspended in DMEM without FBS and then transfected using Sangon Biotech siRNA (75nM final concentration) via DharmaFECT 1 transfection reagent using the manufacturer’s protocol in 6-well format. After 24 h, transfected cells were cultured in a complete DMEM for another 24 h and collected for analysis. Knockdown efficiency was approximately $65\%$ as measured by quantitative real-time polymerase chain reaction (PCR) analysis after 48 h transfection. ## CCK-8 assay The cytotoxicity of CAY10566 was evaluated by CCK-8 assay. Briefly, MCF-7 cells were seeded into 96-well plates at a density of 2,000 cells/well and a volume of 100 μl. Next, 10 μl per well of CCK-8 solution was added. Then, the cells were cultured for another 2 h at 37 °C and the absorbance at 450 nm was recorded according to the manufacturer’s instructions. ## Data analysis For the relative quantitative analysis of GPs C=C location isomers, the peak areas of corresponding diagnostic ions were used. Hierarchical cluster analysis was conducted using HemI67 software (version 1.0.3.7) [76], with Pearson's correlation coefficient for distance measurements and average linkage (default) variance in the clustering method. ## Data Availability All data used to support the findings of this study are available from the corresponding author upon request. ## References 1. Meier U, Gressner AM. Endocrine regulation of energy metabolism: Review of pathobiochemical and clinical chemical aspects of leptin, ghrelin, adiponectin, and resistin. Clin Chem (2004) 50 1511-1525. PMID: 15265818 2. Wymann MP, Schneiter R. Lipid signalling in disease. Nat Rev Mol Cell Biol (2008) 9 162-176. PMID: 18216772 3. Lingwood D, Simons K. Lipid rafts as a membrane-organizing principle. Science (2010) 327 46-50. PMID: 20044567 4. Miao H, Li B, Wang Z, Mu J, Tian Y, Jiang B, Zhang S, Gong X, Shui G, Lam SM. Lipidome atlas of the developing heart uncovers dynamic membrane lipid attributes underlying cardiac structural and metabolic maturation. Research (2022) 2022 research.0006 5. Wang Z, Klipfell E, Bennett BJ, Koeth R, Levison BS, DuGar B, Feldstein AE, Britt EB, Fu X, Chung YM. Gut flora metabolism of phosphatidylcholine promotes cardiovascular disease. Nature (2011) 472 57-63. PMID: 21475195 6. Meikle PJ, Wong G, Barlow CK, Kingwell BA. Lipidomics: Potential role in risk prediction and therapeutic monitoring for diabetes and cardiovascular disease. Pharmacol Ther (2014) 143 12-23. PMID: 24509229 7. Vriens K, Christen S, Parik S, Broekaert D, Yoshinaga K, Talebi A, Dehairs J, Escalona-Noguero C, Schmieder R, Cornfield T. Evidence for an alternative fatty acid desaturation pathway increasing cancer plasticity. Nature (2019) 566 403-406. PMID: 30728499 8. Zhang W, Zhang D, Chen Q, Wu J, Ouyang Z, Xia Y. Online photochemical derivatization enables comprehensive mass spectrometric analysis of unsaturated phospholipid isomers. Nat Commun (2019) 10 79. PMID: 30622271 9. Li J, Condello S, Thomes-Pepin J, Ma X, Xia Y, Hurley TD, Matei D, Cheng JX. Lipid desaturation is a metabolic marker and therapeutic target of ovarian cancer stem cells. Cell Stem Cell (2017) 20 303-314.e5. PMID: 28041894 10. Röhrig F, Schulze A. The multifaceted roles of fatty acid synthesis in cancer. Nat Rev Cancer (2016) 16 732-749. PMID: 27658529 11. Lam SM, Zhang C, Wang Z, Ni Z, Zhang S, Yang S, Huang X, Mo L, Li J, Lee B. A multi-omics investigation of the composition and function of extracellular vesicles along the temporal trajectory of COVID-19. Nat Metab (2021) 3 909-922. PMID: 34158670 12. Lee HJ, Chen Z, Collard M, Chen F, Chen JG, Wu M, Alani RM, Cheng J-X. Multimodal metabolic imaging reveals pigment reduction and lipid accumulation in metastatic melanoma. BMEF (2021) 2021 9860123 13. Chen L, Zhou Z, Hu C, Maitz MF, Yang L, Luo R, Wang Y. Platelet membrane-coated nanocarriers targeting plaques to deliver anti-CD47 antibody for atherosclerotic therapy. Research (2022) 2022 14. Wang J, Xu J, Zang G, Zhang T, Wu Q, Zhang H, Chen Y, Wang Y, Qin W, Zhao S. Research (2022) 2022 9839368. PMID: 35465346 15. Faubert B, Solmonson A, DeBerardinis RJ. Metabolic reprogramming and cancer progression. Science (2020) 368 eaaw5473. PMID: 32273439 16. Bergers G, Fendt S-M. The metabolism of cancer cells during metastasis. Nat Rev Cancer (2021) 21 162-180. PMID: 33462499 17. Pascual G, Avgustinova A, Mejetta S, Martín M, Castellanos A, Attolini CSO, Berenguer A, Prats N, Toll A, Hueto JA. Targeting metastasis-initiating cells through the fatty acid receptor CD36. Nature (2017) 541 41-45. PMID: 27974793 18. Zhang M, di Martino JS, Bowman RL, Campbell NR, Baksh SC, Simon-Vermot T, Kim IS, Haldeman P, Mondal C, Yong-Gonzales V. Adipocyte-derived lipids mediate melanoma progression via FATP proteins. Cancer Discov (2018) 8 1006-1025. PMID: 29903879 19. Wood CB, Habib NA, Thompson A, Bradpiece H, Smadja C, Hershman M, Barker W, Apostolov K. Increase of oleic acid in erythrocytes associated with malignancies. Br Med J (1985) 291 163-165. PMID: 3926106 20. Igal RA. Stearoyl CoA desaturase-1: New insights into a central regulator of cancer metabolism. Biochim Biophys Acta (2016) 1861 1865-1880. PMID: 27639967 21. Marien E, Meister M, Muley T, Pulgar T, Swinnen JV. Phospholipid profiling identifies acyl chain elongation as a ubiquitous trait and potential target for the treatment of lung squamous cell carcinoma. Oncotarget (2016) 7 12582-12597. PMID: 26862848 22. Jafari N, Drury J, Morris AJ, Onono FO, Stevens PD, Gao T, Liu J, Wang C, Lee EY, Weiss HL. De novo fatty acid synthesis-driven sphingolipid metabolism promotes metastatic potential of colorectal cancer. Mol Cancer Res (2019) 17 140-152. PMID: 30154249 23. Young RSE, Bowman AP, Williams ED, Tousignant KD, Bidgood CL, Narreddula VR, Gupta R, Marshall DL, Poad BLJ, Nelson CC. Apocryphal FADS2 activity promotes fatty acid diversification in cancer. Cell Rep (2021) 34 108738. PMID: 33567271 24. Lin Q, Li P, Fang M, Zhang D, Xia Y. Deep profiling of aminophospholipids reveals a dysregulated desaturation pattern in breast cancer cell lines. Anal Chem (2022) 94 820-828. PMID: 34931817 25. Thomas MC, Mitchell TW, Harman DG, Deeley JM, Nealon JR, Blanksby SJ. Ozone-induced dissociation: Elucidation of double bond position within mass-selected lipid ions. Anal Chem (2008) 80 303-311. PMID: 18062677 26. Pham HT, Maccarone AT, Thomas MC, Campbell JL, Mitchell TW, Blanksby SJ. Structural characterization of glycerophospholipids by combinations of ozone- and collision-induced dissociation mass spectrometry: The next step towards "top-down" lipidomics. Analyst (2014) 139 204-214. PMID: 24244938 27. Baba T, Campbell JL, Le Blanc JCY, Baker PRS. Distinguishing cis and trans isomers in intact complex lipids using electron impact excitation of ions from organics mass spectrometry. Anal Chem (2017) 89 7307-7315. PMID: 28613874 28. Campbell JL, Baba T. Near-complete structural characterization of phosphatidylcholines using electron impact excitation of ions from organics. Anal Chem (2015) 87 5837-5845. PMID: 25955306 29. Brodbelt JS, Morrison LJ, Santos I. Ultraviolet photodissociation mass spectrometry for analysis of biological molecules. Chem Rev (2020) 120 3328-3380. PMID: 31851501 30. Williams PE, Klein DR, Greer SM, Brodbelt JS. Pinpointing double bond and sn-positions in glycerophospholipids via hybrid 193 nm ultraviolet photodissociation (UVPD) mass spectrometry. J Am Chem Soc (2017) 139 15681-15690. PMID: 28988476 31. Ma X, Xia Y. Pinpointing double bonds in lipids by Paternò-Büchi reactions and mass spectrometry. Angew Chem (2014) 53 2592-2596. PMID: 24500881 32. Cao W, Cheng S, Yang J, Feng J, Zhang W, Li Z, Chen Q, Xia Y, Ouyang Z, Ma X. Large-scale lipid analysis with C=C location and sn-position isomer resolving power. Nat Commun (2020) 11 375-375. PMID: 31953382 33. Feng Y, Chen B, Yu Q, Li L. Identification of double bond position isomers in unsaturated lipids by m-CPBA epoxidation and mass spectrometry fragmentation. Anal Chem (2019) 91 1791-1795. PMID: 30608661 34. Zhao YY, Zhao H, Zhao X, Jia J, Ma Q, Zhang S, Zhang X, Chiba H, Hui SP, Ma X. Identification and quantitation of C=C location isomers of unsaturated fatty acids by epoxidation reaction and tandem mass spectrometry. Anal Chem (2017) 89 10270-10278. PMID: 28837768 35. Laskin J, Unsihuay D, Su P, Hu H, Dey SK. Imaging and analysis of isomeric unsaturated lipids through online photochemical derivatization of C=C bonds. Angew Chem (2021) 29 7559-7563 36. Li H-F, Zhao J, Cao W, Zhang W, Xia Y, Ouyang Z. Site-specific photochemical reaction for improved C=C location analysis of unsaturated lipids by ultraviolet photodissociation. Research (2022) 2022 9783602. PMID: 35252873 37. Zhang W, Jian R, Zhao J, Liu Y, Xia Y. Deep-lipidotyping by mass spectrometry: Recent technical advances and applications. J Lipid Res (2022) 63 100219. PMID: 35489417 38. Fang M, Rustam Y, Palmieri M, Sieber OM, Reid GE. Evaluation of ultraviolet photodissociation tandem mass spectrometry for the structural assignment of unsaturated fatty acid double bond positional isomers. Anal Bioanal Chem (2020) 412 2339-2351. PMID: 32006064 39. Klein DR, Feider CL, Garza KY, Lin JQ, Eberlin LS, Brodbelt JS. Desorption electrospray ionization coupled with ultraviolet photodissociation for characterization of phospholipid isomers in tissue sections. Anal Chem (2018) 90 10100-10104. PMID: 30080398 40. Li Z, Cheng S, Lin Q, Cao W, Yang J, Zhang M, Shen A, Zhang W, Xia Y, Ma X. Single-cell lipidomics with high structural specificity by mass spectrometry. Nat Commun (2021) 12 2869. PMID: 34001877 41. Ma X, Zhao X, Li J, Zhang W, Cheng JX, Ouyang Z, Xia Y. Photochemical tagging for quantitation of unsaturated fatty acids by mass spectrometry. Anal Chem (2016) 88 8931-8935. PMID: 27560604 42. Mohammadzadeh F, Mosayebi G, Montazeri V, Darabi M, Fayezi S, Shaaker M, Rahmati M, Baradaran B, Mehdizadeh A, Darabi M. Fatty acid composition of tissue cultured breast carcinoma and the effect of stearoyl-CoA desaturase 1 inhibition. J Breast Cancer (2014) 17 136-142. PMID: 25013434 43. Ackerman D, Tumanov S, Qiu B, Michalopoulou E, Spata M, Azzam A, Xie H, Simon MC, Kamphorst JJ. Triglycerides promote lipid homeostasis during hypoxic stress by balancing fatty acid saturation. Cell Rep (2018) 24 2596-2605.e5. PMID: 30184495 44. Vincent BM, Tardiff DF, Piotrowski JS, Aron R, Lucas MC, Chung CY, Bacherman H, Chen YQ, Pires M, Subramaniam R. Inhibiting stearoyl-CoA desaturase ameliorates α-synuclein cytotoxicity. Cell Rep (2018) 25 2742-2754.e31. PMID: 30517862 45. Ollila S, Hyvönen MT, Vattulainen I. Polyunsaturation in lipid membranes: Dynamic properties and lateral pressure profiles. J Phys Chem B (2007) 111 3139-3150. PMID: 17388448 46. Rysman E, Brusselmans K, Scheys K, Timmermans L, Derua R, Munck S, van Veldhoven PP, Waltregny D, Daniëls VW, Machiels J. De novo lipogenesis protects cancer cells from free radicals and chemotherapeutics by promoting membrane lipid saturation. Cancer Res (2010) 70 8117-8126. PMID: 20876798 47. Kuo T-H, Chung HH, Chang HY, Lin CW, Wang MY, Shen TL, Hsu CC. Deep lipidomics and molecular imaging of unsaturated lipid isomers: A universal strategy initiated by mCPBA epoxidation. Anal Chem (2019) 91 11905-11915. PMID: 31408322 48. Wang J, Yu L, Schmidt RE, Su C, Huang X, Gould K, Cao G. Characterization of HSCD5, a novel human stearoyl-CoA desaturase unique to primates. Biochem Biophys Res Commun (2005) 332 735-742. PMID: 15907797 49. Kulkoyluoglu-Cotul E, Arca A, Madak-Erdogan Z. Crosstalk between estrogen signaling and breast cancer metabolism. Trends Endocrinol Metab (2019) 30 25-38. PMID: 30471920 50. Ruihua W, Hansen Z, Xiaochao Z, Xu Z, Song J. Metabolic discrimination of breast cancer subtypes at the single-cell level by multiple microextraction coupled with mass spectrometry. Anal Chem (2019) 91 3667-3674. PMID: 30702862 51. He M, Guo S, Li Z. In situ characterizing membrane lipid phenotype of breast cancer cells using mass spectrometry profiling. Sci Rep (2015) 5 11298. PMID: 26061164 52. Linher-Melville K, Zantinge S, Sanli T, Gerstein H, Tsakiridis T, Singh G. Establishing a relationship between prolactin and altered fatty acid β-oxidation via carnitine palmitoyl transferase 1 in breast cancer cells. BMC Cancer (2011) 11 56. PMID: 21294903 53. Thörn K, Hovsepyan M, Bergsten P. Reduced levels of SCD1 accentuate palmitate-induced stress in insulin-producing β-cells. Lipids Health Dis (2010) 9 1-8. PMID: 20053284 54. Dobrzyn P, Dobrzyn A, Miyazaki M, Cohen P, Asilmaz E, Hardie DG, Friedman JM, Ntambi JM. Stearoyl-CoA desaturase 1 deficiency increases fatty acid oxidation by activating AMP-activated protein kinase in liver. Proc Natl Acad Sci USA (2004) 101 6409-6414. PMID: 15096593 55. Dobrzyn A, Dobrzyn P, Lee SH, Miyazaki M, Ntambi JM. Stearoyl-CoA desaturase-1 deficiency reduces ceramide synthesis by downregulating serine palmitoyltransferase and increasing β-oxidation in skeletal muscle. Am J Physiol Endocrinol Metab (2005) 288 E599-E607. PMID: 15562249 56. Chen R, Chen L. Solute carrier transporters: Emerging central players in tumour immunotherapy. Trends Cell Biol (2021) 186-201. PMID: 34511324 57. Abraham BK, Fritz P, Mcclellan MJCCR. Prevalence of CD44+/CD24/low cells in breast cancer may not be associated with clinical outcome but may favor distant metastasis. Clin Cancer Res (2005) 11 1154-1159. PMID: 15709183 58. Freund A, Chauveau C, Brouillet JP, Lucas A, Lacroix M, Licznar A, Vignon F, Lazennec G. IL-8 expression and its possible relationship with estrogen-receptor-negative status of breast cancer cells. Oncogene (2003) 22 256-265. PMID: 12527894 59. Li W, Ma H, Zhang J, Zhu L, Wang C, Yang Y. Unraveling the roles of CD44/CD24 and ALDH1 as cancer stem cell markers in tumorigenesis and metastasis. Sci Rep (2017) 7 13856. PMID: 29062075 60. Fong MY, Zhou W, Liu L, Alontaga AY, Chandra M, Ashby J, Chow A, O’Connor STF, Li S, Chin AR. Breast-cancer-secreted miR-122 reprograms glucose metabolism in premetastatic niche to promote metastasis. Nat Cell Biol (2015) 17 183-194. PMID: 25621950 61. Zhou W, Fong MY, Min Y, Somlo G, Liu L, Palomares MR, Yu Y, Chow A, O’Connor STF, Chin AR. Cancer-secreted miR-105 destroys vascular endothelial barriers to promote metastasis. Cancer Cell (2014) 25 501-515. PMID: 24735924 62. Weverwijk A, Koundouros N, Iravani M, Ashenden M, Gao Q, Poulogiannis G, Jungwirth U, Isacke CM. Metabolic adaptability in metastatic breast cancer by AKR1B10-dependent balancing of glycolysis and fatty acid oxidation. Nat Commun (2019) 10 2698. PMID: 31221959 63. Sheridan C, Kishimoto H, Fuchs RK, Mehrotra S, Bhat-Nakshatri P, Turner CH, Goulet R, Badve S, Nakshatri H. CD44+/CD24-breast cancer cells exhibit enhanced invasive properties: An early step necessary for metastasis. Breast Cancer Res (2006) 8 R59. PMID: 17062128 64. Li M, Xian H-c, Tang Y-J, Liang X-h, Tang Y-l. Fatty acid oxidation: Driver of lymph node metastasis. Cancer Cell Int (2021) 21 339. PMID: 34217300 65. Wang C, Shao L, Pan C, Ye J, Ding Z, Wu J, du Q, Ren Y, Zhu C. Elevated level of mitochondrial reactive oxygen species via fatty acid β-oxidation in cancer stem cells promotes cancer metastasis by inducing epithelial–mesenchymal transition. Stem Cell Res Ther (2019) 10 175. PMID: 31196164 66. Loo Ser Y, Toh LP, Xie WH, Pathak E, Tan W, Ma S, Lee MY, Shatishwaran S, Yeo JZZ, Yuan J. Fatty acid oxidation is a druggable gateway regulating cellular plasticity for driving metastasis in breast cancer. Sci Adv (2021) 7 eabh2443. PMID: 34613780 67. Wu X, Somlo G, Yu Y, Palomares MR, Li AX, Zhou W, Chow A, Yen Y, Rossi JJ, Gao H. De novo sequencing of circulating miRNAs identifies novel markers predicting clinical outcome of locally advanced breast cancer. J Transl Med (2012) 10 42. PMID: 22400902 68. Vivas-García Y, Falletta P, Liebing J, Louphrasitthiphol P, Feng Y, Chauhan J, Scott DA, Glodde N, Chocarro-Calvo A, Bonham S. Lineage-restricted regulation of SCD and fatty acid saturation by MITF controls melanoma phenotypic plasticity. Mol Cell (2020) 77 120-137.e9. PMID: 31733993 69. Tracz-Gaszewska Z, Dobrzyn P. Stearoyl-CoA desaturase 1 as a therapeutic target for the treatment of cancer. Cancers (2019) 11 948. PMID: 31284458 70. Zhao G, Tan Y, Cardenas H, Vayngart D, Wang Y, Huang H, Keathley R, Wei JJ, Ferreira CR, Orsulic S. Ovarian cancer cell fate regulation by the dynamics between saturated and unsaturated fatty acids. Proc Natl Acad Sci USA (2022) 119 e2203480119. PMID: 36197994 71. Lien EC, Westermark AM, Zhang Y, Yuan C, Li Z, Lau AN, Sapp KM, Wolpin BM, Vander Heiden MG. Low glycaemic diets alter lipid metabolism to influence tumour growth. Nature (2021) 599 302-307. PMID: 34671163 72. Holder AM, Gonzalez-Angulo AM, Chen H, Akcakanat A, do KA, Fraser Symmans W, Pusztai L, Hortobagyi GN, Mills GB, Meric-Bernstam F. High stearoyl-CoA desaturase 1 expression is associated with shorter survival in breast cancer patients. Breast Cancer Res Treat (2013) 137 319-327. PMID: 23208590 73. Chajès V, Joulin V, Clavel-Chapelon F. The fatty acid desaturation index of blood lipids, as a biomarker of hepatic stearoyl-CoA desaturase expression, is a predictive factor of breast cancer risk. Curr Opin Lipidol (2011) 22 6-10. PMID: 20935562 74. Tang F, Guo C, Ma X, Zhang J, Su Y, Tian R, Shi R, Xia Y, Wang X, Ouyang Z. Rapid in situ profiling of lipid C═C location isomers in tissue using ambient mass spectrometry with photochemical reactions. Anal Chem (2018) 90 5612-5619. PMID: 29624380 75. Ma X, Chong L, Tian R, Shi R, Hu TY, Ouyang Z, Xia Y. Identification and quantitation of lipid C=C location isomers: A shotgun lipidomics approach enabled by photochemical reaction. Proc Natl Acad Sci USA (2016) 113 2573. PMID: 26903636 76. Deng W, Wang Y, Liu Z, Cheng H, Xue Y. HemI: A toolkit for illustrating heatmaps. PLOS ONE (2014) 9 e111988. PMID: 25372567
--- title: IL-17/CXCL5 signaling within the oligovascular niche mediates human and mouse white matter injury authors: - Guanxi Xiao - Rosie Kumar - Yutaro Komuro - Jasmine Burguet - Visesha Kakarla - Ida Azizkhanian - Sunil A. Sheth - Christopher K. Williams - Xinhai R. Zhang - Michal Macknicki - Andrew Brumm - Riki Kawaguchi - Phu Mai - Naoki Kaneko - Harry V. Vinters - S. Thomas Carmichael - Leif A. Havton - Charles DeCarli - Jason D. Hinman journal: Cell reports year: 2023 pmcid: PMC10026849 doi: 10.1016/j.celrep.2022.111848 license: CC BY 4.0 --- # IL-17/CXCL5 signaling within the oligovascular niche mediates human and mouse white matter injury ## SUMMARY Cerebral small vessel disease and brain white matter injury are worsened by cardiovascular risk factors including obesity. Molecular pathways in cerebral endothelial cells activated by chronic cerebrovascular risk factors alter cell-cell signaling, blocking endogenous and post-ischemic white matter repair. Using cell-specific translating ribosome affinity purification (RiboTag) in white matter endothelia and oligodendrocyte progenitor cells (OPCs), we identify a coordinated interleukin-chemokine signaling cascade within the oligovascular niche of subcortical white matter that is triggered by diet-induced obesity (DIO). DIO induces interleukin-17B (IL-17B) signaling that acts on the cerebral endothelia through IL-17Rb to increase both circulating and local endothelial expression of CXCL5. In white matter endothelia, CXCL5 promotes the association of OPCs with the vasculature and triggers OPC gene expression programs regulating cell migration through chemokine signaling. Targeted blockade of IL-17B reduced vessel-associated OPCs by reducing endothelial CXCL5 expression. In multiple human cohorts, blood levels of CXCL5 function as a diagnostic and prognostic biomarker of vascular cognitive impairment. ## In brief Xiao et al. demonstrate that a high-fat diet disrupts brain white matter and exacerbates the response to a subcortical ischemic stroke by $30\%$. Using cell-specific gene expression studies, they show that this occurs through dysregulated immune signaling between blood vessels and oligodendrocyte progenitor cells acting through IL-17/CXCL5 signaling. ## Graphical Abstract ## INTRODUCTION Cerebral small vessel disease is an age-related entity affecting brain white matter. The resulting white matter lesions accumulate over time1 and contribute to disability,2 dementia,3–5 and death.6 Cerebral small vessel injury is significantly worsened by chronic cardiovascular risk factors such as hypertension, diabetes, and obesity.7–10 In particular, abdominal obesity and its associated metabolic disturbances in blood pressure, lipids, and blood sugar control increase the risk of developing white matter lesions on magnetic resonance imaging (MRI)11–14 and increase the likelihood of lacunar brain infarction or stroke.15 While the pathologic changes associated with cerebral small vessel disease are well known,16,17 the molecular pathways that drive small vessel injury in the brain are largely unknown. Emerging data suggest that an interaction between cerebral vessels and cells of the oligodendrocyte lineage play a key role in maintaining white matter homeostasis.18–20 A subset of platelet-derived growth factor receptor alpha-positive (PDGFRα+) oligodendrocyte progenitor cells (OPCs) closely associate with the vasculature21,22 and use it to migrate in the brain during development.23 Proteins secreted by endothelial cells promote OPC migration and proliferation in vitro.24,25 In the spontaneously hypertensive rat model of cerebral small vessel disease, the OPC population is increased in association with vascular changes, and delays in OPC maturation may be mediated by endothelial secretion of HSP90α.26 Both the diagnosis and treatment of cerebral small vessel disease would be advanced by identifying additional molecular pathways active in cerebral endothelia and driven by chronic cardiovascular risk factors.27 To identify molecular pathways active in the oligovascular niche triggered by chronic cardiovascular risk factors, we used a mouse model of diet-induced obesity (DIO)28 that recapitulates a number of features of human cardiovascular risk.29 We combined this DIO model with a mouse model of subcortical white matter stroke that mimics human lacunar stroke.30,31 In this combined DIO-stroke model, we show that stroke-responsive OPCs are more numerous and persistent after stroke and that post-stroke white matter repair is compromised by DIO. We then use cell-specific translating ribosome affinity purification and RNA sequencing in Tie2-Cre:RiboTag and PDGFRα-CreERT2:RiboTag mice to identify the oligovascular transcriptome after the onset of DIO. This approach led to the identification of an oligovascular signaling cascade acting through the interleukin-17B (IL-17B)/IL-17 receptor b (Rb) isoforms of the IL-17 family in chronically injured cerebral endothelial cells to increase endothelial CXCL5, which can exert paracrine signaling on OPCs. We hypothesized that this coordinated intercellular signaling cascade could drive endothelial-OPC interactions before and after stroke. Further, we speculated that this DIO-induced signaling cascade could act as a functional biomarker for human cerebral small vessel disease and vascular cognitive impairment. Here, we present evidence that IL-17B/IL-17Rb/CXCL5 signaling is activated by a recognized chronic cerebrovascular risk factor, drives intercellular signaling within the oligovascular niche, and marks a subset of human subjects at risk for vascular cognitive impairment. These findings have direct implications for the understanding of human cerebral small vessel disease. ## DIO damages white matter microvasculature and promotes endothelial-OPC interactions Obesity is a significant risk factor for the development of small vessel disease and white matter injury.10,11,13,14 We used a well-established model of DIO28 to model the effects of chronic cardiovascular risk on brain white matter and the vasculature using Tie2-Cre;tdTomato (Ai14) transgenic mice. After 12 weeks on the dietary intervention, mice on control-fat diet (CFD) gained 5.84 ± 0.78 g, while mice on high-fat diet (HFD) gained 23.2 ± 0.76 g, corresponding to a $71\%$ relative weight gain ($$p \leq 0.0004$$). HFD mice also exhibited metabolic disturbances in cholesterol and blood sugar (Figure S1) consistent with the diagnostic criteria for metabolic syndrome.32 Similar changes in weight were induced by DIO in several other transgenic strains used in this study. At 20 weeks of age after the development of obesity, we examined the vasculature and cellular makeup of the white matter. In Tie2-Cre;tdTomato (Ai14) transgenic mice, DIO reduces the volume of tdTomato (tdT)+ vessels and the branch complexity of the vasculature within subcortical white matter ($26\%$, $$p \leq 0.0069$$, and $15.4\%$, $$p \leq 0.0032$$, respectively) (Figures 1A and 1B). In addition to the DIO-induced reduction in white matter microvasculature, we also observed an increase in the percentage of PDGFRα+ OPCs within the corpus callosum (PDGFRα+/DAPI+, $5.01\%$ ± $0.13\%$ versus 5.66 ± $0.22\%$; $$p \leq 0.014$$) and a concordant increase in OPCs associated with vessels, measured as OPCs per unit vessel length (6.46 ± 0.23 versus 8.94 ± 0.31 cells/mm; $p \leq 0.0001$) (Figures 1C and 1D). Notably, DIO also appeared to alter the morphology of OPCs from a predominantly stellate morphology to an intermediate and/or perivascular cell type as reported by Kishida et al.21 This change in OPCs occurs in the absence of difference in the percentage of GST-π-+ mature oligodendrocytes (Figures S2A and S2B) but is associated with thinner myelin sheaths and an increase in the average g ratio (0.88 versus 0.80; *$$p \leq 0.002$$) in DIO mice (Figure 1E). Using a direct RNA hybridization gene expression assay for oligodendrocyte stages, we find that DIO drives an immature OPC-like gene expression profile in white matter compared with control (Figures S2C and S2D; Table S1), suggesting that DIO may compromise myelination by lineage restriction of OPCs. ## DIO impairs post-stroke remyelination The major pathologic consequence of advanced cerebral small vessel disease is subcortical ischemic injury to the white matter. To determine the effect of DIO on ischemic white matter injury, we used an established model of white matter stroke produced by focal stereotactic injection of an eNOS inhibitor producing a permanent focal region of ischemia.30,33 At 7 days after white matter stroke, there was no significant difference in the stroke lesion volume when comparing animals on CFD versus HFD ($$p \leq 0.31$$) (Figure 2A). This ischemic white matter lesion results in a distinct population of stroke-responsive PDGFRα+ OPCs.31,34 In DIO mice, PDGFRα+ stroke-responsive OPCs per lesion were increased compared with control at 7 days post-stroke (Figure 2B). Spatial mapping of stroke-responsive OPCs coupled with nearest neighbor comparative analysis indicates a greater distribution of stroke-responsive OPCs specifically at the peri-infarct lesion margins in DIO mice compared with control (Figures 2B and S3). To determine if DIO impairs OPC differentiation after stroke, we compared PDGFRα+ OPC and GST-π+ mature oligodendrocyte cell counts in three regions of interest spanning the ischemic white matter lesion at 28 days post-stroke. DIO drives a significant change in oligodendrocyte cell populations 28 days after stroke ($$p \leq 0.0011$$, two-way ANOVA, $F = 14.47$) (Figure 2C). Residual stroke-responsive PDGFRα+ OPCs were present at 28 days post-stroke in animals on HFD compared with those on CFD (adjusted $$p \leq 0.0114$$). The number of GST-π+ mature oligodendrocytes within the lesion at 28 days post-stroke was variable and generally reduced in animals on HFD compared with those on CFD (adjusted $$p \leq 0.0654$$). To further assess the effect of DIO on white matter stroke remyelination, we measured peri-infarct myelin basic protein as a function of distance from the stroke core. This measure of functional remyelination after stroke demonstrates reduced peri-infarct MBP+ immunoreactivity at 28 days post-stroke in DIO mice compared with control mice ($p \leq 0.0001$, two-way ANOVA, $F = 3.11$) (Figure 2D), indicating a failure of post-stroke remyelination. ## Molecular profiling of white matter endothelia and OPCs using RiboTAG To identify the molecular pathways induced by DIO that could drive abnormal endothelial-OPC signaling and thereby impair baseline and post-stroke remyelination, we used a cell-specific RiboTAG approach employing Tie2-Cre:RiboTag and PDGFRα-CreERT2:RiboTag mice to tag ribosomes in endothelia and OPCs, respectively, enabling translating ribosome affinity purification (TRAP)35 (Figure 3A). Tie2-Cre:RiboTag mice show robust hemagglutinin (HA) labeling in the cerebrovasculature (Figure 3B). RNA sequencing (RNA-seq) analysis of immunoprecipitated HA+ ribosomes from Tie2-Cre: RiboTag mice show endothelial specificity, with a specific enrichment of endothelial transcripts compared with established marker genes for other perivascular cells including pericytes and OPCs.36 Similarly, PDGFRα-CreERT2:RiboTag mice show significant HA expression in OPCs 4 days after induction with tamoxifen, and OPC transcripts are enriched after TRAP-seq (Figure 3C). In both RiboTag strains, DIO results in a specific gene expression profile (Figures 3D and S4, Tables S2 and S3). Compared with white matter endothelial cells from normal-weight mice, DIO induced 112 up-regulated genes and 60 down-regulated genes (false discovery rate [FDR] < 0.1). Gene Ontology of the up-regulated endothelial genes points to DIO enrichment of immune signaling pathways including C-X-C chemokine signaling and IL receptor activation within white matter endothelia (Figure 3E). Among the top differentially regulated genes, IL17Rb (8.83-fold increase, FDR = 0.090) and its effector chemokine Cxcl5 (11.35-fold increase, FDR = 0.064) were strongly up-regulated genes when comparing DIO versus control animals (Figure 3D) and suggests a cognate inflammatory signaling pathway specific to DIO in white matter endothelia. Furthermore, with the known role of chemokine receptor (CXCR) signaling on OPC migration,23 we reasoned that endothelial up-regulation of an IL-17Rb/CXCL5 signaling cascade in DIO mice may function to promote OPC migration to the vasculature. Gene Ontology of the differentially expressed genes (DEGs) induced in HA+ OPCs from PDGFRα-CreERT2:RiboTag mice on HFD compared with the full murine genome enriched for multiple pathways involved in cell migration (Figure 3E). Pathway analysis of DEGs in HFD HA+ OPCs demonstrated enrichment for downstream chemokine signaling with 31 of 198 chemokine signaling pathway genes differentially expressed in HFD HA+ OPCs (FDR = 1.62 × 10−55)37 (Table 1). ## IL-17Rb and CXCL5 up-regulation in injured white matter vasculature IL-17 signaling involves five IL ligands (A–E) and five cognate receptor isoforms that hetero- and/or homo-dimerize to effect downstream signaling.38 Within our transcriptional dataset, the only IL-17 receptor isoform that was significantly differentially regulated in DIO-affected cerebral endothelial cells was IL-17Rb (Table S4). Among a number of diverse functions, IL-17 receptor activation drives effector chemokine signaling, including CXCL539 as a mechanism of identifying tissue injury. CXCL5 is a member of the C-X-C chemokine family40 that acts as a chemoattractant in other tissues and has been reportedly up-regulated in white matter after peri-natal hypoxia.41 Guided by our RNA-seq data, we hypothesized that DIO may induce IL-17B signaling acting through IL-17Rb resulting in increased endothelial expression of CXCL5, resulting in its secretion both into the bloodstream and into surrounding brain tissue to exert a localized paracrine effect on OPCs (Figure 4A). First, to confirm DIO-induced up-regulation of IL-17Rb/CXCL5 in white matter endothelia observed by TRAP-seq, we performed TRAP-qPCR using independent Tie2-Cre:RiboTag biologic replicates for a subset of differentially regulated genes (Glut-1, Itgb3, Cd180, Hsd3b3, Tnfrsf10b, Il17rb, Cxcl5, and Ttc21a) (Figure 4B). Using TRAP-qPCR, we confirmed the effect of DIO on white matter endothelia with similar degrees of up-regulation for Il17rb and Cxcl5 (3.94 ± 0.07-fold expression, $$p \leq 0.0009$$, and 4.32 ± 0.01-fold expression, $$p \leq 0.0009$$, respectively). Retro-orbital venous blood sampling confirmed increased serum detection of CXCL5 in DIO mice (4,609 ± 407 versus 10,306 ± 1,660 pg/mL, $$p \leq 0.036$$; Figure 4C). Immunofluorescent labeling for IL-17Rb (Figure 4D) and CXCL5 (Figure 4E) in Tie2-Cre;tdTomato (Ai14) mice demonstrated a marked increase in detection of both molecules within white matter cerebral vessels in DIO mice. In peri-infarct tissue 7 days after subcortical white matter stroke, endothelial CXCL5 expression is significantly increased in mice on HFD versus those on CFD as measured by the percentage of CXCL5+ voxels that co-localized with GLUT-1 within the peri-infarct tissue surrounding the stroke (Figure 4F). As in uninjured white matter, the percentage of CXCL5+/GLUT-1+ voxels was significantly increased within the periinfarct tissue in animals on HFD (3.18 ± 0.29 versus 18.19 ± 1.06; $p \leq 0.0001$). Importantly, OPCs were seen in close apposition to CXCL5+ vessel segments in DIO mice, suggesting that this IL-chemokine cascade may drive vascular-OPC signaling and regulate OPC migration (Figure 4F). ## The IL-17/CXCL5 pathway as a vessel-to-OPC signaling paradigm To confirm that IL-17 signaling can drive brain endothelial CXCL5 secretion as suggested by our transcriptional data and working model, we stimulated primary human brain microvascular endothelial cells with recombinant isoforms of IL-17 (A–E). IL-17B, -D, and -E (250 ng/mL) were noted to drive 2-fold increases in the secretion of CXCL5 into conditioned medium ($$p \leq 0.0372$$; Figure 5A). In vitro exposure of O4+ OPCs to increasing doses of recombinant murine CXCL5 resulted in a dose-dependent increase in OPC cell area with cytoskeletal changes suggesting motility ($p \leq 0.0001$, $F = 9.82$ by one-way ANOVA; Figure 5B). To determine the ability of endothelial CXCL5 to signal to OPCs in vivo, we used a combined transgenic and targeted viral gene expression approach (Figure 5C). We designed a pCDH-FLEX-CXCL5-T2A-GFP lentiviral construct to target CXCL5 overexpression to white matter endothelial cells in Tie2-Cre;tdTomato mice. Injection of lentiviral particles expressing either pCDH-FLEX-CXCL5-T2A-GFP or control pCDH-FLEX-GFP into the subcortical white matter of Tie2-Cre;tdTomato mice results in targeted gene expression specifically in white matter vasculature (Figure S5). After 6 weeks of endothelial upregulation of CXCL5-GFP or GFP in normal-weight mice, we measured the distance of individual OPCs from vessels and the cell area of vessel-associated OPCs (Figure 5C). The average distance of OPCs from tdT+ vessels was reduced in CXCL5-GFP-injected animals compared with GFP-injected animals, while the number of PDGFRα+ OPCs in apposition to tdT+ vessels was increased (top panels in Figures 5C, 5E, and 5F), supporting a chemoattractant role for CXCL5 on OPCs. Consistent with the effects of recombinant CXCL5 on OPCs in vitro, endothelial over-expression of CXCL5 in vivo resulted in increased OPC cell area (bottom panels in Figures 5C and 5G). There was no difference in vessel length induced by CXCL5 overexpression in normal-weight mice (0.26 ± 0.05 mm [GFP] versus 0.25 ± 0.06 mm [CXCL5-GFP]; $$p \leq 0.89$$). To block DIO-induced endothelial CXCL5 expression resulting from IL-17Rb activation, we employed repetitive peripheral injections of a function-blocking anti-IL-17B antibody or isotype control immunoglobulin G (IgG) for 6 weeks in Tie2-Cre;tdTomato mice on HFD (Figure 5D). Endothelial CXCL5 expression within the tdT+ vasculature of subcortical white matter was reduced by $60.4\%$ using this approach ($$p \leq 0.018$$, $$n = 4$$/group [grp]; Figure 5H), while IL-17Rb levels were not changed (Figure S5), indicating that DIO-induced increases in endothelial CXCL5 can be at least partially regulated through IL-17B signaling at the endothelial cell surface. Peripheral blocking of IL-17B signaling significantly reduced both the frequency of vessel-associated OPCs as well as the mean vessel-OPC distance in DIO mice (top panels in Figures 5D–5F), while the cell surface area of vessel-associated OPCs was not significantly different in HFD mice administered anti-IL-17B antibody (bottom panels in Figures 5D and 5G). Notably, anti-IL-17B IgG treatment did not significantly alter white matter vessel length in mice on HFD (0.22 ± 0.04 mm [control IgG] versus 0.21 ± 0.04 [anti-IL-17B IgG]; $$p \leq 0.50$$). ## IL-17B and CXCL5 levels in human subjects at risk for cerebrovascular disease With a working model suggesting that DIO drives white matter endothelial CXCL5 expression through IL-17B/IL-17Rb signaling, we sought to establish the relevance of this signaling cascade to human cerebral small vessel disease and vascular cognitive impairment. Using available plasma samples from a single-center cohort study including subjects presenting with acute neurologic symptoms suggestive of stroke,42,43 we assayed plasma levels of IL-17B and CXCL5 using a custom Luminex assay. In those subjects with concurrent blood samples and MRI scans ($$n = 131$$), subjects with detectable levels of IL-17B ($$n = 32$$, mean IL-17B = 47.83 pg/mL) had higher median CXCL5 levels (1,043.0 pg/mL) than in those without detectable IL-17B ($$n = 99$$, 515.3 pg/mL; $p \leq 0.0001$) (Figure 6A). In subjects with tissue-confirmed acute microvessel ischemic lesions, CXCL5 values were higher in those subjects with detectable IL-17B compared with those without measurable IL-17B levels ($$p \leq 0.0157$$) (Figure 6B). In this cohort, the burden of pre-existing cerebral small vessel disease indicated by modified Fazekas scale scoring of white matter hyperintensities is significantly different in IL-17B+ subjects compared with IL-17B− subjects ($p \leq 0.0001$). To confirm CXCL5 expression by white matter endothelia, we examined CXCL5 expression in peri-ventricular white matter from a small post-mortem convenience cohort ($$n = 10$$) of older individuals (86 ± 8 years of age) with measurable amounts of cerebrovascular pathology (Table S5; Figure 6C). The mean percentage of CXCL5+ vessel segments per subject was $71.2\%$ ± $0.08\%$ (17.2 ± 3.4 vessel segments/subject; $$p \leq 0.0005$$) (Figure 6C). Using a separate cohort of 150 subjects with baseline serum sampling and longitudinal cognitive assessment, a mixed-effects regression model adjusted for age, sex, education, and pre-morbid cognitive diagnosis indicates that elevated serum CXCL5 values are significantly associated with level of decline in mean executive function over time (β estimate = 4.61 × 10−5, $$p \leq 0.026$$) (Table S6). ## DISCUSSION Cerebral small vessel disease is increasingly recognized as a substantial contributor to stroke risk and dementia.6 Microvascular injury in the brain is driven by cardiovascular risk factors, yet molecular factors that link systemic vascular risk factors with molecular pathways in the brain are lacking. Here, we use a mouse model of DIO to identify a multicellular inflammatory signaling cascade active in injured white matter before and after ischemic stroke that can also function as a diagnostic and prognostic biomarker for cerebral small vessel disease. Modeling of chronic cerebrovascular risk and pathology using a combined DIO and subcortical white matter stroke model demonstrate that OPCs respond to DIO by vascular association and that their differentiation post-stroke is restricted. Using TRAP-seq in endothelial and OPC transgenic mice, we identify a vascular-OPC signaling cascade acting predominantly through IL-17B-IL-17Rb interaction at the vascular surface to drive endothelial expression of the C-X-C family chemokine CXCL5, promoting OPC chemoattraction to the vasculature. With a combination of in vitro and in vivo studies, we show both that IL-17B regulates endothelial expression of CXCL5 and that OPCs respond to endothelial CXCL5 expression by associating to the vasculature, likely through CXCR-mediated activation of cellular migration. Finally, we extend these findings to the human condition by demonstrating that CXCL5 is present in aged cerebral small vessels and that circulating levels of CXCL5 can identify subjects with imaging or cognitive manifestations of cerebral small vessel disease. Despite advances in single-cell RNA-seq, cell-specific transcriptional profiling using ribosomal tagging remains a valuable tool in parsing out molecular signals from a complex tissue such as the brain.35,44 Here, we utilized EndoRiboTag mice45 in the context of a chronic vascular risk factor model to identify endothelial pathways that appear relevant to human cerebral small vessel disease. A similar vascular profiling approach could be easily applied to identify microvascular injury signals in other organs such as the kidney or retina or conditions that feature microvascular injury including aging, diabetes, or isolated hypertension. Our attempt to translate this vascular profiling approach from mouse to human as a platform for biomarker discovery may represent a unique opportunity to better understand the relationship between cerebrovascular risk factors and human cerebral small vessel disease. Here, we chose to model obesity as it is a leading cardiovascular and cerebrovascular risk factor, is growing in prevalence,46 is associated with white matter changes in humans,11,13–15 and has a reliable animal model.28 Our findings of reductions of white matter vasculature and increased OPCs in DIO mice are similar to those reported in other models of chronic white matter injury.26 While these results reporting vessel-associated OPC morphology in the context of DIO are somewhat discrepant with the heterogeneity of OPC morphologies reported by others,21 this work focused exclusively on subcortical white matter and peri-infarct OPCs, which may have less variation in OPC morphology to begin with. Our results showing ultrastructural changes in myelin in adult-onset DIO are similar to those seen in genetically obese (ob/ob) mice with reductions in myelin47 and increases in OPCs in leptin-deficient ob/ob mice,48 validating this model for the study of chronic white matter injury. OPCs are known to respond early and robustly to white matter ischemic lesions common to the aging human brain.31,49,50 The peri-infarct white matter at the margin of the ischemic lesion, often referred to as the white matter penumbral region,51 is where reparative remyelination can be activated.31 In DIO mice, the stroke-responsive OPC lesion area is $30\%$ larger, and this expanded penumbral region is marked by increased endothelial CXCL5 expression, potentially explaining why more stroke-responsive OPCs are seen at the lesion periphery. Though we did not demonstrate it here, therapeutic targeting of the vasculature in order to regulate remyelination after stroke is an attractive strategy for brain repair. Vessels and OPCs are known to interact both during development and to maintain white matter homeostasis.52 During CNS development, OPCs migrate extensively to distribute throughout the entire CNS, and this migration requires the physical vascular scaffold.23 Cerebral endothelial cells secrete trophic factors that activate Src and Akt signaling pathways to support the survival and proliferation of OPCs.18 However, the full spectrum of molecular pathways that drive the vessel-OPC interaction remain largely unknown. The present data in disease and studies in the developing brain indicate that chemokines are critical. In vivo time-lapse imaging reveals that in the developing mouse brain, OPCs interact with vasculature and migrate along the vessels to the destined cerebral regions dependent on CXCR4 activation in OPCs, which binds to endothelial secreted ligand CXCL12, and promotes their attraction to cerebral vasculature.53 Our study illustrates a similar phenomenon, with DIO-induced endothelial expression of CXCL5 promoting the association of OPCs to the vasculature within adult white matter in vivo. Transcriptional profiling of OPCs in DIO using PDGFRα RiboTAG mice further imply that chemokine signaling pathways play a significant role in regulating a migratory interaction between endothelial cells (ECs) and OPCs. Based on the Gene Ontology analysis* from DIO OPCs, this interaction may promote white matter angiogenesis in the chronic state. Though much is known about the IL-17 superfamily, comparatively little is known about IL-17B and IL-17Rb signaling.38 Using both gain- and loss-of-function studies in vitro and in vivo, we clearly demonstrate that IL-17B can act on brain endothelia to produce CXCL5. The precise source of IL-17B is unclear, though DIO is known to promote Th17 T cells that may function as a primary source of this cytokine.54 Beyond its potential paracrine action on OPCs in the white matter, CXCL5 is secreted by ECs. As such, we hypothesized that circulating CXCL5 could also function as a disease biomarker. In two small, but independent, cohort studies, we show that circulating CXCL5 can function as a cross-sectional diagnostic biomarker for white matter injury on MRI and, in a longitudinal cohort, may associate with future cognitive impairment. An emerging concept places the cerebral EC at the center of the pathophysiology relevant to cerebral small vessel disease.55 Because they act as the conduit between the brain and systemic insults such as hypertension, diabetes, and the metabolic disturbances of obesity, the cerebral endothelia represent an attractive target for understanding disease pathogenesis. From the data presented here, intercellular inflammatory signaling involving the IL-chemokine pathway may be central to white matter injury and post-ischemic myelin repair. Molecular pathways triggered by chronic cerebrovascular risk factors can directly alter injury response and repair after stroke by acting through vascular regulation of myelination. ## Limitations of the study Despite our translational results from mouse to human, this study has important limitations. Regional transcriptional profiling from endothelia only in the subcortical white matter limits the ability to generalize this oligovascular signaling response to other brain regions. Additionally, while we demonstrate the ability of IL-17B to signal through the IL-17Rb receptor to trigger CXCL5 expression in murine and human microvascular ECs, we did not identify a source for circulating IL-17B. If identified, this could drive a therapeutic strategy for white matter repair by targeting the source of IL-17B in obesity. Finally, both cohorts of human subjects are relatively small, and though significant, the magnitude of the effect on diagnosis or prognosis is small. Future studies can expand on these findings using combined IL-17B and CXCL5 measurements in larger, coordinated cohorts enriched for subjects at risk for vascular cognitive impairment. ## Lead contact Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Jason D. Hinman (jhinman@mednet.ucla.edu). ## Materials availability Plasmids generated in this study are deposited in Addgene. Mouse lines generated in this study are available to share upon contact with the lead contact. Anti-IL-17B antibody used in Luminex assay may be available from the manufacturer upon request (Biotechne). ## Data and code availability RNA-seq data have been deposited at GEO: GSE217356 and are publicly available as of the date of publication. Accession numbers are listed in the key resources table. Microscopy data reported in this paper will be shared by the lead contact upon request. All original code has been deposited at https://doi.org/10.17605/OSF.IO/2YMB4 and is publicly available as of the date of publication. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. ## Animals All animal studies presented here were approved by the UCLA Animal Research Committee ARC#2014-067-01B, accredited by the AAALAC. Mice were housed under UCLA regulation with a 12-hour dark-light cycle. All mice used in the study were male. Wild-type C57Bl/6 mice fed ad lib on $60\%$kCal from fat chow (Research Diets, Inc.) (HFD) (Strain #380050) or $10\%$kCal from fat chow (Research Diets, Inc.) (CFD) (Strain #380056) were purchased directly from Jackson Labs at 17 weeks of age and allowed to acclimate for 2 weeks prior to experimental use. Weights (g) were measured weekly. The PDGFRα- CreERT2/Rpl22-HA and Tie2-Cre/Rpl22-HA transgenic strain were generated by crossing PDGFRα- CreERT2 mice (Jackson Labs Strain #018280 - B6N.Cg-Tg(Pdgfra-cre/ERT)467Dbe/J) and Tie2-Cre (Jackson Labs Strain #008863-B6.Cg-Tg(Tek-cre)1Ywa/J) with Rpl22-fl-Rpl22-HA (Jackson Labs Strain #011029 - B6N.129-Rpl22tm1.1Psam/J). The Tie2-Cre;tdTomato mice were generated by crossing Tie2-Cre mice with flox-stop tdTomato mice (Jackson Labs Strain #007908 – B6;129S6-Gt(ROSA)26Sortm14(CAG-tdTomato)Hze/J). Diet-induced obesity was induced in transgenic mice by ad lib feeding with $60\%$kCal from fat chow (HFD) or $10\%$kCal from fat chow (CFD) (Research Diets, Inc.). Genotyping was performed by transgene specific qPCR (Transnetyx). For OPC RNA-sequencing, tamoxifen (Sigma) was dissolved in corn oil and injected i.p. ( 50mg/kg) once to PDGFRα-CreERT2/Rpl22-HA ($$n = 6$$) mice and animals were euthanized and tissue collected 48 hrs later for RiboTag pulldown as described. All animal studies presented here were approved by the UCLA Animal Research Committee, accredited by the AAALAC. Mice were housed under UCLA regulation with a 12 hour dark-light cycle. All mice used in the study were male. Wild-type C57Bl/6 mice fed ad lib on $60\%$kCal from fat chow (HFD) (Strain #380050) or $10\%$kCal from fat chow (CFD) (Strain #380056) were purchased directly from Jackson Labs at 17 weeks of age and allowed to acclimate for 2 weeks prior to experimental use. The PDGFRα- CreERT2/Rpl22-HA and Tie2-Cre/Rpl22-HA transgenic strain were generated by crossing PDGFRα- CreERT2 mice (Jackson Labs Strain #018280 - B6N.Cg-Tg(Pdgfra-cre/ERT)467Dbe/J) and Tie2-Cre (Jackson Labs Strain #008863-B6.Cg-Tg(Tek-cre)1Ywa/J) with Rpl22-flRpl22-HA (Jackson Labs Strain #011029 - B6N.129-Rpl22tm1.1Psam/J). The Tie2-Cre;tdTomato mice were generated by crossing Tie2-Cre mice with flox-stop tdTomato mice (Jackson Labs Strain #007908 – B6;129S6-Gt(ROSA)26Sortm14(CAG-tdTomato)Hze/J). Diet-induced obesity was induced in transgenic mice by ad lib feeding with $60\%$kCal from fat chow (HFD) or $10\%$kCal from fat chow (CFD) (Research Diets, Inc.). Weights (g) were measured weekly. For OPC RNA-sequencing, tamoxifen (Sigma) was dissolved in corn oil and injected i.p. ( 50mg/kg) once to PDGFRα-CreERT2/Rpl22-HA ($$n = 6$$) mice and animals were euthanized and tissue collected 48 hrs later for RiboTag pulldown as described. ## ASPIRE study cohort Patients presenting for emergency evaluation of stroke or cerebrovascular disease were recruited and provided blood samples and neuroimaging data as approved by the UCLA Institutional Review Board (IRB # 14-001798) as previously reported.42 Serum levels of IL-17B and CXCL5 were measured in technical duplicate using a custom Luminex assay (R&D Systems). Manufacturer protocol was followed and antigen binding within the assay was measured on a Luminex 200 System and analyzed using Milliplex Analyst 5.1. Modified Fazekas scores were determined by blinded analysis of T2-weighted FLAIR images by two independent reviewers. ASPIRE study data are available at https://osf.io/92erq/. ## Post-mortem cohort Subjects were selected from a subset of 950 UCDavis ADC Neuropathology Core samples based on a priori selection criteria. All subjects consented to autopsy. A convenience cohort of ten elderly individuals with variable amounts of cerebrovascular disease and low Braak and Braak scores were selected for analysis. Age and sex information is provided in Table S2. ## Longitudinal cohort UCD ADRC Longitudinal Diversity Cohort consists of demographically diverse individuals recruited through both clinical and community sources.56 Formal written consent was obtained for all participants prior to the collection of data. For this study, this highly demographically diverse cohort consists of $58\%$ non-Hispanic Caucasians (Whites), $19\%$ African Americans (Blacks) and $13\%$ Hispanics, $52\%$ female, average age 78 + 7.3 years with average educational attainment of 14.7 + 4.0 years ranging from 0–20 years and various medical comorbidities common to the general population. Longitudinal cognitive testing utilized the Spanish English Neuropsychological Assessment Scale.57,58 Participants for this study were assessed 6.3 + 3.6 times ranging from 1–17 times. Serum levels of CXCL5 were measured in technical duplicate using a custom Luminex assay as above. ## Human brain microvascular endothelial cell culture Primary Human Brain Microvascular Endothelial Cells (HBMECs) (Cell Systems) between P5-P9 were maintained at 37°C until confluence with manufacturer recommended media containing serum with media exchange every two days. Maintenance cultures were replated into a 96-well filter bottom plate and cultured until near confluence. Cultures were mixed sex and not authenticated. ## White matter stroke Subcortical white matter ischemic injury was induced as previously described33 using three stereotactic injections of the irreversible eNOS inhibitor, L-Nio (L-N⁵-(1-Iminoethyl) ornithine, dihydrochloride; Calbiochem) into the subcortical white matter under sensorimotor cortex. Animals ($$n = 8$$/grp) were sacrificed at 7- or 28-days post-stroke and analyzed for tissue outcomes. ## Translating ribosome affinity purification and RNA-sequencing HA-tagged ribosomal associated RNAs from cerebral white matter endothelia or OPCs were isolated following published protocol.35 Post-immunoprecipitation RNA samples were purified by Nucleospin miRNA kit (Machary-Nagel). Normalized RNA amounts (ng) underwent cDNA library generation using the TrueSeq with Ribozero kit preparation (Illumina), pooled and sequenced using 69 bp paired end reads on a Illumina HiSeq 4000 sequencer. Samples were sequenced over 4 lanes for an average of read count of 62.1± 10.7 million per sample (Tie2-Cre:RiboTag) and 75.9 ± 11.1 million per sample (PDGFRα- CreERT2:RiboTag). Reads were aligned to the mouse genome using STAR (v.mm10). Differential gene expression analysis was performed using EdgeR assuming an FDR <0.1 as significant. Gene ontology analysis was performed using GOrilla59 and Enrichr.60 Chemokine pathway analysis was performed using the KEGG pathway resource61 and verified using the STRING database resource.37 Selected genes were verified by qPCR using independent TRAP isolates. ## RNA hybridization assay Wild-type C57Bl/6 mice ($$n = 4$$/grp) were placed on CFD or HFD starting at 8 weeks of age and after 12 weeks on CFD or HFD were sacrificed. The subcortical white matter was freshly dissected. RNA was isolated using the Nucleospin miRNA kit (Machary-Nagel). RNA samples were allowed to directly hybridize with a custom RNA probe set for 120 oligodendrocyte/myelin gene set derived from Zhang et al.36 with 40 genes each corresponding to the major oligodendrocyte stages including oligodendrocyte progenitor cells (OPC), pre-myelinating oligodendrocytes (PMO), and (myelinating oligodendrocytes (MO). Hybridized mRNA species were detected using the nCounter detection system (Nanostring) and normalized to five housekeeping genes (Supplemental Data File 1). Normalized counts for each probe set were divided into three major oligodendrocyte subtypes (OPC, PMO, and MO) and compared by differential gene expression analysis. Additional comparisons were performed using normalized read counts from Zhang et al. using a tSNE data reduction analysis. ## IL-17 treatment and CXCL5 measurement Two days after seeding, HBMECs were stimulated with culture medium containing 250 ng/mL of mouse IL-17A, B, C, D, or E (R&D Systems, Inc.). Conditioned media from triplicate culture conditions was collected after 48 hours and human CXCL5 levels measured using a human CXCL5 Quantikine Elisa Kit (R&D Systems, Inc.). Absorbance values measured at 450 nm and absorbance at 570 nm was used for background subtraction. Background subtracted absorbance values were converted to pg/mL concentrations based on standard curve concentrations. ## Microscopy and imaging Animals were euthanized with a lethal dose of isoflurane, transcardially perfused with PBS followed by $4\%$ paraformaldehyde in 0.1 M sodium phosphate buffer, brains removed, post-fixed for 24 hrs and cryoprotected for 48 hrs in $30\%$ sucrose in PBS. Forty micron coronal cryosections and immunostaining were performed essentially as described.30 The following primary antibodies were used: mouse anti-NF200 (1:200, Sigma), rabbit anti-MBP (1:500, Calbiochem), goat anti-PDGFRα (1:500; Neuromics), mouse anti-HA (1:1000, Biolegend), rabbit-Gst-π (1:1000, Millipore), rabbit anti-IL-17Rb (1:500, Santa Cruz Biotech), rat anti-CXCL5 (1:250, R&D) in PBS containing $5\%$ goat or donkey serum and $0.3\%$ Triton-X 100 (Sigma) overnight at 4°C. Secondary antibody labeling was performed using donkey anti-mouse, donkey anti-rabbit, donkey anti-rat or donkey anti-goat Fab2-Alexa conjugated antibodies (Jackson Immunoresearch, Inc.). All microscopic images were obtained using a Nikon C2 confocal microscope. ## Electron microscopy Wild-type C57Bl/6 mice ($$n = 6$$/grp) on CFD or HFD were transcardially perfused with a $2\%$ glutaraldehyde solution, post-fixed for 24 hrs, hemisected in the sagittal plane and 2 mm cubes including the corpus callosum were dissected and embedded in plastic resin for ultrastructural analysis as previously described.31 One-micron, plastic embedded toluidine blue stained sections were used to select transcallosal fibers underneath sensorimotor cortex by light microscopy. Three electron micrographs were obtained at a primary magnification of 7200X using a JEOL 100 CX transmission electron microscope and a representative electron micrograph of high technical quality from each animal was used for quantitation of fiber diameter, axon diameter, myelin thickness, and g-ratio. ## Lentiviral injection A plasmid containing the open reading frame of the murine CXCL5 sequence with a 3′ stop codon was purchased from Origene (#MR200761). The pCDH-EF1-FLEX-EGFP-CMV-2A-TagBFP2-SC dual promoter lentiviral backbone was created by subcloning the FLEX-GFP sequence between the loxP sites from the pAAV-FLEX-GFP vector (Addgene #28304) into the pCDH-EF1-MCS-CMV-2A-pTagBFP2-SC dual promoter lentiviral construct using restriction digestion. The pCDH-EF1-FLEX-EGFP-CMV-2A-TagBFP2-SC backbone was linearized by removing the GFP sequence between the loxP sites using restriction digestion with XhoI and EcoRI (New England Biolabs). The murine CXCL5 sequence was PCR amplified in a HiFi DNA Assembly reaction (New England Biolabs) such that it was subcloned in the 3’>5′ position in between the loxP sites. The resulting reaction was transformed into Stbl3 E.coli cells and positive clones were identified by restriction digestion and verified by DNA sequencing. Subsequently, a 3’>5′ T2A-copGFP sequence was added 5′ to the murine CXCL5 sequence. The donor T2A-copGFP vector (pCDH-EF1-MCS-copGFP; System Biosciences) was PCR amplified and subcloned into pCR-Blunt II TOPO (ThermoFisher Scientific) for amplification and utilized in a HiFi DNA Assembly reaction. The resulting reaction was transformed as above and positive clones were identified by restriction digestion and DNA sequencing. DNA amplification was performed using an Endotoxin-Free PureLink Plasmid Midiprep Kit (ThermoFisher Scientific). Resulting DNA was quantified and used in lentiviral packaging. Control GFP and CXCL5-GFP lentivirus were packaged in human 293 cells (ATCC cat. no. CRL-11268) and concentrated by ultracentrifugation on a sucrose column. 200 nL of concentrated virus was injected into the subcortical white matter and allowed to express for 6 weeks. ## Anti-IL-17B antibody administration Anti-mIL-17B function blocking antibody (R&D, AF1709) was diluted with $0.9\%$ saline to a concentration of 1 mg/mL. Normal Goat isotype-matched IgG (R&D, AB-108-C) was used as control. Tie2-Cre;tdTomato mice were fed with high fat diet starting at 8 weeks old and weighed weekly. Aliquots of 50μg of anti-mIL-17B IgG or control IgG were prepared and administered in a blinded fashion every 72 hours by intraperitoneal injection from 14 weeks old and analyzed 48 hours after the last injection at 20 weeks old. ## Microscopic analysis To measure microvascular complexity, Tie2-Cre;tdTomato expressing vessels were used for the analyses of vessel volume, vessel length, and junction point. Vessel volume was measured by Imaris software with automated “Add surface” function. Volume of small particle less than 30μm3 was subtracted to eliminate the background interference. The masked volume that created by Imaris was identified as vessel volume. Vessel length and junction point were analyzed by AngioTool. The parameters for AngioTool measurement were set as “Diameter 5–40”, “Intensity 40–255” and “Particles less than 10000”. Analysis of the spatial distribution of stroke-responsive OPCs was performed as follows. The boundary of increased PDGFR-α-+ cells and the loss of GST-π-+ cells was identified in each of three sections per animal ($$n = 3$$ animals/group). Using Imaris software, the x,y,z position of each PDGFR-α-+ cell relative to the user defined center point ($x = 0$, $y = 0$, $z = 0$) of the elliptical stroke region was determined using the automated “Add Spots” function. Individual cell areas were generated by Imaris with “Add Surface” function. Because the z-axis was limited (10 μm), a two-dimensional grid analysis was performed using a 2D modification of the previously reported 3D spatial density estimator using a smoothing parameter of $k = 8.$ The local cell density in each position within the overlaid grid is compared statistically as previously described. Therefore, a p-value map is generated for each position in the grid and thresholded ($p \leq 0.05$) to reveal regions with significant density differences. The size of PDGFRα+ OPC was measured individually by Imaris with automated “Add surface” function. Voxel of small particle less than 800 was subtracted to eliminate the background interference. The masked area of PDGFRα+ OPC that created by Imaris was identified as the size of OPC. OPC-vessel distance was measured by Imaris with “Add spot” function. For PDGFRα+ OPC location, nucleus with Dapi staining was used as a reference. The distance of Tie2-Cre;tdTomato vessel to PDGFRα+ OPC was measured with the function of “Spot to Spot closest distance”. The levels of CXCL5/IL-17Rb in IgG/IL-17B treated mice white matter were measured by Imaris “Coloc” function. The percentages of CXCL5/IL-17Rb that colocalized with Tie2cre;tdTomato positive vessels were measured as voxel areas. For GLUT-1/CXCL5 colocalization measurement, GLUT-1 positive vessels were masked by Imaris with “Add surface” to create new GLUT-1 and CXCL5 channels. The percentages of GLUT-1/CXCL5 colocalization in new channels were measured as voxel areas by Imaris “Coloc” function. ## Immunohistochemistry for human brain samples Case selection was made from a subset of 950 UC Davis ADC Neuropathology Core samples based on a priori selection criteria: low Braak and Braak scores, at least $80\%$ with some pathologic evidence of cerebrovascular disease sufficient to cause dementia. The most recent cases available were selected based on the selection criteria. Immunohistochemistry was performed using formalin (Medical Chemical Corporation, 575A) fixed paraffin embedded tissue sections cut at 6μm. Sections where placed on positive charged slides (Fisherbrand, 12-550-15) then incubated overnight at 60°C. De-paraffinization was accomplished with three 5min xylene (Fisher Scientific, X3P) washes. The samples were rehydrated with graded concentrations of alcohol (American MasterTech, ALREACS) diluted with deionized water. Endogenous peroxidase was blocked with a $3\%$ solution of hydrogen peroxide (Fisher Scientific, H325–500) 20min incubation. Heat-induced epitope retrieval used a citrate buffer (BioCare Medical, CB910M). The slides incubated in the buffer at 90°C for 45min. Blocking used $2.5\%$ normal horse serum (Vector, S2012) for 60min. Antigen specificity was elucidated by incubating the slides for 90min in CXCL$\frac{5}{6}$ (1:100, Abcam, ab198505). Primary antibody detection was amplified with a 45min incubation using a secondary antibody (Vector, MP-7401). A 5 second counterstain used hematoxylin (Richard Allan Scientific, 7221). The samples were dehydrated with graded alcohols and three xylene washes before being coverslipped. ## QUANTIFICATION AND STATISTICAL ANALYSIS The number of animals used in each experiment is listed in the Results section. Vessel densities and oligodendrocyte population cell counts as a fraction of total cells were determined by averaging counts from 5 fields of view (FOV) throughout the corpus callosum across a minimum of three sections 240 μm apart. Per animal averages were generated and significance between groups determined using an unpaired Welch’s t-test (α = 0.05). Measurements of white matter ultrastructural features were determined using 6 FOVs and averaged across animals and compared at the feature level separately using Mann-Whitney U test between groups (α = 0.05). Determination of stroke lesion area was performed by sampling lesion area ($$n = 3$$–5 40 μm sections) across groups ($$n = 4$$/grp) and using the sampled distribution to create bootstrapped area distribution ($$n = 25$$) representing a full area sampling of the approximate 1 mm lesion created by the stroke model. This area distribution was averaged across animals in each group and compared using a Mann-Whitney U test between groups (α = 0.05). Spatial analysis of stroke-responsive OPCs were determined as above. Cell counts at 28d post-stroke were determined across three sections 240 μm apart with lesion core and edge analyses determined using a two-way ANOVA (α = 0.05) with post-hoc Holm-Sidak test to correct for multiple comparisons. Post-stroke myelination was determined using a Chi-square comparison of distributions. Gene expression differences were determined at the individual gene level using unpaired Welch’s t-test (α = 0.05). CXCL5 values in conditioned media were analyzed using a Krusal-Wallis test with false discovery rate correction. Human serum CXCL5 levels were log10 transformed and compared by Mann-Whitney U test. Fazekas scale scores were compared using an ordinal shift Chi-square analysis. Human CXCL5+ vessel segments were compared by two-tailed, one-sample t-test assuming no expression of CXCL5 in non-injured tissue. Unless otherwise stated, all other comparisons were determined using a one-way ANOVA with post-hoc Holm-Sidak test to correct for multiple comparisons. Statistical analysis was performed using GraphPad Prism 7 software. Data are shown as mean ± SEM. ## References 1. Gouw AA, van der Flier WM, Fazekas F, van Straaten ECW, Pantoni L, Poggesi A, Inzitari D, Erkinjuntti T, Wahlund LO, Waldemar G. Progression of white matter hyperintensities and incidence of new lacunes over a 3-year period: the Leukoaraiosis and Disability study. Stroke (2008) 39 1414-1420. DOI: 10.1161/STROKEAHA.107.498535 2. Inzitari D, Pracucci G, Poggesi A, Carlucci G, Barkhof F, Chabriat H, Erkinjuntti T, Fazekas F, Ferro JM, Hennerici M. Changes in white matter as determinant of global functional decline in older independent outpatients: three year follow-up of LADIS (leukoaraiosis and disability) study cohort. BMJ (2009) 339 b2477. DOI: 10.1136/bmj.b2477 3. Jokinen H, Gouw AA, Madureira S, Ylikoski R, van Straaten ECW, van der Flier WM, Barkhof F, Scheltens P, Fazekas F, Schmidt R. Incident lacunes influence cognitive decline: the LADIS study. Neurology (2011) 76 1872-1878. DOI: 10.1212/WNL.0b013e31821d752f 4. Koga H, Takashima Y, Murakawa R, Uchino A, Yuzuriha T, Yao H. Cognitive consequences of multiple lacunes and leukoaraiosis as vascular cognitive impairment in community-dwelling elderly individuals. J. Stroke Cerebrovasc. Dis (2009) 18 32-37. DOI: 10.1016/j.jstrokecerebrovasdis.2008.07.010 5. Reed BR, Eberling JL, Mungas D, Weiner M, Kramer JH, Jagust WJ. Effects of white matter lesions and lacunes on cortical function. Arch. Neurol (2004) 61 1545-1550. DOI: 10.1001/archneur.61.10.1545 6. Debette S, Markus HS. The clinical importance of white matter hyperintensities on brain magnetic resonance imaging: systematic review and meta-analysis. BMJ (2010) 341 c3666. DOI: 10.1136/bmj.c3666 7. van Dijk EJ, Breteler MMB, Schmidt R, Berger K, Nilsson LG, Oudkerk M, Pajak A, Sans S, de Ridder M, Dufouil C. The association between blood pressure, hypertension, and cerebral white matter lesions: cardiovascular determinants of dementia study. Hypertension (2004) 44 625-630. DOI: 10.1161/01.HYP.0000145857.98904.20 8. de Leeuw FE, de Groot JC, Oudkerk M, Witteman JCM, Hofman A, van Gijn J, Breteler MMB. Hypertension and cerebral white matter lesions in a prospective cohort study. Brain (2002) 125 765-772. PMID: 11912110 9. Jongen C, van der Grond J, Kappelle LJ, Biessels GJ, Viergever MA, Pluim JPW. Automated measurement of brain and white matter lesion volume in type 2 diabetes mellitus. Diabetologia (2007) 50 1509-1516. DOI: 10.1007/s00125-007-0688-y 10. Manschot SM, Brands AMA, van der Grond J, Kessels RPC, Algra A, Kappelle LJ, Biessels GJ. Brain magnetic resonance imaging correlates of impaired cognition in patients with type 2 diabetes. Diabetes (2006) 55 1106-1113. PMID: 16567535 11. Bokura H, Yamaguchi S, Iijima K, Nagai A, Oguro H. Metabolic syndrome is associated with silent ischemic brain lesions. Stroke (2008) 39 1607-1609. DOI: 10.1161/STROKEAHA.107.508630 12. Park K, Yasuda N. Association between metabolic syndrome and minimal leukoaraiosis. Stroke (2009) 40 e5. DOI: 10.1161/STROKEAHA.108.531012 13. Park K, Yasuda N, Toyonaga S, Tsubosaki E, Nakabayashi H, Shimizu K. Significant associations of metabolic syndrome and its components with silent lacunar infarction in middle aged subjects. J. Neurol. Neurosurg. Psychiatry (2008) 79 719-721. DOI: 10.1136/jnnp.2007.134809 14. Yin ZG, Cui M, Zhou SM, Yu MM, Li R, Zhou HD. Association between metabolic syndrome and white matter lesions in middle-aged and elderly patients. Eur. J. Neurol (2014) 21 1032-1039. DOI: 10.1111/ene.12433 15. Kwon HM, Kim BJ, Park JH, Ryu WS, Kim CK, Lee SH, Ko SB, Nam H, Lee SH, Lee YS, Yoon BW. Significant association of metabolic syndrome with silent brain infarction in elderly people. J. Neurol (2009) 256 1825-1831. DOI: 10.1007/s00415-009-5201-8 16. Fisher CM. Capsular infarcts: the underlying vascular lesions. Arch. Neurol (1979) 36 65-73. PMID: 420625 17. Esiri MM, Wilcock GK, Morris JH. Neuropathological assessment of the lesions of significance in vascular dementia. J. Neurol. Neurosurg. Psychiatry (1997) 63 749-753. PMID: 9416809 18. Arai K, Lo EH. An oligovascular niche: cerebral endothelial cells promote the survival and proliferation of oligodendrocyte precursor cells. J. Neurosci (2009) 29 4351-4355. DOI: 10.1523/JNEUROSCI.0035-09.2009 19. Pham LDD, Hayakawa K, Seo JH, Nguyen MN, Som AT, Lee BJ, Guo S, Kim KW, Lo EH, Arai K. Crosstalk between oligodendrocytes and cerebral endothelium contributes to vascular remodeling after white matter injury. Glia (2012) 60 875-881. DOI: 10.1002/glia.22320 20. Maki T, Maeda M, Uemura M, Lo EK, Terasaki Y, Liang AC, Shindo A, Choi YK, Taguchi A, Matsuyama T. Potential interactions between pericytes and oligodendrocyte precursor cells in perivascular regions of cerebral white matter. Neurosci. Lett (2015) 597 164-169. DOI: 10.1016/j.neulet.2015.04.047 21. Kishida N, Maki T, Takagi Y, Yasuda K, Kinoshita H, Ayaki T, Noro T, Kinoshita Y, Ono Y, Kataoka H. Role of perivascular oligodendrocyte precursor cells in angiogenesis after brain ischemia. J. Am. Heart Assoc (2019) 8 e011824. DOI: 10.1161/JAHA.118.011824 22. Marques S, Zeisel A, Codeluppi S, van Bruggen D, Mendanha Falcão A, Xiao L, Li H, Häring M, Hochgerner H, Romanov RA. Oligodendrocyte heterogeneity in the mouse juvenile and adult central nervous system. Science (2016) 352 1326-1329. DOI: 10.1126/science.aaf6463 23. Tsai HH, Niu J, Munji R, Davalos D, Chang J, Zhang H, Tien AC, Kuo CJ, Chan JR, Daneman R, Fancy SPJ. Oligodendrocyte precursors migrate along vasculature in the developing nervous system. Science (2016) 351 379-384. DOI: 10.1126/science.aad3839 24. Hayakawa K, Pham LDD, Som AT, Lee BJ, Guo S, Lo EH, Arai K. Vascular endothelial growth factor regulates the migration of oligodendrocyte precursor cells. J. Neurosci (2011) 31 10666-10670. DOI: 10.1523/JNEUROSCI.1944-11.2011 25. Hayakawa K, Seo JH, Pham LDD, Miyamoto N, Som AT, Guo S, Kim KW, Lo EH, Arai K. Cerebral endothelial derived vascular endothelial growth factor promotes the migration but not the proliferation of oligodendrocyte precursor cells in vitro. Neurosci. Lett (2012) 513 42-46. DOI: 10.1016/j.neulet.2012.02.004 26. Rajani RM, Quick S, Ruigrok SR, Graham D, Harris SE, Verhaaren BFJ, Fornage M, Seshadri S, Atanur SS, Dominiczak AF. Reversal of endothelial dysfunction reduces white matter vulnerability in cerebral small vessel disease in rats. Sci. Transl. Med (2018) 10 eaam9507. DOI: 10.1126/scitranslmed.aam9507 27. Corriveau RA, Koroshetz WJ, Gladman JT, Jeon S, Babcock D, Bennett DA, Carmichael ST, Dickinson SLJ, Dickson DW, Emr M. Alzheimer’s disease-related dementias summit 2016: national research priorities. Neurology (2017) 89 2381-2391. DOI: 10.1212/WNL.0000000000004717 28. Surwit RS, Kuhn CM, Cochrane C, McCubbin JA, Feinglos MN. Diet-induced type II diabetes in C57BL/6J mice. Diabetes (1988) 37 1163-1167. PMID: 3044882 29. Kennedy AJ, Ellacott KLJ, King VL, Hasty AH. Mouse models of the metabolic syndrome. Dis. Model. Mech (2010) 3 156-166. DOI: 10.1242/dmm.003467 30. Hinman JD, Rasband MN, Carmichael ST. Remodeling of the axon initial segment after focal cortical and white matter stroke. Stroke (2013) 44 182-189. DOI: 10.1161/STROKEAHA.112.668749 31. Sozmen EG, Rosenzweig S, Llorente IL, DiTullio DJ, Machnicki M, Vinters HV, Havton LA, Giger RJ, Hinman JD, Carmichael ST. Nogo receptor blockade overcomes remyelination failure after white matter stroke and stimulates functional recovery in aged mice. Proc. Natl. Acad. Sci. USA (2016) 113 E8453-E8462. DOI: 10.1073/pnas.1615322113 32. Huang PL. A comprehensive definition for metabolic syndrome. Dis. Model. Mech (2009) 2 231-237. DOI: 10.1242/dmm.001180 33. Nunez S, Doroudchi MM, Gleichman AJ, Ng KL, Llorente IL, Sozmen EG, Carmichael ST, Hinman JD. A versatile murine model of subcortical white matter stroke for the study of axonal degeneration and white matter neurobiology. J. Vis. Exp (2016) 53404. DOI: 10.3791/53404 34. Sozmen EG, DiTullio DJ, Rosenzweig S, Hinman JD, Bridges SP, Marin MA, Kawaguchi R, Coppola G, Carmichael ST. White matter stroke induces a unique oligo-astrocyte niche that inhibits recovery. J. Neurosci (2019) 39 9343-9359. DOI: 10.1523/JNEUROSCI.0103-19.2019 35. Sanz E, Yang L, Su T, Morris DR, McKnight GS, Amieux PS. Cell-type-specific isolation of ribosome-associated mRNA from complex tissues. Proc. Natl. Acad. Sci. USA (2009) 106 13939-13944. DOI: 10.1073/pnas.0907143106 36. Zhang Y, Chen K, Sloan SA, Bennett ML, Scholze AR, O’Keeffe S, Phatnani HP, Guarnieri P, Caneda C, Ruderisch N. An RNA-sequencing transcriptome and splicing database of glia, neurons, and vascular cells of the cerebral cortex. J. Neurosci (2014) 34 11929-11947. DOI: 10.1523/JNEUROSCI.1860-14.2014 37. Szklarczyk D, Gable AL, Lyon D, Junge A, Wyder S, Huerta-Cepas J, Simonovic M, Doncheva NT, Morris JH, Bork P. STRING v11: protein-protein association networks with increased coverage, supporting functional discovery in genome-wide experimental datasets. Nucleic Acids Res (2019) 47 D607-D613. DOI: 10.1093/nar/gky1131 38. Gaffen SL. Structure and signalling in the IL-17 receptor family. Nat. Rev. Immunol (2009) 9 556-567. DOI: 10.1038/nri2586 39. Chen K, Eddens T, Trevejo-Nunez G, Way EE, Elsegeiny W, Ricks DM, Garg AV, Erb CJ, Bo M, Wang T. IL-17 receptor signaling in the lung epithelium is required for mucosal chemokine gradients and pulmonary host defense against K. pneumoniae. Cell Host Microbe (2016) 20 596-605. DOI: 10.1016/j.chom.2016.10.003 40. Strieter RM, Polverini PJ, Kunkel SL, Arenberg DA, Burdick MD, Kasper J, Dzuiba J, Van Damme J, Walz A, Marriott D. The functional role of the ELR motif in CXC chemokine-mediated angiogenesis. J. Biol. Chem (1995) 270 27348-27357. PMID: 7592998 41. Wang LY, Tu YF, Lin YC, Huang CC. CXCL5 signaling is a shared pathway of neuroinflammation and blood-brain barrier injury contributing to white matter injury in the immature brain. J. Neuroinflammation (2016) 13 6. DOI: 10.1186/s12974-015-0474-6 42. Altendahl M, Maillard P, Harvey D, Cotter D, Walters S, Wolf A, Singh B, Kakarla V, Azizkhanian I, Sheth SA. An IL-18-centered inflammatory network as a biomarker for cerebral white matter injury. PLoS One (2020) 15 e0227835. DOI: 10.1371/journal.pone.0227835 43. Azizkhanian I, Sheth SA, Iavarone AT, Lee S, Kakarla V, Hinman JD. Plasma lipid profiling identifies biomarkers of cerebral microvascular disease. Front. Neurol (2019) 10 950. DOI: 10.3389/fneur.2019.00950 44. Heiman M, Kulicke R, Fenster RJ, Greengard P, Heintz N. Cell type-specific mRNA purification by translating ribosome affinity purification (TRAP). Nat. Protoc (2014) 9 1282-1291. DOI: 10.1038/nprot.2014.085 45. Cleuren ACA, van der Ent MA, Jiang H, Hunker KL, Yee A, Siemieniak DR, Molema G, Aird WC, Ganesh SK, Ginsburg D. The in vivo endothelial cell translatome is highly heterogeneous across vascular beds. Proc. Natl. Acad. Sci. USA (2019) 116 23618-23624. DOI: 10.1073/pnas.1912409116 46. 46.U.S. Department of Health & Human Services (2018). Nutrition, Physical Activity, and Obesity: Data, Trends and Maps (U.S. Department of Health & Human Services).. Nutrition, Physical Activity, and Obesity: Data, Trends and Maps (2018) 47. Sena A, Sarliève LL, Rebel G. Brain myelin of genetically obese mice. J. Neurol. Sci (1985) 68 233-243. PMID: 2989440 48. Udagawa J, Nimura M, Otani H. Leptin affects oligodendroglial development in the mouse embryonic cerebral cortex. Neuroendocrinol. Lett (2006) 27 177-182. PMID: 16670672 49. Miyamoto N, Maki T, Shindo A, Liang AC, Maeda M, Egawa N, Itoh K, Lo EK, Lok J, Ihara M, Arai K. Astrocytes promote oligodendrogenesis after white matter damage via brain-derived neurotrophic factor. J. Neurosci (2015) 35 14002-14008. DOI: 10.1523/JNEUROSCI.1592-15.2015 50. Sozmen EG, Kolekar A, Havton LA, Carmichael ST. A white matter stroke model in the mouse: axonal damage, progenitor responses and MRI correlates. J. Neurosci. Methods (2009) 180 261-272. DOI: 10.1016/j.jneumeth.2009.03.017 51. Maillard P, Fletcher E, Harvey D, Carmichael O, Reed B, Mungas D, DeCarli C. White matter hyperintensity penumbra. Stroke (2011) 42 1917-1922. DOI: 10.1161/STROKEAHA.110.609768 52. Arai K, Lo EH. Oligovascular signaling in white matter stroke. Biol. Pharm. Bull (2009) 32 1639-1644. PMID: 19801821 53. Banisadr G, Frederick TJ, Freitag C, Ren D, Jung H, Miller SD, Miller RJ. The role of CXCR4 signaling in the migration of transplanted oligodendrocyte progenitors into the cerebral white matter. Neurobiol. Dis (2011) 44 19-27. DOI: 10.1016/j.nbd.2011.05.019 54. Ahmed M, Gaffen SL. IL-17 in obesity and adipogenesis. Cytokine Growth Factor Rev (2010) 21 449-453. DOI: 10.1016/j.cytogfr.2010.10.005 55. Quick S, Moss J, Rajani RM, Williams A. A vessel for change: endothelial dysfunction in cerebral small vessel disease. Trends Neurosci (2021) 44 289-305. DOI: 10.1016/j.tins.2020.11.003 56. Hinton L, Carter K, Reed BR, Beckett L, Lara E, DeCarli C, Mungas D. Recruitment of a community-based cohort for research on diversity and risk of dementia. Alzheimer Dis. Assoc. Disord (2010) 24 234-241. DOI: 10.1097/WAD.0b013e3181c1ee01 57. Mungas D, Reed BR, Marshall SC, González HM. Development of psychometrically matched English and Spanish language neuropsychological tests for older persons. Neuropsychology (2000) 14 209-223. PMID: 10791861 58. Mungas D, Reed BR, Crane PK, Haan MN, González H. Spanish and English Neuropsychological Assessment Scales (SENAS): further development and psychometric characteristics. Psychol. Assess (2004) 16 347-359. PMID: 15584794 59. Eden E, Navon R, Steinfeld I, Lipson D, Yakhini Z. GOrilla: a tool for discovery and visualization of enriched GO terms in ranked gene lists. BMC Bioinf (2009) 10 48. DOI: 10.1186/1471-2105-10-48 60. Gundersen GW, Jones MR, Rouillard AD, Kou Y, Monteiro CD, Feldmann AS, Hu KS, Ma’ayan A. GEO2Enrichr: browser extension and server app to extract gene sets from GEO and analyze them for biological functions. Bioinformatics (2015) 31 3060-3062. DOI: 10.1093/bioinformatics/btv297 61. 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
--- title: Class Ib MHC–Mediated Immune Interactions Play a Critical Role in Maintaining Mucosal Homeostasis in the Mammalian Large Intestine authors: - Suryasarathi Dasgupta - Igor Maricic - Jay Tang - Stephen Wandro - Kelly Weldon - Carolina S. Carpenter - Lars Eckmann - Jesus Rivera-Nieves - William Sandborn - Rob Knight - Peter Dorrestein - Austin D. Swafford - Vipin Kumar journal: ImmunoHorizons year: 2021 pmcid: PMC10026853 doi: 10.4049/immunohorizons.2100090 license: CC BY 4.0 --- # Class Ib MHC–Mediated Immune Interactions Play a Critical Role in Maintaining Mucosal Homeostasis in the Mammalian Large Intestine ## Abstract Lymphocytes within the intestinal epithelial layer (IEL) in mammals have unique composition compared with their counterparts in the lamina propria. Little is known about the role of some of the key colonic IEL subsets, such as TCRαβ+CD8+ T cells, in inflammation. We have recently described liver-enriched innate-like TCRαβ+CD8αα regulatory T cells, partly controlled by the non-classical MHC molecule, Qa-1b, that upon adoptive transfer protect from T cell–induced colitis. In this study, we found that TCRαβ+CD8αα T cells are reduced among the colonic IEL during inflammation, and that their activation with an agonistic peptide leads to significant Qa-1b–dependent protection in an acute model of colitis. Cellular expression of Qa-1b during inflammation and corresponding dependency in peptide-mediated protection suggest that Batf3-dependent CD103+CD11b− type 1 conventional dendritic cells control the protective function of TCRαβ+CD8αα T cells in the colonic epithelium. In the colitis model, expression of the potential barrier-protective gene, Muc2, is enhanced upon administration of a Qa-1b agonistic peptide. Notably, in steady state, the mucin metabolizing Akkermansia muciniphila was found in significantly lower abundance amid a dramatic change in overall microbiome and metabolome, increased IL-6 in explant culture, and enhanced sensitivity to dextran sulfate sodium in Qa-1b deficiency. Finally, in patients with inflammatory bowel disease, we found upregulation of HLA-E, a Qa-1b analog with inflammation and biologic non-response, in silico, suggesting the importance of this regulatory mechanism across species. ## INTRODUCTION Immune cells have a distinct pattern of distribution across the mammalian intestine both longitudinally and transversely, influenced at least in part by the microenvironmental niches in different anatomic locations [1]. Thus, intestinal epithelial layers (IELs), which are lymphocytes localized in the epithelial barrier in the colon, may be part of the initial cellular defense that the host mounts to various insults from the luminal side and help decide the fate of the tissue resulting in homeostasis or pathology [2, 3]. Comparatively little is known of colonic IELs compared with their counterparts in the underlying lamina propria (LP) tissue during inflammation [4]. Recent single-cell RNA sequencing (scRNA-seq) studies in human colon have indicated the specialized environment in the epithelial layer where immune cells reside [5]. In addition, how the microbiome, as another component of the gut epithelial microenvironment, helps shape this immune cell mediation in barrier function is poorly understood. TCRαβ+CD8+ T cells are a major component of gut epithelial barrier community in both mice and humans [2, 3]. They are of at least two varieties based on the CD8 dimers: innate-like (also named unconventional or natural) CD8+ T cells bearing CD8αα homodimers, and adaptive (conventional or inducible) CD8+ T cells carrying CD8αβ heterodimers (2–4). It is noteworthy that CD8αα is expressed by four different cell types in the IEL: TCRαβ+CD8αα+CD8αβ+, TCRαβ+CD8αα+ CD8αβ−, TCRγδ+CD8αα+, and TCR−CD8αα+. Because IELs are enriched among intestinal cells expressing CD8αα homodimers [2], we hypothesized that a careful analysis of TCRαβ+ CD8αα+ T cells may provide a biomarker for epithelial barrier function, a quintessential element that is compromised in inflammatory bowel disease (IBD) and its murine models. Although in human intestine, the presence of CD8αα+ T cells has been controversial [2, 6], recent evidence from scRNA-seq profiling reveals the TYROBP+ CD3+CD8+ IEL population as natural and the TYROBP− CD3+CD8+ IEL population as the inducible counterpart to the murine cells in human colon [7, 8]. Although both of these cell types were enriched in NK cell markers, the TYROBP+ population was reduced during inflammation whereas there was an expansion in the TYROBP− population [8]. The latter behavior suggests a homeostatic role for the TYROBP+CD8+ T IELs. Functionally, these innate-like CD8+ T cells have been mostly studied in the context of the small intestinal epithelial layer in mice. We have recently demonstrated innate-like TCRαβ+CD8αα+ regulatory T cells (Tregs) that, upon adoptive transfer, protect recipient mice from T cell–induced colitis in a perforin-dependent manner [9]. These cells have a polyclonal TCR repertoire and an activated/memory phenotype and are distinct from NKT and mucosa-associated invariant T cells and, accordingly, are present in CD1d−/− and Mr1−/− mice as well as in germ-free mice. Additionally, these TCRαβ+ CD8αα+ T cells are Foxp3 negative, and a sizeable portion of them are restricted by the non-classical MHC molecule Qa-1b [9, 10]. Recently, Qa-1b has been demonstrated to be a high-affinity functional ligand for the CD8αα homodimer [11]. In addition, Qa-1b-restricted CD8+ T cells have previously been reported to be immunoregulatory T cells in experimental autoimmune encephalomyelitis (EAE) and lupus (12–15). The Qa-1-restricted TCRαβ+CD8+ T cells express CD44, CD122, and the NK inhibitory receptor Ly49CIFH (16–18). In addition to PLZF, innate-like regulatory TCRαβ+CD8+ T cells also express the transcription factor Helios [9, 19]. How these innate-like T cells coordinate function, especially in the context of inflammatory assault in the much thicker mucus layer and a higher load of commensal microbiota adjacent to the colonic epithelium, is poorly understood. In this study, we demonstrate a unique immunoregulatory niche in the intestinal epithelial barrier in the steady state and in inflammation that can be specifically manipulated by engaging class Ib MHC-dependent TCRαβ+CD8αα+ T cell-type 1 conventional dendritic cell (cDC1)-mediated immune-microbiome interactions, to maintain mucosal homeostasis in colon. Additionally, using in silico methods and IBD datasets, we also probe human relevance for this unique pathway. ## Mice C57BL/6 (B6) mice and Batf3−/− mice on a B6 background were purchased from The Jackson Laboratory (Bar Harbor, ME). Qa-1b−/− mice (originally provided by Dr. P. Jensen, University of Utah) on a B6 background were bred in our own facility in the University of California San Diego. For the microbiome comparison studies, age-matched wild-type (WT) B6 mice of different sexes were purchased and maintained for >6 mo, and differently aged mice feces were collected separately and compared. For colitis experiments, mice were kept in our animal house in the same room for at least 2 wk before experimentation. Animal studies were carried out in strict accordance with the recommendations of the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocols were reviewed and approved by the Institutional Animal Care and Use Committee of the University of California San Diego. ## Isolation of colonic cells Colonic cell isolation was performed as previously published with a few modifications [20]. Briefly, colons were removed, cleaned of mesentery and feces, and opened up longitudinally. They were then transversely cut into small pieces of ~1 cm in length, and intraepithelial and LP fractions were extracted following the manufacturer’s protocol (LP dissociation kit, a MACSmix tube rotator, Miltenyi Biotec). Following two rounds of predigestion solution using the MACSmix tube rotator, the IEL fraction was generated. The remaining tissue was washed to remove the EDTA, and supernatant was again pooled with the IEL fraction (third wash). The colon pieces were then incubated in digestion solution (containing enzymes) at 37°C and gently shaken as mentioned using the MACSmix tube rotator. Finally, tissue pieces were disrupted using the m_intestine_01 program of the gentleMACS dissociator and passed first through a 100-μm filter and then through a 70-μm filter to obtain the LP fraction. For the small intestinal IEL fraction, a similar method was followed. For mesenteric lymph node (MLN), lymph nodes were carefully isolated and made free of extra fat and then physically mashed gently using the plunger of a 5-ml syringe while passing through a 70-μm filter to obtain single-cell suspensions. ## Flow cytometry Single-cell suspensions from different tissue sources were first incubated with anti-CD16/CD32 (Fc block) in FACS buffer ($0.02\%$ sodium azide/$2\%$ FBS/PBS) and then surface stained with Abs against CD45 (clone 30-F11), CD45R/B220 (clone RA3–6B2), CD4 (GK1.5 or RM4–5), TCRb (H57–597), CD8a (53–6.7), CD8b (YTS156.7.7), CD8b0.2 (53–5.8), CD25 (7D4 or PC61), CD122 (TM-b1), Ly49CIFH (14B11), and CD44 (IM7) and subsequently stained intracellularly for Foxp3 (3G3) and Helios (22F6) for the T cell panel. Separate surface labeling was performed on the same single-cell suspension samples for the myeloid cell panel with Abs against CD45 (clone 30-F11), MHC class II (M$\frac{5}{114}$), CD11c (HL3 or N418), CD11b (M$\frac{1}{70}$), CD103 (2E7), CD3 (17A2 or 145–2C11), CD19 (6D5), Siglec H [551], PDCA-1 (129C1), CD45R/B220 (clone RA3–6B2), and Qa-1b (6A8.6F10.1A6). Abs were purchased from BD Biosciences, BioLegend, or Thermo Fisher Scientific. Following surface staining for each sample, live/dead staining was performed using eFluor 780 fixable viability live/dead dye staining (eBioscience), following the manufacturer’s protocol. Intracellular transcription factor staining was performed using Foxp3/transcription factor staining buffer set (eBioscience) according to manufacturer’s protocol. Samples were further resuspended in BD Biosciences stabilizing fixative and stored at 4°C until acquisition. Cells were analyzed on a FACSCalibur or FACSCanto (BD Biosciences) at the Flow Cytometry Research Core Facility (Veterans Affairs San Diego Healthcare System, San Diego, CA), and all analyses were performed using FlowJo v10 software (Tree Star). ## Acute DSS-induced colitis Mice were supplied with $2.5\%$ or $1.5\%$ DSS (molecular mass 40–50 kDa, Affymetrix, USB) in drinking water for 7 d with replenishment on day 3. After 7 d, DSS containing water was replaced with regular drinking water and mice were sacrificed at the peak of disease or at different time points as indicated. Body weight and water consumption were monitored regularly and mice were sacrificed by CO2 euthanasia followed by cervical dislocation. Mice were immunized i.p. ( 50 μg/mouse) with either a Qa-1b–binding 9-mer peptide LFFVLSSLL, which activates regulatory CD8+ T cells in B6 mice (H. Sheng, I. Marrero, I. Maricic, and V. Kumar, manuscript in preparation), emulsified in IFA or buffer in the control group once 4 d before starting DSS water. ## Histology and gross parameters in colitis model After sacrifice, colons were isolated and cleaned. Cleaned colons were measured for length and stool softness. Colon length change indicates a decrease in colon length in centimeters from mean length (8 cm for male and 7.5 cm for female) of untreated B6 colons of matched age, sex, and similar housing. Stool softness was scored on a scale of 0–3: 0, normal; 1, mildly soft stool; 2, soft but not runny stool; 3, completely runny and bloody stool. The entire mouse colon barring the last 1 cm was either used for FACS or rolled up into “Swiss rolls” and fixed in Bouin’s fixative (RICCA Chemical, Arlington, TX). The fixed colon rolls were then sectioned and stained with H&E at the University of California San Diego Histology Core. The processed slides were scored in a blinded fashion by a trained pathologist for ulcerations in micrometers and then converted to centimeters. Finally, the colon length decrease in centimeters, stool softness score, and histological ulcer score in centimeters were summed to arrive at the cumulative colitis score on a scale of 0–20. Exceptions were made for Fig. 3E and Supplemental Fig. 2B, where histology scoring was not available due to utilization of tissue for a different purpose, and, accordingly, for these figures the y-axis was altered to reflect this change. ## Colon explant culture and analysis of supernatant The last 1 cm of colon from the distal end was cut and washed twice in PBS. The washed piece of colon was then put in culture overnight in a 24-well plate with 1 μl of RPMI 1640 without serum but containing $1\%$ penicillin and streptomycin. Following incubation, culture fluid was carefully collected and centrifuged, and supernatant was stored at −80°C for analysis later. ELISA or a cytometric bead array (CBA) kit was used to measure cytokines in the supernatant. ## Quantitative real-time PCR of murine genes Total RNA was isolated from proximal and distal colon tissues using an RNeasy mini kit (Qiagen, 74104). RNA was quantified using a NanoDrop 2000c spectrophotometer (Thermo Fisher Scientific) and reverse transcribed using a Maxima first-strand cDNA synthesis kit for quantitative real-time PCR (Thermo Fisher Scientific, K1672). Real-time PCR was performed using Maxima SYBR Green/ROX quantitative PCR (qPCR) master mix (2×) (Thermo Fisher Scientific, K0241) on a StepOnePlus real-time PCR system (Applied Biosystems) with specific PCR primer pairs. Data were analyzed using relative mRNA gene expression over mRNA housekeeping gene (L32) expression. ## CBA assay To measure IL-6, TNF-α, IFN-γ, and IL-17 released in supernatant from colon explant culture, a CBA mouse Th1/Th2/Th17 cytokine kit (BD Biosciences, 560485) was used. A CBA assay was performed according to the manufacturer’s protocol. Data were analyzed using FCAP Array (Soft Flow). ## 16S rRNA gene sequencing data acquisition and processing Samples were processed and for 16S sequencing as previously described [21, 22]. Briefly, the V4 hypervariable region of the 16S rRNA gene was amplified using barcoded 515f-806r primers and followed by 2 × 150-bp sequencing on the Illumina MiSeq. Raw sequences were processed in QIIME 2 within Qiita and denoised to sequence variants with deblur (23–25). α Diversity was calculated using the Shannon index after rarefaction to 6300 sequences per sample. β Diversity, measured by robust Aitchison distance, was calculated and a biplot was created using DEICODE [26]. Statistical significance of difference between bacterial composition of genotypes was determined by permutational ANOVA (PERMANOVA) using the robust Aitchison distance matrix. The data have now been submitted to the European Nuclear Archive under accession number ERP132208 (http://www.ebi.ac.uk/ena/data/view/ERP132208). ## Akkermansia qPCR For the quantitative detection of Akkermansia muciniphila DNA via qPCR, we used the following primers 5′-CAGCACGT-GAAGGTGGGGAC-3′ (forward) and 5′-CCTTGCGGTTGGCTT CAGAT-3′ (reverse) as described previously [27]. Each amplification reaction was carried out in a total volume of 10 μl with 5 μl of SsoAdvanced universal SYBR Green supermix (Bio-Rad Laboratories), 0.5 μl of each primer (10 μM stock; Integrated DNA Technologies), 3.5 μl of PCR-grade water (Invitrogen), and 1 μl of purified genomic DNA. Standard curves for qPCR data analysis were generated based on 10-fold serial dilutions (from 5 to 0.0005 ng/μl) of cultured A. muciniphila. We then estimated gene copy counts by converting the detected DNA concentration by qPCR using the following equation: Gene copies = (DNA concentration (nanograms) × Avogadro’s number/amplicon length (base pairs) × Daltons) × 1.0E+09, where amplicon length was determined using National Center for Biotechnology Information reference sequence NR_042817.1. ## Metabolomics data acquisition and processing Mass spectrometric data were generated and processed using a previously published protocol [28]. In brief, samples were lyophilized, resuspended in $50\%$ MeOH $50\%$ water (Optima liquid chromatography–mass spectrometry grade; Fisher Scientific, Fair Lawn, NJ), and analyzed on an ultra-HPLC system (UltiMate 3000; Thermo Fisher Scientific, Waltham, MA) coupled to a Maxis quadruple time of flight mass spectrometer (Bruker Daltonics, Bremen, Germany) with a Kinetex C18 column (Phenomenex, Torrance, CA) in the positive electrospray ionization mode using a linear gradient of the mobile phase of water with $0.1\%$ formic acid (v/v) and acetonitrile with $0.1\%$ formic acid (v/v) (liquid chromatography–mass spectrometry-grade solvents; Fisher Chemical). MZmine2 (version MZmine-2.37.corr16.4) [18] was used for MS1 feature finding as previously described [28], followed by feature-based molecular networking using GNPS [29]. The feature table was normalized by converting each sample to relative abundances. β Diversity was calculated and biplot was created using DEICODE. Statistical significance of difference between metabolome composition of genotypes was determined by PERMANOVA using the robust Aitchison distance matrix. Log ratios of bile acids were calculated by summing intensities of bile acid subtypes within each sample, followed by taking the log ratio of primary-to-primary–conjugated bile acids within each sample. ## Bioinformatics analysis of human datasets Human gene expression data were downloaded from the Gene Expression Omnibus database (accession numbers GSE16879 [https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE16879], GSE73661 [https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE73661], and GSE100833 [https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE100833]). The data were then background corrected, normalized with the robust multiarray averaging procedure, and transformed with an oligonucleotide package [30] from R with Bioconductor version 3.10. The gene expression difference was examined by two-way ANOVA in R. Similarly, human proteomics data were downloaded from the Proteomics Identification Database with accession number PXD012284 (https://www.ebi.ac.uk/pride/archive/projects/PXD012284). The data were variance stabilization–normalized and missing values were imputed with the DEP package [31] from R with Bioconductor version 3.10. Human scRNA data were downloaded from Single Cell Portal by accession number SCP259 (https://singlecell.broadinstitute.org/single_cell/study/SCP259/intra-and-inter-cellular-rewiring-of-the-human-colon-during-ulcerative-colitis), with cell type assignment from the original publication [5]. The differential expression analysis was performed by the MAST generalized linear model framework [32] at default parameters with implementation in the Seurat package version 3.1.4 [33]. ## Statistical analysis Data were analyzed using GraphPad Prism v7 software (GraphPad Software). Data are reported as mean ± SEM. The two-tailed unpaired t test, Student t test, or Mann–Whitney U test was used when comparing two groups of unpaired data. A one-way ANOVA with the Bonferroni multiple comparison posttest was used when comparing three or more groups. Significance was assessed using two-tailed tests and is indicated as follows: $p \leq 0.05$, $p \leq 0.01$, $p \leq 0.001$, *$p \leq 0.0001.$ ## TCRαβ+CD8+ T cell niche in colonic IEL is different from other mucosal compartments in steady state To gain a better understanding of steady-state composition of TCRαβ+CD8+ T cells, we first analyzed the presence of innate-like TCRαβ+CD4−B220−CD8αα+ T cells and their adaptive counterpart TCRαβ+CD4−B220−CD8αβ+ T cells in different gut mucosal compartments: colon IEL and LP fractions, small intestinal IEL, and MLN from naive B6 mice. We found that the frequency of TCRαβ+CD8αα+ T cells in colonic IEL (as compared with other compartments) was much higher than that of CD8αβ+ T cells (Fig. 1A). Next, we determined whether the cell surface phenotype of TCRαβ+CD8αα+ cells is similar to that displayed by regulatory CD8+ T cells. We found that the TCRαβ+CD8αα+ T cell population among the colon IEL expresses two markers, CD44 and the NK inhibitory receptors Ly49CIFH, whereas the expression of CD122 is low/no. In contrast, neither the TCRαβ+CD8αα T cells in colon LP nor CD8αβ+ T cells in colonic IEL or colon LP express Ly49CIFH (Fig 1B). Consistent with their regulatory phenotype, colonic TCRαβ+CD8αα+ T cells predominantly were Helios+ Foxp3− in the colon IEL fraction in comparison with the colon LP cells (Supplemental Fig. 1A). Similar to the unique niche of CD8+ T cells in murine colonic IEL, in silico analysis of CD8+ cells in the epithelial layer versus LP in healthy human colon also revealed a differential gene expression profile underlining the uniqueness that intraepithelial residence provides to these lymphocytes in the steady state (Supplemental Fig. 1B). Our murine data indicate a unique phenotype of the colon IEL TCRαβ+CD8αα+ T cells containing potential regulatory CD8+ populations in the steady state in naive B6 mice. ## Colonic IEL TCRαβ+CD8αα+ and CD8αβ+ T cells correlate differentially with induction and restoration phases in an acute model of colitis Next, we determined the dynamics of different TCRαβ+CD8+ T cell subsets during acute inflammation in the colon. We used body weight loss as a general disease parameter in an acute model of DSS colitis. Mice started to lose body weight on day 5 after initiation of DSS ($2.5\%$), with the peak of weight loss occurring on day 9 and subsequent weight gain (Fig. 2A). We found that the percentages of TCRαβ+CD8αα+ but not CD8αβ+ T cells in the colon IEL were markedly decreased during the peak disease phase (Fig. 2B, 2C). In contrast, their percentage in the LP was not significantly affected under these conditions. In comparison, TCRαβ+CD8αβ+ T cells increased by 10-fold among the colon IELs during the most severe phase of the disease, whereas CD4 T cells increased later in the recovery phase (Fig. 2C). In the LP fraction, we could only observe an increase in CD4+ T cells during the restoration phase (Fig. 2D). Thus, a significant increase in TCRαβ+CD8αβ+ T cells in the colon IEL during inflammation suggests that they might contribute to disease, whereas a significant decrease in TCRαβ+CD8αα+ T cells is consistent with a regulatory role. The quantification of absolute cell numbers also concurred with a significant accumulation of TCRαβ+CD8αβ+ T cells (Fig. 2E). Although the decrease in TCRαβ+CD8αα+ T cells was not significant, their increase was significant during the restoration phase (Fig. 2E). In addition, an enhancement in CD4 T cell numbers in IEL and LP was observed, but no significant alterations were found in CD8+ T cell subset numbers in LP (Fig. 2E, 2F). ## Administration of an agonistic peptide for the TCRαβ+CD8αα+ T cells protects colitis in WT mice but not in Qa-1b−/− mice We have recently identified an agonistic 9-mer peptide that stimulates Qa-1 restricted regulatory TCRαβ+CD8αα+ T cells in vitro and in vivo and protects mice from EAE (H. Sheng, I. Marrero, I. Maricic, and V. Kumar, manuscript in preparation). Since TCRαβ+CD8αα+ T cells are decreased during colitis, we wanted to investigate whether prophylactic in vivo stimulation of these T cells with the agonistic peptide would confer protection in the acute model of colitis. Accordingly, we found that the administration of the agonistic peptide in WT B6 mice resulted in a significant reduction in gross colon parameters and cumulative colitis scores on the day of peak body weight loss (Fig. 3A, 3B). Explant cultures from peptide-treated mice also showed significant reduction in proinflammatory cytokine production (Fig. 3C). Although we observed that the peptide-treated mice gained weight after DSS treatment, we found that this weight gain was due to the IFA used to emulsify the peptide, and, therefore, in IFA-only controls, body weight was enhanced but without any effect on other colonic disease parameters (Supplemental Fig. 2A, 2B). To avoid this confounding variable, for further experiments with the peptide, we omitted body weight loss as a disease parameter and focused instead on other disease parameters to measure colitis. Next, to determine whether the peptide-mediated protection is Qa-1–dependent, we administered peptide in mice deficient in Qa-1b. In contrast to WT mice, peptide treatment failed to provide protection in Qa-1b−/− mice as observed by gross colonic parameters and cytokines in explant cultures (Fig. 3D–F). These data indicate that the protection from colitis induced following peptide treatment is Qa-1b-dependent. ## CD103+CD11b− cDCs populate colonic IEL during peak of DSS-induced colitis and uniquely upregulate Qa-1b We have previously demonstrated that DC-mediated Ag cross-presentation is required to induce TCRαβ+CD8αα+ T cells and facilitate their immunoregulatory function in EAE [34]. There are several well-known subsets of professional APCs in mammalian intestine, and some of them have been implicated in immunoregulation (35–37). Of these subsets, cDC1s, characterized as CD103+CD11b− in the murine intestine, are also known to display enhanced Ag cross-presentation properties in several situations [38, 39]. Recently, intestinal cDC1s in mice have been reported to maintain tolerance to epithelial Ags involving a Foxp31CD8+ Treg-inducing mechanism [40]. As a first step toward determining whether DCs play a role in mediating the anti-inflammatory functions of CD8αα+ T cells, we characterized the phenotypes of DCs in the intraepithelial compartment, where TCRαβ+CD8αα+ T cells mainly reside, during colitis. We found a significant increase in the percentage of several DC subsets in the IEL fraction, including DCs that are barely present in the IEL fraction in the steady state, such as CD103+CD11b− cDCs, but they were mostly found in the LP fraction. Additionally, DCs that we found to be present in the steady-state IEL included plasmacytoid DCs (pDCs) and CD103−CD11b+ cDCs (Fig. 4A, 4B, Supplemental Fig. 2C). Thus, we found that CD103+CD11b− cDCs (cDC1s), but not pDCs or CD103−CD11b+ cDCs (cDC2s), significantly decreased, by frequency and numbers, among colon IEL after day 8, which correlated with the change in TCRαβ+CD8+ T cells at that time. Based on absolute numbers, we also found a general increase in both LP and IEL on day 16 compared with other days (Supplemental Fig. 2C). Next, we determined the expression of Qa-1b among various DC subsets and macrophages during colitis in colon IEL from WT mice. Analysis of the surface levels (geometric mean fluorescence intensity) of Qa-1b revealed that CD103+CD11b− cDCs in colon IEL are the major Qa-1b-expressing DC population (Fig. 4C, 4D). Thus, a significant correlation between the numeric changes in Qa-1b+ CD103+CD11b− cells and in the relevant subsets of TCRαβ+CD8+ T cells suggests that cDC1s may be important for the peptide-mediated Qa-1b-dependent protection from colitis. ## Peptide-mediated protection from DSS-induced colitis is lost in Batf3−/− mice lacking cDC1s CD103+CD11b− cDCs are known to be developmentally controlled by various transcription factors, including Batf3 [41]. Batf3 also controls CD8αα+ cDCs in the spleen and other lymphoid tissues that, along with CD103+ cDCs form the cDC1s, are known for playing a role in intestinal antiviral CD8+ T cell-mediated immunity and tolerance induction (42–45). We first confirmed that Batf3−/− mice have a substantially reduced number of CD103+CD11b− cDCs in the colon (Supplemental Fig. 3A). Additionally, consistent with a compromised CD8+ T cell function in Batf3−/− mice [46], the frequency of TCRαβ+CD8αβ+ T cells was also reduced in both colonic IEL and LP in these mice (Supplemental Fig. 3B). Importantly, we found that the protection induced by the Qa-1 agonistic peptide against acute colitis (with $2.5\%$ DSS) was lost in Batf3−/− mice in comparison with WT mice (Fig. 5). Importantly, however, note that the severity of colitis observed in Batf3−/− mice was apparently lower in comparison with that in the WT mice. To address this potential confounding factor, we induced colitis in Batf3−/− mice with an increased dose of DSS ($3.5\%$), but again observed that peptide-mediated protection was lost in these mice (Supplemental Fig. 3C). These data suggest that the immune regulation of colitis is dependent on the presence of the Batf3-dependent Qa-1b+ cDC1 population in the inflamed epithelial layer of the colon. ## Peptide-mediated protection results in a selected upregulation of barrier function genes including Muc2 in distal but not in proximal colonic tissue We wanted to determine whether the CD8+ T cell/CD103+ DC–mediated immune regulation impacts epithelial functions. We used qRT-PCR analysis to investigate changes in the expression of colon-expressed genes known for barrier protection in peptide-treated mice in comparison with the control group following DSS-induced colitis. We analyzed proximal and distal colon separately since they have displayed differential gene expression [47, 48] and observed that only in the distal, but not proximal, colon peptide administration enhanced the barrier function gene, Muc2 (Fig 6). Notably, Muc2 deficiency causes spontaneous colitis in mice [49]. Muc2 has been shown to induce tolerogenic DCs in mice and humans and maintains gut homeostasis and oral tolerance [50]. The other important gene found to be upregulated in the distal colon was Krüppel-like factor 4 (Klf4), and there were nonsignificant trends observed in the junctional molecule occludin (Ocln), but there were no trends observed in other molecules investigated, that is, MUC4, ZO1, and CLDN4 (Fig. 6). These data suggest that peptide-mediated activation of TCRαβ+CD8αα+ T cells in colon enhances expression of genes whose products play a key role in the gut barrier function. ## In the steady state, lack of Qa-1b is associated with a dramatically altered microbiome and metabolome The gut microbiome impacts intestinal barrier function and can modulate colonic inflammation [51, 52]. Because we observed enhancement of mucin gene expression after administration of the Qa-1b–agonistic peptide in the colitis model, we examined whether the Qa-1b pathway has any impact on intestinal microbiota or metabolome that could affect the epithelial barrier even during steady state. We therefore collected fecal samples for 16S rRNA gene amplicon sequencing and untargeted metabolomics analysis from WT and Qa-1b−/− mice of different ages and sex housed in our animal facility and at two different time points separated by 6 mo. We adopted this randomized sampling to avoid potential confounding influences, such as cage-associated effect, on the microbiota other than Qa-1b deficiency. Principal-component analysis (PCA) of the robust Aitchison distances based on the 16S sequencing data revealed that the microbiome profiles of WT and Qa-1b−/− mice were significantly different (PERMANOVA $F = 73.5$, $p \leq 0.001$) (Fig. 7A). The* genera Prevotella and Akkermansia appeared to drive the separation between the genotypes with lesser contributions by the genera Bacteroides and Bacillus along with an undetermined genus from the family Rikenellaceae. Additionally, α diversity, as measured by the Shannon index, was significantly higher in Qa-1b−/− mice compared with WT (Mann-Whitney $U = 3166$, $$p \leq 3.8$$e 12, Fig. 7B). Further analysis of microbes that were differentially abundant between the groups were examined via Songbird [53], revealing that Qa-1b−/− mice had higher proportions of the genera Prevotella, Bacteroides, and Lactobacillus, whereas WT mice were associated with higher proportions of Akkermansia and Bacillus. Differential abundance of Akkermansia was of particular interest because the genus is known to affect mucin production and regulate responses of the gut epithelium in mice and in humans (54–56). To confirm that abundance of A. muciniphila was different between WT and Qa-1b−/− mice, we measured Akkermansia abundance by qPCR and found that the absolute abundance of Akkermansia in Qa-1b−/− mice was significantly lower than in WT mice (Mann-Whitney $U = 603$, $$p \leq 4.6$$e 11) (Fig. 7C). As a measure of the potential functional consequences of the microbial changes, we performed untargeted metabolomics on the same fecal samples used for 16S analysis. Similar to the microbiome, PCA of the β diversity based on robust Aitchison distances revealed that the metabolite profile of Qa-1b−/− mice was significantly different from WT (PERMANOVA $F = 14.5$, $p \leq 0.001$) (Fig. 7D). Several of the metabolites most associated with Qa-1b−/− mice are known, as are novel bile acids [57], such as cholic acid and hyocholic acid. Bile acid abundance and composition vary due to both host and microbial contributions [57], and bile acids may be conjugated in the liver, which affects their functions in fat absorption in the intestine. Qa-1−/− mice had a higher ratio of unconjugated to conjugated primary bile acids compared with WT mice (Mann-Whitney $U = 178$, $$p \leq 3.6$$e−10) (Fig. 7E) based on Songbird analysis [53]. ## In steady state, Qa-1b−/− mice harbor a proinflammatory environment in colon and are accordingly more sensitive to DSS-induced colitis Next, we investigated whether the colonic tissues in steady state in Qa-1b−/− mice harbor an enhanced proinflammatory milieu in the steady state. We found enhanced lymphocyte infiltrations into colonic tissue of Qa-1b−/− in comparison with WT mice (Fig. 8A). Consistently, TCRαβ+CD8αβ+ adaptive T cells are significantly increased in steady state in the IEL in these mice (Supplemental Fig. 2D). Additionally, in the tissue explant culture, we also observed significantly higher levels of IL-6 in Qa-1b−/− mice in comparison with WT mice (Fig. 8A). Because CD4 Tregs are important to control inflammation, we also investigated CD4+Foxp3+ Tregs to probe whether Qa-1b deficiency compromised colonic CD4+ Tregs. We did not find a significant reduction in colon IEL and colon LP fractions of CD4+Foxp3+ Tregs in Qa-1b−/− mice in comparison with WT mice (Supplemental Fig. 2E). We then investigated whether Qa-1b−/− mice, because of this proinflammatory milieu, respond differentially in the induction of colitis. We chose a dose of DSS that does not precipitate properly into colitis in WT mice in our animal house, i.e., $1.5\%$ DSS instead of the regular $2.5\%$ DSS. When feeding age-matched mice with $1.5\%$ DSS, we observed that whereas the WT mice did not respond with a change in body weight loss, Qa-1b−/− counterparts suffered significant loss of body weight (Fig. 8B). Accordingly, Qa-1b−/−mice treated with DSS also developed enhanced gross and histopathological features of colitis (Fig. 8C, 8D). Explant culture of colon tissue demonstrated upward trends in proinflammatory cytokines from Qa-1b−/− mice treated with low-dose DSS (Fig. 8E). Taken together, our data demonstrate a proinflammatory milieu in colon when the Qa-1b-based regulatory pathway is absent. ## IBD patients display a unique pattern of expression of HLA-E, an ortholog for the murine Qa-1b Similar to the murine non-classical MHC molecules, humans also express a conserved class Ib MHC molecule, HLA-E, which shares several features with the murine counterpart [58, 59]. Therefore, we determined whether HLA-E expression is altered in IBD patients and more specifically in any specific cell subset recently reported in IBD [5]. We first compared differential gene expression of two class Ib MHC molecules (HLA-E and CD1d) and one class Ia MHC molecule (HLA-A) between non-inflamed and inflamed colon from descending and ascending colon in refractory Crohn’s disease (CD) using the CERTIFI cohort (GSE100833) [60]. Human gene expression data were downloaded from the Gene Expression Omnibus database (60–62). Interestingly, in the descending colon, but not the ascending colon, we observed significantly enhanced gene expression of HLA-E, but not CD1d or HLA-A (Fig. 9A). The enhanced HLA-E gene expression was also supported by increased protein expression from a different colonic CD cohort [63], which included treatment-free samples and samples from patients treated with various medications (Fig. 9B). Human proteomics data were downloaded from the Proteomics Identification Database [63]. Next, expression analysis of therapeutic response studies with Infliximab and vedolizumab allowed us to look into the induced restitution phase of the inflammatory disease in patients. In one study, colonic mucosal biopsies were obtained from ulcerative colitis (UC) patients treated with either vedolizumab or infliximab. Samples from before and after therapy along with outcome (response by histological healing to the therapeutic intervention or not) were compared with non-IBD controls [62]. In another study, results for before and after the first infliximab treatment in ileal versus colonic CD were compared with those for non-IBD controls [61]. In UC colon datasets and in colonic CD, but not in ileal CD, we observed decreased gene expression of HLA-E and not of CD1d and HLA-A (observed in CD) in therapeutic response and also not in non-responders (Fig. 9C, 9D). Next, on a cellular level, taking cue from our murine studies, mapping prominent myeloid cell subsets in UC obtained from a published scRNA-seq dataset [5], we observed overall enhanced expression of HLA genes similar to our observations as in Fig. 9A. Interestingly, only in cDC1, but not in other myeloid subsets, there was an enhancement of HLA-E expression by both gene average expression level and by abundance (Fig. 9E, Supplemental Table I). Thus, a similarity in the cell-specific expression profile of HLA-E in humans and Qa-1b in mice in anatomically similar inflamed tissues suggests potential involvement of this regulatory axis in IBD. ## DISCUSSION In this study we show the dynamic nature of TCRαβ+CD8αα+ versus TCRαβ+CD8αβ+ T cell subsets that reside in the epithelial layer of colon in the steady state and during colitis. Notably, in the induction phase of the inflamed state, the frequency of TCRαβ+CD8αα+ T cells decrease with a significant increase in the frequency of the TCRαβ+CD8αβ+ T cells, and in resolution phase the reverse trends are observed. Expression of a non-classical class I MHC molecule, Qa-1b, which is known to be involved in regulatory CD8+ T cell-mediated immune regulation, is significantly upregulated in the Batf3 controlled cDC1s in inflamed epithelium. An agonistic peptide that stimulates Qa-1b-restricted regulatory TCRαβ+CD8αα+ T cells confers protection in the acute model of colonic inflammation in a Qa-1b- and Batf3-dependent manner, and enhances expression of known barrier-protective genes in distal colon. Consistent with a crucial role of the Qa-1b regulatory axis, deficiency of Qa-1b molecules in the steady state leads to a proinflammatory milieu in distal colonic tissue with a distinct microbiota, including a significant decrease in mucin-metabolizing bacteria and a significant alteration in the conjugated versus unconjugated primary bile acids. Expression of the human ortholog of Qa-1b, HLA-E, is overall enhanced with inflammation in colonic tissues from IBD patients and in biologic non-response subjects. Among myeloid cell subsets in humans, similar to our observations in mice, HLA-E expression in cDC1s also increases in comparison with all other myeloid cell subsets examined. Collectively, these studies identify a novel colonic immune regulatory mechanism involved in the maintenance of gut-barrier homeostasis. CD8+ T cells with regulatory activity to counter proinflammatory T cells have been reported for a long time. Reports on CD8+ Treg phenotype have also been published (9, 10, 12, 14–19, 64, 65). Although both class Ia and class Ib MHC molecules have been implicated, a large number of studies have focused on regulatory CD8+ T cells restricted by the class Ib MHC molecule, Qa-1b (9, 10, 12, 14–17, 64, 65). Recently, in both EAE and in patients with multiple sclerosis, classical class I MHC–restricted CD8+ Tregs have also been recently described [66]. Several studies have also shown that CD8+ Tregs were either reactive to novel peptide epitopes or peptides derived from the TCR b-chain [12, 14, 15] and inhibited EAE by suppressing the anti-myelin oligodendrocyte glycoprotein encephalitogenic CD4+ T cells. We have recently identified a novel immunoregulatory TCRαβ+CD8αα+ T cell subset in both mice and humans with innate-like features that is dependent on the transcription factor PLZF capable of controlling both EAE and T cell-induced colitis [9]. Consistently, glatiramer acetate-mediated induction of Qa-1 restricted lymph node-derived CD8+ T cells has been shown to also prevent DSS-induced colitis [67]. Our data indicate that in contrast to TCRαβ+CD8αβ+ T cells in colonic IEL or colonic LP CD8+ T cells, TCRαβ+ CD8αα+ T cells displaying a regulatory phenotype in steady-state CD44+CD122loLy49CIFH+ T cells mostly reside within the colonic IEL CD8αα+ T cells in comparison with CD8αβ+ T cells. Consistently, Helios+ Foxp3−CD8+ Tregs are also enriched in the colonic IEL CD8αα+ T cells in the steady state (data not shown). These data indicate that colonic IEL is a special repository of potential immunoregulatory TCRαβ+CD8αα+ T cells. Thus, TCRαβ+CD8αα+ T cells in colonic IEL are reduced in the inflammatory disease phase and are restored in the resolution phase, in sharp contrast to the adaptive CD8αβ+ T cells, a situation analogous to TYROBP+/− IELs in human colon in UC [8]. Importantly, in addition, note that these alterations in the ratio of CD8αα+ T versus CD8αβ+ T cells are restricted to the colonic IEL population and are not reflected in the colonic LP counterparts in mice, thus indicating epithelial layer damage and restoration in acute DSS colitis. Collectively, our data suggest that the TCRαβ+CD8αα+ T cells, or a subpopulation thereof, within colonic IEL support the integrity of the epithelial barrier from inflammatory assault by interacting with their immediate and prominent neighbors, that is, the epithelial cells or other immune cells within the epithelium. Notably, during inflammation, Qa-1b molecules are significantly expressed on the cDC1 subset defined by the CD103+CD11b− phenotype. Usually this subset resides in LP, but we observed its accumulation in the inflamed epithelial fraction. This is similar to another recent report where this subset of DCs was shown to communicate with epithelial cells in conferring protection in the colitis model [68]. These DC subsets are known to be regulated by the transcription factor Batf3, and consequently the Qa-1b peptide-dependent regulation failed to protect Batf3−/− mice. Loss of protection from colitis in Qa-1b−/− mice further confirms an important role for the class Ib MHC-based regulatory axis. It is likely that in Qa-1b−/− mice, either the Qa-1b-bound peptide is not presented to TCRαβ+CD8αα+ T cells and therefore these cells are not activated and thus unable to control colitis. Because most of the CD44+CD122loLy49CIFH+ CD8αα+ T cell population in colonic IEL is not lost in Qa-1b−/− mice (data not shown), in the future we will investigate whether these cells are functionally or transcriptionally compromised in the absence of Qa-1b molecules. In summary, these data clearly demonstrate that the cDC1−CD8+ Treg regulatory axis, operating in the inflamed epithelium, nurtures the barrier in the time of distress or damage. Importantly, investigating recently published scRNA-seq datasets from UC and healthy colon biopsy [5], we observed a small upward trend in HLA-E expression in cDC1s, but in contrast to all other myeloid cells studied. It is noteworthy that there is well-known lineage marker conservation between human and mouse cDC1s in the form of Clec9a/XCR1/BTLA/Necl2 positivity and CD14/Sirpα negativity [69, 70]. Our data demonstrating the dependence of a Qa-1b peptide-mediated protection on cDC1s suggest that it will be important to investigate whether a similar HLA-E peptide-dependent cDC1 immune regulatory axis is operational in humans. Interestingly, epithelial and stromal cells have high expression of HLA-E in human colonic datasets in UC [[8] and data not shown] and could play an important role in activating regulatory CD8+ T cells in human gut. Our finding of the Qa-1b-dependent peptide-augmenting gene expression of the well-known barrier protective molecules, such as Klf4 and Muc2, provides a link of the immune system to the microbiome and non-immune compartment. Klf4 is a transcription factor with fundamental roles demonstrated in intestinal epithelial cells, including differentiation of colonic goblet cells [71, 72] and maintenance of cell morphology and polarity [73]. Notably, epithelium-specific ablation of Klf4 results in enhanced pathology in the colitis-associated colorectal cancer model [74]. In future studies we will further investigate role of the Qa-1b agonist peptide in goblet cell differentiation and mucus production in the context of colonic epithelium. Muc2−/− mice elicit spontaneous colitis, affirming its overall protective role in gut inflammation [49]. Mucus secretion is a property of the specialized epithelial cells called goblet cells, and indeed the CD103+ DCs have been shown to capture luminal Ags from Goblet cells and help in protection [75]. However, because Muc2 has been shown to enhance tolerogenic DCs in gut [50], there may be the possibility for a feedback mechanism between Muc2 and tolerogenic DCs induced by the peptide. Interestingly, we observed enhanced Muc2 expression specifically in the distal colon, a region that was also used for our colon explant assays. In UC patients, pathology can be region specific. Thus, this immunoregulatory mechanism, if the biology holds true in humans, may be targeted for patients suffering from inflammation at distal or sigmoid colon. Important support data came from our in silico human studies, in which we observed that in colonic CD patients, HLA-E was augmented preferentially in descending colon as opposed to ascending colon. CD8+ Tregs have been recently demonstrated to be induced by commensal microbes in the context of protection from type 1 diabetes [76]. We found that Qa-1b−/− mice also had a deficiency in mucin-metabolizing microbiota A. muciniphila, whose members, like A. muciniphila, have been described as beneficial commensals in gut, and their potential in IBD therapeutics is being considered (77–79). Thus, this unique immune regulatory axis involves interactions among regulatory CD8+ T cells and mucin-metabolizing bacteria that can be exploited for therapeutics. We also found a preponderance of primary bile acids in the absence of the regulatory axis in the stool of Qa-1b−/− mice, reminiscent of a recent finding in IBD patient samples [80]. Indeed, these enormous changes in microbiome and metabolome in the steady state was associated with a proinflammatory status in the colonic tissue in Qa-1b−/− mice. Notably, the altered unconjugated/conjugated bile acid profile in these mice may suggest either altered liver function and/or reduced secondary modification by gut bacteria and needs further investigation. Among the cytokines, IL-6, which is one of the prominent proinflammatory cytokines in murine colitis models and IBD patients [81, 82], was found to be significantly enhanced in Qa-1b−/− mice without additional inflammatory stimulus. Whether the altered microbiome was the cause of the proinflammatory environment in the steady state in the absence of Qa-1b needs to be further investigated. Because Qa-1b deficiency did not lead to a loss of CD4+Foxp3+ Tregs in colon, the functional loss of the TCRαβ+CD8αα+ Treg-centered regulatory axis accounts for the control of immunoregulation in the colon, thereby suggesting their important role in the maintenance of epithelial barrier function. It is likely that in the steady state, interactions exist between gut microbiota and Qa-1b-expressing cDC1s that may cross-present microbial peptides to TCRαβ+CD8αα+ T cells to maintain membrane integrity. Importantly, these data indicate that both CD4 Tregs as well as CD8+ Tregs are important for the maintenance of immune homeostasis in colon. Our findings also are important in that by defining agonistic peptides for both class Ia and class Ib MHC-restricted CD8+ Tregs, we provide a unique biologics or synthetic chemistry opportunity for potential novel therapeutics for IBD. With literature being enhanced with scRNA-seq and other deep-sequencing studies from IBD patients, human-relevant CD8+ Treg mechanisms should be revealed in the near future. ## References 1. Mowat AM, Agace WW. Regional specialization within the intestinal immune system. Nat. Rev. Immunol (2014) 14 667-685. PMID: 25234148 2. Olivares-Villagómez D, Van Kaer L. Intestinal intraepithelial lymphocytes: sentinels of the mucosal barrier. Trends Immunol (2018) 39 264-275. PMID: 29221933 3. Hu MD, Edelblum KL. Sentinels at the frontline: the role of intraepithelial lymphocytes in inflammatory bowel disease. Curr. Pharmacol. Rep (2017) 3 321-334. PMID: 29242771 4. Cheroutre H, Lambolez F, Mucida D. The light and dark sides of intestinal intraepithelial lymphocytes. Nat. Rev. Immunol (2011) 11 445-456. PMID: 21681197 5. Smillie CS, Biton M, Ordovas-Montanes J, Sullivan KM, Burgin G, Graham DB, Herbst RH, Rogel N, Slyper M, Waldman J. Intra- and inter-cellular rewiring of the human colon during ulcerative colitis. Cell (2019) 178 714-730. PMID: 31348891 6. Mayassi T, Jabri B. Human intraepithelial lymphocytes. Mucosal Immunol (2018) 11 1281-1289. PMID: 29674648 7. Parikh K, Antanaviciute A, Fawkner-Corbett D, Jagielowicz M, Aulicino A, Lagerholm C, Davis S, Kinchen J, Chen HH, Alham NK. Colonic epithelial cell diversity in health and inflammatory bowel disease. Nature (2019) 567 49-55. PMID: 30814735 8. Corridoni D, Antanaviciute A, Gupta T, Fawkner-Corbett D, Aulicino A, Jagielowicz M, Parikh K, Repapi E, Taylor S, Ishikawa D. Single-cell atlas of colonic CD8+ T cells in ulcerative colitis. Nat. Med (2020) 26 1480-1490. PMID: 32747828 9. Sheng H, Marrero I, Maricic I, Fanchiang SS, Zhang S, Sant’Angelo DB, Kumar V. Distinct PLZF. J. Immunol (2019) 203 2150-2162. PMID: 31554695 10. Tang X, Maricic I, Purohit N, Bakamjian B, Reed-Loisel LM, Beeston T, Jensen P, Kumar V. Regulation of immunity by a novel population of Qa-1-restricted CD8αα. J. Immunol (2006) 177 7645-7655. PMID: 17114434 11. Goodall KJ, Nguyen A, McKenzie C, Eckle SBG, Sullivan LC, Andrews DM. The murine CD94/NKG2 ligand, Qa-1. J. Biol. Chem (2020) 295 3239-3246. PMID: 31992596 12. Jiang H, Chess L. The specific regulation of immune responses by CD8. Annu. Rev. Immunol (2000) 18 185-216. PMID: 10837057 13. Hu D, Ikizawa K, Lu L, Sanchirico ME, Shinohara ML, Cantor H. Analysis of regulatory CD8 T cells in Qa-1-deficient mice. Nat. Immunol (2004) 5 516-523. PMID: 15098030 14. Lu L, Werneck MB, Cantor H. The immunoregulatory effects of Qa-1. Immunol. Rev (2006) 212 51-59. PMID: 16903905 15. Smith TR, Kumar V. Revival of CD8. Trends Immunol (2008) 29 337-342. PMID: 18514574 16. Fanchiang SS, Cojocaru R, Othman M, Khanna R, Brooks MJ, Smith T, Tang X, Maricic I, Swaroop A, Kumar V. Global expression profiling of peripheral Qa-1-restricted CD8αα. Hum. Immunol (2012) 73 214-222. PMID: 21889557 17. Kim HJ, Wang X, Radfar S, Sproule TJ, Roopenian DC, Cantor H. CD8. Proc. Natl. Acad. Sci. USA (2011) 108 2010-2015. PMID: 21233417 18. Rifa’i M, Kawamoto Y, Nakashima I, Suzuki H. Essential roles of CD8. J. Exp. Med (2004) 200 1123-1134. PMID: 15520244 19. Kim HJ, Barnitz RA, Kreslavsky T, Brown FD, Moffett H, Lemieux ME, Kaygusuz Y, Meissner T, Holderried TA, Chan S. Stable inhibitory activity of regulatory T cells requires the transcription factor Helios. Science (2015) 350 334-339. PMID: 26472910 20. Liu H, Dasgupta S, Fu Y, Bailey B, Roy C, Lightcap E, Faustin B. Subsets of mononuclear phagocytes are enriched in the inflamed colons of patients with IBD. BMC Immunol (2019) 20 42. PMID: 31718550 21. Thompson LR, Sanders JG, McDonald D, Amir A, Ladau J, Locey KJ, Prill RJ, Tripathi A, Gibbons SM, Ackermann G. A communal catalogue reveals earth’s multiscale microbial diversity. Nature (2017) 551 457-463. PMID: 29088705 22. Maricic I, Marrero I, Eguchi A, Nakamura R, Johnson CD, Dasgupta S, Hernandez CD, Nguyen PS, Swafford AD, Knight R. Differential activation of hepatic invariant NKT cell subsets plays a key role in progression of nonalcoholic steatohepatitis. J. Immunol (2018) 201 3017-3035. PMID: 30322964 23. Bolyen E, Rideout JR, Dillon MR, Bokulich NA, Abnet CC, Al-Ghalith GA, Alexander H, Alm EJ, Arumugam M, Asnicar F. Reproducible, interactive, scalable and extensible microbiome data science using QIIME 2. [Published erratum appears in 2019 Nat. Biotechnol. Immunol (2019) 37 852-857 24. Gonzalez A, Navas-Molina JA, Kosciolek T, McDonald D, Vázquez-Baeza Y, Ackermann G, DeReus J, Janssen S, Swafford AD, Orchanian SB. Qiita: rapid, web-enabled microbiome meta-analysis. Nat. Methods (2018) 15 796-798. PMID: 30275573 25. Amir A, McDonald D, Navas-Molina JA, Kopylova E, Morton JT, Zech Xu Z, Kightley EP, Thompson LR, Hyde ER, Gonzalez A, Knight R. Deblur rapidly resolves single-nucleotide community sequence patterns. mSystems (2017) 2 e00191-16. PMID: 28289731 26. Martino C, Morton JT, Marotz CA, Thompson LR, Tripathi A, Knight R, Zengler K. A novel sparse compositional technique reveals microbial perturbations. mSystems (2019) 4 e00016-19. PMID: 30801021 27. Schneeberger M, Everard A, Gómez-Valadés AG, Matamoros S, Ramírez S, Delzenne NM, Gomis R, Claret M, Cani PD. Akkermansia muciniphila inversely correlates with the onset of inflammation, altered adipose tissue metabolism and metabolic disorders during obesity in mice. Sci. Rep (2015) 5 16643. PMID: 26563823 28. Gauglitz JM, Morton JT, Tripathi A, Hansen S, Gaffney M, Carpenter C, Weldon KC, Shah R, Parampil A, Fidgett AL. Metabolome-informed microbiome analysis refines metadata classifications and reveals unexpected medication transfer in captive cheetahs. mSystems (2020) 5 e00635-19. PMID: 32156796 29. Wang M, Carver JJ, Phelan VV, Sanchez LM, Garg N, Peng Y, Nguyen DD, Watrous J, Kapono CA, Luzzatto-Knaan T. Sharing and community curation of mass spectrometry data with Global Natural Products Social Molecular Networking. Nat. Biotechnol (2016) 34 828-837. PMID: 27504778 30. Carvalho BS, Irizarry RA. A framework for oligonucleotide microarray preprocessing. Bioinformatics (2010) 26 2363-2367. PMID: 20688976 31. Zhang X, Smits AH, van Tilburg GB, Ovaa H, Huber W, Vermeulen M. Proteome-wide identification of ubiquitin interactions using UbIA-MS. Nat. Protoc (2018) 13 530-550. PMID: 29446774 32. Finak G, McDavid A, Yajima M, Deng J, Gersuk V, Shalek AK, Slichter CK, Miller HW, McElrath MJ, Prlic M. MAST: a flexible statistical framework for assessing transcriptional changes and characterizing heterogeneity in single-cell RNA sequencing data. Genome Biol (2015) 16 278. PMID: 26653891 33. Butler A, Hoffman P, Smibert P, Papalexi E, Satija R. Integrating single-cell transcriptomic data across different conditions, technologies, and species. Nat. Biotechnol (2018) 36 411-420. PMID: 29608179 34. Smith TR, Tang X, Maricic I, Garcia Z, Fanchiang S, Kumar V. Dendritic cells use endocytic pathway for cross-priming class Ib MHC-restricted CD8αα. J. Immunol (2009) 182 6959-6968. PMID: 19454693 35. Coombes JL, Powrie F. Dendritic cells in intestinal immune regulation. Nat. Rev. Immunol (2008) 8 435-446. PMID: 18500229 36. Stagg AJ. Intestinal dendritic cells in health and gut inflammation. Front. Immunol (2018) 9 2883. PMID: 30574151 37. Gross M, Salame TM, Jung S. Guardians of the gut—murine intestinal macrophages and dendritic cells. Front. Immunol (2015) 6 254. PMID: 26082775 38. Gutiérrez-Martínez E, Planès R, Anselmi G, Reynolds M, Menezes S, Adiko AC, Saveanu L, Guermonprez P. Cross-presentation of cell-associated antigens by MHC class I in dendritic cell subsets. Front. Immunol (2015) 6 363. PMID: 26236315 39. Ng SL, Teo YJ, Setiagani YA, Karjalainen K, Ruedl C. Type 1 conventional CD103. Front. Immunol (2018) 9 3043. PMID: 30622538 40. Joeris T, Gomez-Casado C, Holmkvist P, Tavernier SJ, Silva-Sanchez A, Klotz L, Randall TD, Mowat AM, Kotarsky K, Malissen B, Agace WW. Intestinal cDC1 drive cross-tolerance to epithelial-derived antigen via induction of FoxP3. Sci. Immunol (2021) 6 41. Grajales-Reyes GE, Iwata A, Albring J, Wu X, Tussiwand R, Kc W, Kretzer NM, Briseño CG, Durai V, Bagadia P. Batf3 maintains autoactivation of Irf8 for commitment of a CD8α. Nat. Immunol (2015) 16 708-717. PMID: 26054719 42. Edelson BT, Kc W, Juang R, Kohyama M, Benoit LA, Klekotka PA, Moon C, Albring JC, Ise W, Michael DG. Peripheral CD103. J. Exp. Med (2010) 207 823-836. PMID: 20351058 43. Sun T, Rojas OL, Li C, Ward LA, Philpott DJ, Gommerman JL. Intestinal Batf3-dependent dendritic cells are required for optimal antiviral T-cell responses in adult and neonatal mice. Mucosal Immunol (2017) 10 775-788. PMID: 27600308 44. Hasegawa H, Matsumoto T. Mechanisms of tolerance induction by dendritic cells in vivo. Front. Immunol (2018) 9 350. PMID: 29535726 45. Esterházy D, Loschko J, London M, Jove V, Oliveira TY, Mucida D. Classical dendritic cells are required for dietary antigen-mediated induction of peripheral T. Nat. Immunol (2016) 17 545-555. PMID: 27019226 46. Hildner K, Edelson BT, Purtha WE, Diamond M, Matsushita H, Kohyama M, Calderon B, Schraml BU, Unanue ER, Diamond MS. Batf3 deficiency reveals a critical role for CD8α. Science (2008) 322 1097-1100. PMID: 19008445 47. Neumann PA, Koch S, Hilgarth RS, Perez-Chanona E, Denning P, Jobin C, Nusrat A. Gut commensal bacteria and regional Wnt gene expression in the proximal versus distal colon. Am. J. Pathol (2014) 184 592-599. PMID: 24418259 48. Takeuchi K, Bjarnason I, Laftah AH, Latunde-Dada GO, Simpson RJ, McKie AT. Expression of iron absorption genes in mouse large intestine. Scand. J. Gastroenterol (2005) 40 169-177. PMID: 15764147 49. Van der Sluis M, De Koning BA, De Bruijn AC, Velcich A, Meijerink JP, Van Goudoever JB, Büller HA, Dekker J, Van Seuningen I, Renes IB, Einerhand AW. Muc2-deficient mice spontaneously develop colitis, indicating that MUC2 is critical for colonic protection. Gastroenterology (2006) 131 117-129. PMID: 16831596 50. Shan M, Gentile M, Yeiser JR, Walland AC, Bornstein VU, Chen K, He B, Cassis L, Bigas A, Cols M. Mucus enhances gut homeostasis and oral tolerance by delivering immunoregulatory signals. Science (2013) 342 447-453. PMID: 24072822 51. Natividad JM, Verdu EF. Modulation of intestinal barrier by intestinal microbiota: pathological and therapeutic implications. Pharmacol. Res (2013) 69 42-51. PMID: 23089410 52. Zuo T, Ng SC. The gut microbiota in the pathogenesis and therapeutics of inflammatory bowel disease. Front. Microbiol (2018) 9 2247. PMID: 30319571 53. Morton JT, Marotz C, Washburne A, Silverman J, Zaramela LS, Edlund A, Zengler K, Knight R. Establishing microbial composition measurement standards with reference frames. Nat. Commun (2019) 10 2719. PMID: 31222023 54. Labarta-Bajo L, Gramalla-Schmitz A, Gerner RR, Kazane KR, Humphrey G, Schwartz T, Sanders K, Swafford A, Knight R, Raffatellu M, Zúňiga EI. CD8 T cells drive anorexia, dysbiosis, and blooms of a commensal with immunosuppressive potential after viral infection. Proc. Natl. Acad. Sci. USA (2020) 117 24998-25007. PMID: 32958643 55. Xu Y, Wang N, Tan HY, Li S, Zhang C, Feng Y. Function of Akkermansia muciniphila in obesity: interactions with lipid metabolism, immune response and gut systems. Front. Microbiol (2020) 11 219. PMID: 32153527 56. Hänninen A, Toivonen R, Pöysti S, Belzer C, Plovier H, Ouwerkerk JP, Emani R, Cani PD, De Vos WM. Akkermansia muciniphila induces gut microbiota remodelling and controls islet autoimmunity in NOD mice. Gut (2018) 67 1445-1453. PMID: 29269438 57. Quinn RA, Melnik AV, Vrbanac A, Fu T, Patras KA, Christy MP, Bodai Z, Belda-Ferre P, Tripathi A, Chung LK. Global chemical effects of the microbiome include new bileacid conjugations. Nature (2020) 579 123-129. PMID: 32103176 58. Leibson PJ. Cytotoxic lymphocyte recognition of HLA-E: utilizing a nonclassical window to peer into classical MHC. Immunity (1998) 9 289-294. PMID: 9768748 59. Zeng L, Sullivan LC, Vivian JP, Walpole NG, Harpur CM, Rossjohn J, Clements CS, Brooks AG. A structural basis for antigen presentation by the MHC class Ib molecule, Qa-1b. J. Immunol (2012) 188 302-310. PMID: 22131332 60. Peters LA, Perrigoue J, Mortha A, Iuga A, Song WM, Neiman EM, Llewellyn SR, Di Narzo A, Kidd BA, Telesco SE. A functional genomics predictive network model identifies regulators of inflammatory bowel disease. Nat. Genet (2017) 49 1437-1449. PMID: 28892060 61. Arijs I, De Hertogh G, Lemaire K, Quintens R, Van Lommel L, Van Steen K, Leemans P, Cleynen I, Van Assche G, Vermeire S. Mucosal gene expression of antimicrobial peptides in inflammatory bowel disease before and after first infliximab treatment. PLoS One (2009) 4 e7984. PMID: 19956723 62. Arijs I, De Hertogh G, Lemmens B, Van Lommel L, de Bruyn M, Vanhove W, Cleynen I, Machiels K, Ferrante M, Schuit F. Effect of vedolizumab (anti-α4β7-integrin) therapy on histological healing and mucosal gene expression in patients with UC. Gut (2018) 67 43-52. PMID: 27802155 63. Pierre N, Salée C, Massot C, Blètard N, Mazzucchelli G, Smargiasso N, Morsa D, Baiwir D, De Pauw E, Reenaers C. Proteomics highlights common and distinct pathophysiological processes associated with ileal and colonic ulcers in Crohn’s disease. J. Crohn’s Colitis (2020) 14 205-215. PMID: 31282946 64. Kim HJ, Cantor H. Regulation of self-tolerance by Qa-1-restricted CD8+ regulatory T cells. Semin. Immunol (2011) 23 446-452. PMID: 22136694 65. Varthaman A, Khallou-Laschet J, Clement M, Fornasa G, Kim HJ, Gaston AT, Dussiot M, Caligiuri G, Herbelin A, Kaveri S. Control of T cell reactivation by regulatory Qa-1-restricted CD8. J. Immunol (2010) 184 6585-6591. PMID: 20488793 66. Saligrama N, Zhao F, Sikora MJ, Serratelli WS, Fernandes RA, Louis DM, Yao W, Ji X, Idoyaga J, Mahajan VB. Opposing T cell responses in experimental autoimmune encephalomyelitis. Nature (2019) 572 481-487. PMID: 31391585 67. Yao Y, Han W, Liang J, Ji J, Wang J, Cantor H, Lu L. Glatiramer acetate ameliorates inflammatory bowel disease in mice through the induction of Qa-1-restricted CD8. Eur. J. Immunol (2013) 43 125-136. PMID: 23002042 68. Muzaki AR, Tetlak P, Sheng J, Loh SC, Setiagani YA, Poidinger M, Zolezzi F, Karjalainen K, Ruedl C. Intestinal CD103. Mucosal Immunol (2016) 9 336-351. PMID: 26174764 69. Merad M, Sathe P, Helft J, Miller J, Mortha A. The dendritic cell lineage: ontogeny and function of dendritic cells and their subsets in the steady state and the inflamed setting. Annu. Rev. Immunol (2013) 31 563-604. PMID: 23516985 70. Murphy TL, Grajales-Reyes GE, Wu X, Tussiwand R, Briseño CG, Iwata A, Kretzer NM, Durai V, Murphy KM. Transcriptional control of dendritic cell development. Annu. Rev. Immunol (2016) 34 93-119. PMID: 26735697 71. Katz JP, Perreault N, Goldstein BG, Lee CS, Labosky PA, Yang VW, Kaestner KH. The zinc-finger transcription factor Klf4 is required for terminal differentiation of goblet cells in the colon. Development (2002) 129 2619-2628. PMID: 12015290 72. Gersemann M, Becker S, Kübler I, Koslowski M, Wang G, Herrlinger KR, Griger J, Fritz P, Fellermann K, Schwab M. Differences in goblet cell differentiation between Crohn’s disease and ulcerative colitis. Differentiation (2009) 77 84-94. PMID: 19281767 73. Yu T, Chen X, Zhang W, Li J, Xu R, Wang TC, Ai W, Liu C. Krüppel-like factor 4 regulates intestinal epithelial cell morphology and polarity. PLoS One (2012) 7 e32492. PMID: 22384261 74. Yang VW, Liu Y, Kim J, Shroyer KR, Bialkowska AB. Increased genetic instability and accelerated progression of colitis-associated colorectal cancer through intestinal epithelium-specific deletion of Klf4. Mol. Cancer Res (2019) 17 165-176. PMID: 30108164 75. McDole JR, Wheeler LW, McDonald KG, Wang B, Konjufca V, Knoop KA, Newberry RD, Miller MJ. Goblet cells deliver luminal antigen to CD103. Nature (2012) 483 345-349. PMID: 22422267 76. Shimokawa C, Kato T, Takeuchi T, Ohshima N, Furuki T, Ohtsu Y, Suzue K, Imai T, Obi S, Olia A. CD8. Nat. Commun (2020) 11 1922. PMID: 32321922 77. Naito Y, Uchiyama K, Takagi T. A next-generation beneficial microbe: Akkermansia muciniphila. J. Clin. Biochem. Nutr (2018) 63 33-35. PMID: 30087541 78. Cani PD, de Vos WM. Next-generation beneficial microbes: the case of Akkermansia muciniphila. Front. Microbiol (2017) 8 1765. PMID: 29018410 79. Earley H, Lennon G, Balfe Á, Coffey JC, Winter DC, O’Connell PR. The abundance of Akkermansia muciniphila and its relationship with sulphated colonic mucins in health and ulcerative colitis. Sci. Rep (2019) 9 15683. PMID: 31666581 80. Bushman FD, Conrad M, Ren Y, Zhao C, Gu C, Petucci C, Kim MS, Abbas A, Downes KJ, Devas N. Multi-omic analysis of the interaction between Clostridioides difficile infection and pediatric inflammatory bowel disease. Cell Host Microbe (2020) 28 422-433. PMID: 32822584 81. Atreya R, Neurath MF. Involvement of IL-6 in the pathogenesis of inflammatory bowel disease and colon cancer. Clin. Rev. Allergy Immunol (2005) 28 187-196. PMID: 16129903 82. Bernardo D, Vallejo-Díez S, Mann ER, Al-Hassi HO, Martínez-Abad B, Montalvillo E, Tee CT, Murugananthan AU, Núñez H, Peake ST. IL-6 promotes immune responses in human ulcerative colitis and induces a skin-homing phenotype in the dendritic cells and Tcells they stimulate. Eur. J. Immunol (2012) 42 1337-1353. PMID: 22539302
--- title: Spatial variation of the gut microbiome in response to long-term metformin treatment in high-fat diet-induced type 2 diabetes mouse model of both sexes authors: - Laila Silamiķele - Rihards Saksis - Ivars Silamiķelis - Patrīcija Paulīne Kotoviča - Monta Brīvība - Ineta Kalniņa - Zane Kalniņa - Dāvids Fridmanis - Jānis Kloviņš journal: Gut Microbes year: 2023 pmcid: PMC10026874 doi: 10.1080/19490976.2023.2188663 license: CC BY 4.0 --- # Spatial variation of the gut microbiome in response to long-term metformin treatment in high-fat diet-induced type 2 diabetes mouse model of both sexes ## ABSTRACT Antidiabetic drug metformin alters the gut microbiome composition in the context of type 2 diabetes and other diseases; however, its effects have been mainly studied using fecal samples, which offer limited information about the intestinal site-specific effects of this drug. Our study aimed to characterize the spatial variation of the gut microbiome in response to metformin treatment by using a high-fat diet-induced type 2 diabetes mouse model of both sexes. Four intestinal parts, each at the luminal and mucosal layer level, were analyzed in this study by performing 16S rRNA sequencing covering six variable regions (V1-V6) of the gene and thus allowing to obtain in-depth information about the microbiome composition. We identified significant differences in gut microbiome diversity in each of the intestinal parts regarding the alpha and beta diversities. Metformin treatment altered the abundance of different genera in all studied intestinal sites, with the most pronounced effect in the small intestine, where Lactococcus increased remarkably. The abundance of Lactobacillus was substantially lower in male mice compared to female mice in all locations, in addition to an enrichment of opportunistic pathogens. Diet type and intestinal layer had significant effects on microbiome composition at each of the sites studied. We observed a different effect of metformin treatment on the analyzed subsets, indicating the multiple dimensions of metformin’s effect on the gut microbiome. ## Introduction Despite being widely studied, metformin is an antidiabetic drug with a still debated molecular action mechanism. Many studies have shown that its action is at least partially effectuated via the gut microbiome.1–3⁠ Gut microbiome composition and density vary along the gastrointestinal tract longitudinally and cross-sectionally.⁠4 Components contributing to this diversity include nutrient availability, chemical gradients, oxygen levels, mucosal structure, the diameter of the lumen, and cellular composition.5⁠⁠⁠ The gut microbiome is dominated by four phyla – Bacteroidetes (Gram-negative, anaerobic), Firmicutes (Gram-positive, anaerobic or obligate or facultatively aerobic), Actinobacteria (Gram-positive, anaerobic), and Proteobacteria (Gram-negative, aerobic or facultative anaerobic) in both mice and humans.6,7 The functional niche for each of these phyla is different as Bacteroidetes mainly produces acetate and propionate, but Firmicutes – butyrate.⁠⁠8 Functional differences between distinct regions of the gut correspond to the distribution of the microbial genera with respect to both identity and abundance. Studies in mice, piglets, and humans have shown that the proximal part of the small intestine – duodenum-jejunum, which is characterized by faster transit and facilitation of simple sugar and amino acid metabolism, is dominated by facultative anaerobes Lactobacillus, Proteobacteria, and obligate anaerobes Bacteroides, distal small intestine – ileum by Fusobacterium and Escherichia and the cecum and colon where the passage is slower and metabolism advantages fermentation of complex polysaccharides arising from undigested fibers or host mucus is prevailed by saccharolytic Bacteroidales, Clostridiales, and Prevotella.4,9,10⁠⁠⁠⁠ In healthy mice, Lactobacillaceae dominates the stomach and small intestine, while in the large intestine, the most abundant families are Bacteroidaceae, Prevotellaceae, Rikenellaceae, Lachnospiraceae, and Ruminococcaceae.11⁠ In the cecum of high-fat diet (HFD)-fed mice, the higher abundance of Lachnospiraceae, Blautia, and Oscillibacter, among others, has been shown; in addition, the colon and feces of HFD-fed mice are enriched in Bacteroides and Proteobacteria compared to control diet (CD)-fed ones.12,13⁠ Microbial communities differ on the transverse axis as well. Bacteria associated with the gut mucosa remain located at the same place; thus, these depend on the available substrate, whereas bacteria in the gut lumen can freely associate with various substrates. The mucosa is inhabited by aerotolerant taxa such as Proteobacteria and Actinobacteria, Clostridium, Lactobacillus, Enterococcus, and Akkermansia, while other bacteria predominate toward the lumen – Bacteroides, Bifidobacterium, Streptococcus, Enterobacteriaceae, and Ruminoccoccus among others.4,14⁠⁠ The mucosa-associated microbiota varies in different intestinal parts. Although fecal samples representing luminal content can provide information on global microbiota composition to a certain extent, research on mucosa-associated microbial communities can provide critical information on the interaction between gut microbiota and host.15⁠⁠ Absorption of metformin has been shown to vary between different intestinal sites, with the most prominent absorption occurring in the proximal small intestine.16⁠ Half of the administered dose reaches the distal small intestine, accumulating in the mucosal layer, and $30\%$ of the dose is eliminated in the feces.17⁠ Due to various absorption rates in different parts of the gastrointestinal tract, it can be expected that the metformin’s effect on microbiome composition and the abundance of different microorganisms is not uniform. Furthermore, metformin accumulates in mucosal tissue,18⁠ thus the exposition time and presumably the effect on resident bacteria differ in various sites. These differences necessitate an in-depth analysis of the effect of metformin treatment on the gut microbiome composition at various gastrointestinal locations and layers. The effect of metformin treatment has mainly been studied in cecal and fecal samples only, providing only limited information about the various aspects of the interaction between the gut microbiome and metformin. Short-term effects of metformin have been shown in different parts of the small intestine in the luminal layer in a recent study.19⁠ Our study adds information about the long-term effects of metformin therapy on different intestinal segments, not only in the luminal layer but also in the gut mucosa. Substantial advantages of our research include the analysis of both sexes and sequencing of six variable regions of the 16S rRNA gene, providing novel and trustworthy information on metformin’s effects. In addition, we employed a factorial animal experiment, including all the relevant control groups, allowing us to analyze the effect of various factors that potentially influence the interaction between metformin and the gut microbiome. Knowledge of the spatial variation of the gut microbiome in response to metformin treatment could potentially allow for more targeted microbiome modifications. In the future, this could alleviate the side effects experienced by metformin users, thereby increasing the applicability of the medication. ## Microbial composition analysis In total, 192 microbiome samples representing four different intestinal regions at the luminal and mucosal layers collected from 24 mice were sequenced. One of the samples was excluded from further analysis due to possible mislabeling. Mice representing three experimental units were included in each group representing each of the eight treatment arms described in this study (Figure 1). Figure 1.Experimental design of the study ($$n = 24$$) and intestinal sites studied. Abbreviations: HFD – high-fat diet; CD – control diet; M – male; F – female; Met+ – receiving metformin treatment for 10 weeks; Met- – not receiving metformin treatment. To summarize the microbiome composition in each analyzed site, we compiled a list of the top 20 location-specific genera, as shown in Figures 2 and 3. We observed variation in microbial composition both longitudinally and cross-sectionally. Both parts of the small intestine were dominated by Lactobacillus in HFD-fed mice, followed by Pseudomonas and Microbacterium in the mucosal layer and Pseudomonas and Lactococcus in the luminal layer (Figure 2a,b,e and f). Likewise, Lactobacillus and Pseudomonas prevailed in CD-fed mice, followed by Lactococcus in the mucosal layer (Figure 3a,b). In the luminal layer, Lactobacillus was followed by Pseudomonas and Microbacterium in the proximal part of the small intestine and by Lactococcus and Streptococcus – in the distal part (Figure 3e,f). The relative abundance of Lactococcus was increased in the distal small intestine of CD-fed male mice receiving metformin treatment. In contrast, in HFD-fed mice, this effect was more pronounced in female mice in both parts of the small intestine. Streptococcus was more abundant in the small intestine of female HFD-fed mice than in males, and its relative abundance was higher in the metformin-treated animals. Figure 2.Microbiome composition at different sites of high-fat diet-fed mice at the genus level – top 20 genera are shown. Mucosal layer: (a) proximal small intestine; (b) distal small intestine; (c) cecum; (d) colon. Luminal content layer: (e) proximal small intestine; (f) distal small intestine; (g) cecum; (h) colon. Samples representing mice receiving metformin treatment are highlighted. Red and blue bars under each plot indicate females and males, respectively. Figure 3.Microbiome composition at different sites of control diet-fed mice at the genus level – top 20 genera are shown. Mucosal layer: (a) proximal small intestine; (b) distal small intestine; (c) cecum; (d) colon. Luminal content layer: (e) proximal small intestine; (f) distal small intestine; (g) cecum; (h) colon. Samples representing mice receiving metformin treatment are highlighted. Red and blue bars under each plot indicate females and males, respectively. In HFD-fed mice, Blautia had the highest relative abundance in the cecum at both layers (Figures 2c,g). Mucispirillum, Lachnoclostridium, and Bacteroides were among the other top genera in the cecum. In CD-fed mice, Mucispirillum showed the highest relative abundance in the mucosa, while luminal content was enriched in Bacteroidales representatives (Figures 3c,g). Blautia and Lachnoclostridium had high relative abundance in both layers. Similar to the cecum, in the colon of HFD-fed mice top bacteria were Bacteroides in the mucosa, followed by Blautia and Lachnoclostridium, and Blautia in the luminal layer, followed by Bacteroides and other Bacteroidales members (Figures 2d,h). Similar results were observed in CD-fed mice (Figures 3d,h). ## Alpha and beta diversity analysis Shannon diversity index analysis revealed significant differences in alpha diversity between all the studied intestinal segments, with the most pronounced differences present between the distal part of the small intestine and two other locations, cecum, $H = 55.90$, p-value<0.001 and large intestine, $H = 55.67$, p-value<0.001 (Figure 4). Similar results were observed for Pielou’s evenness and Faith’s phylogenetic diversity (data not shown). There were no significant differences in alpha diversity between intestinal layers for all the metrics analyzed. When the effect of metformin treatment on alpha diversity in each intestinal site was evaluated, we did not observe any significant differences after adjustment for multiple testing for all the metrics analyzed. Figure 4.Microbiome alpha diversity analysis expressed as Shannon index: (a) between intestinal parts; (b) between intestinal layers. Microbiome alpha diversity analysis in response to metformin treatment expressed as Shannon index, $$n = 12$$ per group: (c) between intestinal parts; (d) between intestinal layers. PSI – proximal small intestine; DSI – distal small intestine; CEC – cecum; COL – colon; M – mucosa; L – lumen. adjusted $p \leq 0.05$; * adjusted $p \leq 0.01.$ Beta diversity was evaluated by ordination analysis. Biplots using PCA were generated, including each of the studied groups based on the diet type and metformin treatment status (Figures 5 and 6). When taken together, samples representing both parts of the small intestine clustered separately from the samples of the cecum and colon (Figure 5). Sphingobium, Sarcina, Propionibacterium, Pseudomonas, and Microbacterium representatives were the principal microbial identifiers of the proximal small intestine. Lactococcus, Lactobacillus, Streptococcus, Enterococcus, and Staphylococcus were the main drivers of the distal small intestine. In turn, cecum samples were identified by Mucispirillum, Anaerotruncus, Blautia, Ruminiclostridium, and Bacteroides. Colon was characterized by Desulfovibrio, Parabacteroides, Eubacterium_coprostanoligenes_group members, Alistipes, and Bacteroidales_S24-7_group (now known as Muribaculaceae) representatives. Figure 5.Beta diversity of all samples taken together was estimated using principal components analysis on centered log-ratio transformed values. Intestinal part, diet, and metformin treatment status are indicated. PSI – proximal small intestine; DSI – distal small intestine; CEC – cecum; COL – colon. In the corresponding intestinal part, continuous and dashed ellipses represent Met+ and Met- subsets, respectively. Figure 6.Beta diversity in each of the groups in response to 10 weeks long metformin treatment in each of the intestinal sites, $$n = 6$$ per group: (a) mucosa; (b) lumen; (c) proximal small intestine; (d) distal small intestine; (e) cecum; (f) colon. A separate analysis of each intestinal part (Figure 6) revealed Chryseobacterium, Propionibacterium, Corynebacterium_1, and Sphingobium as the principal identifiers of metformin treatment in the proximal small intestine. The distal part was characterized by Lactococcus, Streptococcus, Enterococcus, Proteus, and Staphylococcus in CD-fed mice and Blautia, Ruminiclostridium_9, Sphingobium, Bacteroides, and Roseburia in HFD-fed mice. The main identifiers of metformin treatment in the cecum of HFD-fed mice were Eubacterium representatives, Mucispirillum, Ruminococcaceae_UCG-003, and Streptococcus and Lachnospiraceae_FCS020_group, Ruminiclostridium representatives, Lactococcus, and Blautia in CD-fed mice. At the same time, Sphingobium, Curvibacter, Microbacterium, Pseudomonas, and Mucispirillum were the strongest identifiers in the colon. ## Metformin treatment-mediated effects on the abundance of bacteria in different intestinal sites When samples were contrasted regarding metformin treatment status (Met+ vs. Met- groups, $$n = 12$$ per group), up to 41 different genera were significantly differentially abundant in any of the parts of the small intestine; 28 genera in the cecum and 31 in the colon; and up to 11 genera in each of the intestinal layers studied (Figure 7). Figure 7.Differentially abundant genera in response to metformin treatment in different intestinal parts and layers (expressed as LogFC), $$n = 12$$ per group: (a) mucosa; (b) lumen; (c) proximal small intestine; (d) distal small intestine; (e) cecum; (f) colon. Blue bars represent genera with increased abundance among metformin users, and red bars – with decreased abundance. Dots of the corresponding color indicate all the individual features assigned to the genus. Metformin treatment showed a varied effect in different sites of the intestine. The effect on the abundance of Bacteroidales_S24-7_group members was inversed in each layer – metformin increased the abundance of this genus in mucosa samples but decreased it in luminal content samples (Figure 7a,b). Ruminiclostridium increased in the mucosa and lumen, while Lachnoclostridium decreased in both layers. Several genera were significantly affected only in one of the layers. The abundance of Roseburia declined in luminal content samples. In contrast, the abundances of Micrococcus, Pseudomonas, and Methylophilus were significantly affected only in the mucosa. Analysis of each of the intestinal parts separately revealed that metformin had a stronger effect on the abundance of the bacteria in the small intestine. Bacteria with increased abundances in response to metformin treatment only in the proximal small intestine include Duganella, Chryseobacterium, Anaeroplasma, Undibacterium, Corynebacterium, Mucispirillum, and Methylophilus (Figure 7c). In turn, Eubacterium_halli_group members were increased in the proximal small intestine but decreased in the cecum. Genera augmented uniquely in the distal part of the small intestine, include Faecalibaculum and Tepidimonas, while Proteus was depleted (Figure 7d). The abundance of Roseburia and Micrococcus was increased in both parts of the small intestine, while Ochrobactrum was decreased in these parts. Pseudomonas was affected in opposite directions in each part of the small intestine, with the abundance of the genus decreasing in the proximal part and increasing in the distal part. The most pronounced effect of metformin on the increased abundance was detected for Lactococcus in the distal small intestine; it was also increased in the colon, though to a lesser extent. Blautia was enriched in the distal part of the small intestine and cecum but diminished in the colon samples. Ruminiclostridium increased in all sites except cecum, where it was one of the most depleted genera (Figure 7e). Several genera were altered in the cecum and colon in opposite directions. Methylobacterium and Ruminococcaceae_UCG-014 were enriched in both parts, while Bacteroides was decreased. In turn, Clostridiales_vadinBB60_group members were decreased in the cecum and increased in the colon. Butyricicoccus was significantly increased uniquely in the colon, but Gastranaerophilales was decreased (Figure 7f). Ruminiclostridium_5 was the most reduced genus in response to metformin treatment in the colon, but it was increased in the proximal small intestine. The abundance of Parabacteroides was decreased in both parts of the small intestine and colon but was not affected in the cecum. A similar pattern was detected for Lachnospiraceae_UCG-008, with the only difference being that it was augmented in the distal part of the small intestine. ## Sex-related differences in the abundance of microbiome members in the intestinal sites studied The effect of sex was evaluated by contrasting the samples from male mice of all experimental groups with corresponding samples from female mice at each intestinal site separately, $$n = 12$$ per group. In total, 11 genera in mucosal samples and 8 genera in luminal content samples were differentially abundant between males and females. When each of the intestinal parts was analyzed separately, 51 genera in the proximal small intestine; 40 genera in the distal small intestine; 28 genera in the cecum; and 30 genera in the colon were significantly differentially abundant between sexes (Figure 8). Figure 8.Differentially abundant genera between sexes in different intestinal parts and layers (expressed as LogFC), $$n = 12$$ per group: (a) mucosa; (b) lumen; (c) proximal small intestine; (d) distal small intestine; (e) cecum; (f) colon. Blue bars represent genera with increased abundance among males, and red bars – with decreased abundance. Dots of the corresponding color indicate all the individual features assigned to the genus. Lactobacillus showed the most pronounced differences between sexes, with being decreased in males both in mucosal and luminal content samples (Figures 8a,b). Members of the Bacteroidales_S24-7_group were also depleted in both layers in males compared to females. In contrast, Proteus was increased in males in both mucosa and lumen, while Staphylococcus and Ruminococcaceae_UCG-003 were increased, and Ruminiclostridium_5 was decreased only in the lumen. The abundance of Lactobacillus was strongly reduced in all of the intestinal parts of males (distinct features in the cecum). Among the genera with significantly different abundance only in the proximal small intestine, Roseburia was enriched, while Chryseobacterium, Undibacterium, and Corynebacterium were depleted in males relative to females (Figure 8c). [ Eubacterium]_hallii_group was affected by the sex in the small intestine – in males, it was decreased in the proximal part but increased in the distal part of the small intestine; however, Enterorhabdus was enriched in both parts. Lachnoclostridium and Ruminiclostridium_5 were increased in the distal part of the small intestine in males compared to females (Figure 8d). Meanwhile, the abundance of Proteus was increased in both the distal small intestine and colon. The abundance of Anaerotruncus was markedly decreased in the cecum of males (Figure 8e). Ruminococcaceae_UCG-003 was increased in males only in the colon. Parabacteroides, Staphylococcus, and Prevotellaceae_UCG-001 representatives were enriched in the cecum and colon in males compared to females (Figure 8e,f). ## Diet-induced effects on the abundance of intestinal microbiome representatives To assess dietary effects, samples from all HFD-fed mice were compared with corresponding samples from CD-fed mice at each intestinal site separately, $$n = 12$$ per group. Diet significantly affected the abundance of 20 genera in the mucosa, 22 genera in the lumen, 55 and 41 genera in the proximal and distal small intestine, respectively, 36 genera in the cecum, and 42 genera in the colon (Figure 9). Figure 9.Differentially abundant genera between high-fat diet-fed and control diet-fed mice in different intestinal parts and layers (expressed as LogFC), $$n = 12$$ per group: (a) mucosa; (b) lumen; (c) proximal small intestine; (d) distal small intestine; (e) cecum; (f) colon. Blue bars represent genera with increased abundance among HFD-fed mice, and red bars – with decreased abundance. Dots of the corresponding color indicate all the individual features assigned to the genus. The abundance of Lactobacillus increased in HFD-fed mice in both studied layers LogFC = 2.79 ± 0.48, FDR<0.001 and LogFC = 2.44 ± 0.53, FDR = 0.002 for mucosa and lumen, respectively (Figures 9a,b). In contrast, Ruminiclostridium_5, Mollicutes_RF9, Lachnoclostridium, Marvinbryantia, and Clostridiales_vadinBB60_group members were significantly lowered in both layers in response to HFD feeding. Ralstonia and Staphylococcus were decreased in HFD-fed mice solely in mucosa samples. Parabacteroides and Ruminiclostridium_9 were significantly increased in the lumen of HFD-fed mice exclusively. Uniquely to the proximal small intestine, the abundance of Paucimonas and Oscillibacter increased, while Undibacterium decreased in HFD-fed mice (Figure 9c). Genera, which significantly decreased in both parts of the small intestine in response to HFD feeding, include Micrococcus, Enterorhabdus, and Candidatus_Saccharimonas (Figures 9c,d). Staphylococcus was depleted in the cecum of HFD-fed mice (Figure 9e). In turn, Eubacterium_oxidoreducens_group members were increased only in the colon (Figure 9F). Mollicutes_RF9 was decreased in both the cecum and colon of HFD-fed mice, while Ruminococcaceae_UCG-003 was increased in these parts. In HFD-fed mice, representatives of two genera, Ruminiclostridium_5 and Clostridiales_vadin_BB60_group, were decreased in all intestinal parts compared to CD-fed mice. As described above, these genera were also depleted in both intestinal layers of HFD-fed mice. ## Differentially abundant bacteria between the mucosa and the lumen in all intestinal parts The luminal and mucosal samples from all experimental groups were contrasted in each of the intestinal parts to investigate layer-related differences in microbiome composition, $$n = 12$$ per group. A total of 41 genera in the proximal small intestine, 48 – in the distal small intestine, 32 – in the cecum, and 38 – in the colon were significantly differentially abundant between the lumen and mucosa (Figure 10). Figure 10.Differentially abundant genera between lumen and mucosa in different intestinal parts (expressed as LogFC), $$n = 12$$ per group: (a) proximal small intestine; (b) distal small intestine; (c) cecum; (d) colon. Blue bars represent genera with increased abundance in the lumen, and red bars – with decreased abundance. Dots of the corresponding color indicate all the individual features assigned to the genus. Differential abundance analysis between mucosa and lumen layers in each of the intestinal parts separately revealed substantial spatial variation of genera. Uniquely to the proximal small intestine, its lumen was depleted of Corynebacterium_1 and Micrococcus (Figure 10a). Genera specifically increased in the lumen of the distal part of the small intestine include Erysipelatoclostridium, Intestinibacter, and Marvinbryantia. Anaerotruncus and Lachnospiraceae_UCG-006 also were enriched in the lumen of the distal small intestine (Figure 10b). Several genera were increased in the cecum and colon, including Parabacteroides, Prevotellaceae_UCG-001, Methylobacterium, and Faecalibacterium (Figures 10c,d). The abundance of Mucispirillum was lower in the lumen of the proximal small intestine and cecum. Similarly, Oscillibacter was reduced in the lumen of the proximal small intestine but increased in the cecum. Ruminiclostridium_5 and Enterococcus were enriched in the lumen of the distal small intestine and cecum. Genera with altered abundance, specifically in the distal small intestine and colon, include Microbacterium and Propionibacterium, with reduced abundance in the lumen of both parts, and Lactococcus, with increased abundance in the same sites. Curvibacter was depleted solely in the lumen of the colon. Roseburia and Ruminiclostridium_9 were altered in opposite directions in the same sites. Roseburia was decreased in both parts of the small intestine and increased in the cecum; in turn, Ruminiclostridium_9 was depleted in the small intestine and enriched in the cecum. Ruminiclostridium, together with Alistipes and Sphigobium, were oppositely affected, the former two being enriched in the lumen of the distal small intestine, cecum, and colon, while the latter was reduced in the same sites. Coprococcus_1 and Blautia were differentially abundant between layers in all intestinal parts. Blautia was increased in the lumen of the proximal small intestine and cecum but decreased in the distal part of the small intestine and colon. Coprococcus_1 was enriched in the lumen in all parts of the intestine except the colon, where the abundance of the genus was reduced. ## Interaction of metformin treatment with diet type, sex, and intestinal layer The genera with the most substantial differences in response to metformin treatment in each of the analyzed subsets are summarized in Figure 11. In total, we found 77 genera with a LogFC of at least 1 in any of the analysis subsets (Figure 11). Complete lists of statistically significantly altered genera in response to metformin treatment in each subset are shown in Supplementary Figures S1–S4. Metformin affected representatives of all the main phyla found in the intestine – Bacteroidetes, Firmicutes, Actinobacteria, Proteobacteria, and Verrucomicrobia. Figure 11.Summary of differentially abundant genera between Met+ and Met- mice in different subsets formed by various combinations of the levels of studied factors (expressed as LogFC), $$n = 3$$ per group. Only the genera with an absolute LogFC≥1 in at least one of the subsets are included, indicating the directions of changes (with an absolute LogFC>0.2), if detected, in all subsets. M – mucosa; L – lumen; PSI – proximal small intestine; DSI – distal small intestine; CEC – cecum; COL – colon. Blue triangle – an increase of the abundance among metformin users; red triangle – a decrease among metformin users. Members of Betaproteobacteria: Paucimonas and Alcaligenes were reduced only in the mucosal layer of both parts of the small intestine, while another representative Curvibacter was altered in opposite directions in different sites, including the cecum and colon in the mucosal layer but not in the luminal layer. Deltaproteobacteria member Desulfovibrio was affected by metformin almost in all studied sites. The most remarkable changes in abundance were observed in the proximal small intestine in both layers and the distal small intestine in the mucosal layer of HFD-fed mice. In addition, this genus showed marked sexual dimorphism in response to metformin treatment, with being increased in HFD-fed males in both parts of the small intestine and decreased in females in the mucosal layer and the proximal small intestine and cecum in the luminal layer. In CD-fed mice, Desulfovibrio was reduced in both sexes, though it was more pronounced in females. Sex-related differences in the proximal small intestine were also observed for Pseudomonas in mice fed both diet types, with being reduced in males and increased in females. The abundance of Pseudomonas was altered only in the mucosal layer, except for an increase in the lumen of the distal small intestine of HFD-fed males. The most pronounced changes in the abundance of Actinobacteria members were found in both layers of the proximal small intestine; however, significant changes in at least one genus were observed in all intestinal parts. Bacteroides was increased in both sexes of HFD-fed mice in the lumen of both parts of the small intestine and decreased in the colon. In turn, in the colon of CD-fed mice, Bacteroides was increased. Bacteroidales_S24–7 group (Muribaculaceae) was altered in almost all studied sites. The strongest reduction in the abundance of the genus was found in the mucosa of the distal small intestine of CD-fed males. The abundance of Akkermansia was significantly affected exclusively in CD-fed male cecum in both layers. Mucispirillum was strongly decreased in the mucosa of the proximal and distal small intestine of HFD-fed females, while it was slightly increased in males. In contrast, CD-fed females had marked increases in the genus at the exact locations. In the luminal layer, Mucispirillum was depleted in the proximal small intestine of CD-fed males and the distal small intestine of HFD-fed females, whereas the genus was increased in the distal small intestine of males fed both diet types and in HFD-fed males in the cecum. A substantial interaction between metformin treatment and sex, diet type, and the intestinal layer was observed regarding the abundance pattern of Bacilli members. Lactobacillus, Enterococcus, Staphylococcus, and Streptococcus, opposite to Lactococcus, were markedly decreased in the lumen of the proximal small intestine of CD-fed males. In contrast, all genera together with Lactococcus were increased in HFD-fed males at the same site. Lactococcus was substantially increased in the distal small intestine of CD-fed mice of both sexes, while in HFD-fed males, it was reduced. Lactococcus was increased in HFD-fed males and CD-fed females in the lumen of the colon. In the mucosa, the abundance of Lactococcus was not affected by metformin in the proximal small intestine. However, it was increased in both sexes of HFD-fed mice and CD-fed males in the distal small intestine, whereas in females, it was reduced. The abundance of Lactobacillus was mainly reduced in response to metformin, though it was augmented in the mucosa of the proximal small intestine of HFD-fed males. Intestinimonas and Clostridiales_vadinBB60_group members were affected only in the cecum and colon in the mucosa layer. A similar pattern was observed in the lumen, except that Clostridiales_vadinBB60_group was strongly increased in CD-fed males in the proximal small intestine and HFD-fed females in the distal small intestine. Butyricicoccus was affected similarly, mainly being augmented only in the mucosa of the cecum of male mice and the colon of HFD-fed males. In the lumen of the cecum, HFD-fed male mice were enriched in Butyricicoccus, while in CD-fed males genus was reduced. Butyricicoccus was increased in HFD-fed females in the lumen of the colon but unchanged in other subsets. Other Clostridiaceae representatives were not affected in the cecum and colon in the luminal layer. Genera representing families Lachnospiraceae, Peptococcaceae, Peptostreptococcaceae, and Ruminococcaceae were affected more in the cecum and colon in both layers. In contrast to this observation, Eubacterium_coprostanoligenes_group members were more affected in both parts of the small intestine in both layers. In general, members of the families mentioned above were enriched in the mucosa of the proximal small intestine of HFD-fed males but decreased in HFD-fed females. Other features common to these genera (with a few exceptions) are a decrease in the lumen of the proximal small intestine and cecum of CD-fed mice; and enrichment in the distal small intestine and colon (predominantly males) of HFD-fed mice. Roseburia showed strong sex-related differences in the abundance changes in response to metformin treatment. The abundance of the genus was increased in the mucosa of the proximal small intestine of male mice fed both types of diet and decreased in females. Roseburia was enriched only in HFD-fed males in the cecum but unchanged in other subsets. Similarly, it was increased in the colon of HFD-fed males, and a decrease was also found in other subsets, CD-fed mice and HFD-fed females. In the lumen, Roseburia was increased in HFD-fed mice of both sexes in all intestinal parts (except a decrease in females in the distal small intestine) and decreased in CD-fed mice of both sexes in the cecum and colon and both parts of the small intestine in males only. ## Discussion The effect of metformin therapy on the composition and function of the gut microbiome has been demonstrated in previous studies;20–26⁠⁠ however, an in-depth analysis of spatial variation of the effects of long-term metformin treatment on the gastrointestinal tract has not yet been performed in vivo. Our study provides novel information on the effects of metformin on both the gastrointestinal mucosa and the luminal layers at four different sites of the gut. Furthermore, our animal experiment included both sexes of mice, thus enabling an additional dimension of analysis of metformin’s effects. Finally, the metformin dosage was calculated to correspond to the therapeutic dose in humans, which makes our study more clinically relevant than experiments in which metformin was administered at the supratherapeutic level. We observed significantly different alpha diversities between all the studied intestinal parts corresponding to anatomical and functional differences of the various sites. The cecum and colon microbiomes were more diverse than those of the small intestine, consistent with a previous report.27⁠ The relative abundance of Lactobacillus decreased along the intestinal tract toward the colon, as described before.9⁠⁠⁠ Beta diversity analysis showed a clear difference between microbial communities of the small intestine and the cecum and colon, similar to a previous study comparing different parts of the gastrointestinal tract of male mice.27⁠ Analysis revealed that metformin treatment has the most pronounced effect on the samples representing the proximal small intestine, followed by the distal small intestine. Cecum samples of all experimental groups clustered most closely, showing the least effect of any studied factors on this intestinal part and suggesting its relative stability. We have shown the effect of metformin treatment, sex, diet type, and intestinal layer on the spatial variation of the gut microbiome by analyzing each of these factors separately. Furthermore, we have investigated the effect of metformin treatment in each of the subsets formed by combinations of levels of the studied factors: intestinal layer, intestinal part, diet type, and sex, and detected substantial variation of metformin’s effects in each of these subsets. Metformin treatment had a more pronounced effect on the microbiome composition in both parts of the small intestine (indicated by higher absolute LogFC values). This finding confirms the hypothesis that the effect of metformin treatment is not uniform in the whole intestine but rather depends on the absorption characteristics of the medication. When all mice were contrasted based on metformin treatment status, several genera were increased uniquely in the proximal small intestine, where according to a previous study,16⁠ the absorption of metformin occurs most. Among these genera, many are aerobic bacteria – Duganella, Chryseobacterium, Undibacterium, Corynebacterium, Methylophilus, and the anaerobes Anaeroplasma and Mucispirillum. Mucispirillum is commensal in the microbiota of humans and various vertebrates. The only species of the genus, Mucispirillum schaedleri, is a core member of the laboratory mouse microbiota throughout the whole gastrointestinal tract. Mucispirillum schaedleri has not been widely identified in human studies due to its low abundance in fecal samples, but it is enriched in the intestinal mucosa.28⁠⁠ Genomic data have indicated that instead of degrading mucosal glycans such as mucin, M. schaedleri predominantly processes monosaccharides, oligopeptides, amino acids, glycerol, and short-chain fatty acids produced by other bacteria.29⁠⁠ Metformin treatment substantially increased the abundance of Lactococcus in the distal part of the small intestine. This is consistent with a previous study with a similar duration and route of metformin treatment but using only aged male mice, where an increase in Lactococcus was observed in fecal samples after treatment with metformin.23⁠ Lactic acid bacteria, including Lactococcus, produce lactate, a substrate for other members of the microbiota to convert it into butyrate.8⁠ We observed a subtle increase in another lactate producer, Lactobacillus, in the distal small intestine, whereas the abundance of the genus was reduced in the proximal small intestine and cecum and unchanged in the colon. Previous studies24,30⁠ have produced conflicting results regarding the effect of metformin on Lactobacillus.31⁠ This could be explained by using fecal samples that do not fully represent the small intestinal microbiota, differences in experimental designs, and species-specific effects not detected in the genus-level analysis.32⁠ This together with the intestinal part-unique effects of metformin treatment, further supports the proximal and the distal part of the small intestine as principal sites of metformin-mediated effects on the gut microbiome.⁠⁠⁠⁠ Our results show that metformin mainly targets genera belonging to Clostridia in all intestinal sites. Altered genera include the representatives of Lachnospiraceae (Blautia, Lachnoclostridium, Lachnospira, Marvinbryantia, Roseburia), Clostridiaceae (Butyricicoccus, Clostridium), Ruminococcaceae (Anaerotruncus, Faecalibacterium, Oscillibacter, Ruminococcus), and Eubacteriaceae (Eubacterium). This is consistent with a similar study in which the effect of short-term metformin treatment on the composition of the microbiome representing the intestinal lumen at different sites was assessed in male mice, showing a reduction of Clostridium and Lachnospiraceae members in response to metformin treatment.19⁠ We found an increase of Roseburia in both parts of the small intestine and depletion of the genus in the cecum and colon in the luminal layer. This could explain the previously described reduced abundance of Roseburia in the feces of HFD-fed mice after metformin treatment,23⁠ as feces mostly correspond to the samples of colonic contents. In addition, we observed a strong sex-specific effect on the spatial variation of the gut microbiome when all mice were contrasted based on sex, as the abundance of Lactobacillus was strongly reduced in all intestinal parts and layers of males. This finding is supported by a study in healthy humans, where an increase of Lactobacillales in female mucosa-associated microbiota was reported.33⁠⁠ Lactobacillus has been shown to be increased in diabetes patients and animals fed with a high-fat diet,2,13,34⁠ consistent with enrichment of the genus in the colon and both intestinal layers of HFD-fed mice compared to CD-fed mice in our study. However, we observed that Lactobacillus was decreased in the proximal small intestine of HFD-fed mice, showing the differences between the composition of the fecal microbiome and the microbiome of different intestinal parts. Previous studies have shown that Lactobacillus spp. reduce blood glucose levels in high-fat diet-induced diabetic male mice;35⁠⁠ however, these are positively correlated with blood glucose levels in European women with T2D,36⁠ and metabolites produced by Lactobacilli may contribute to the glucose modulation of metformin.37⁠⁠ Sex-related differences in Lactobacillus abundance that we observed might partially explain these discrepancies emphasizing the need to perform studies on individuals of both sexes to obtain as complete information as possible. Additionally, an increase in Anaerotruncus has been found in women with metabolic syndrome compared to men.38⁠ Our data support these sex-related differences, narrowing them down to the cecum. We also detected a higher abundance of opportunistic pathogens Proteus in both layers and Staphylococcus in the lumen of male mice compared to females. Proteus was increased in the distal small intestine, Staphylococcus in the cecum, and both genera in the colon of males. This suggests that the microbiota of male mice in this model contains more opportunistic pathogens than females, which could contribute to another layer of phenotypic sexual dimorphism reported in HFD-fed mice.39 The role of other factors not analyzed in this study, including sex hormones, stress, dietary fiber content affecting systemic estrogen levels, the interaction between sex and genotype, and circadian rhythms, also might be driving the sex-related differences in the gut microbiome.40–42⁠ In addition, macronutrient levels and diet composition, including the availability of short-chain fatty acids, contribute to the effect of diet.⁠⁠43⁠ Wang et al. compared microbiome composition between HFD-fed and low-fat diet-fed mice in the cecum and colon and, similar to our results, found a higher abundance of Lachnospiraceae_UCG-006 in the cecum of HFD-fed mice.13⁠ We, however, add that in both parts of the small intestine, Lachnospiraceae_UCG-006 is reduced in HFD-fed mice, indicating that the effect of HFD feeding on this genus is not uniform throughout the intestine. Furthermore, we did not observe significant differences in the abundance of Blautia in the cecum, which was reported to be increased in this intestinal part, although it was enriched in both parts of the small intestine in our data. ⁠ Analysis between the lumen and mucosa in each of the intestinal parts revealed that Intestinibacter, Erysipelatoclostridium, and Marvinbryantia are genera enriched in the lumen of the distal small intestine uniquely. All these genera have not been extensively studied; Intestinibacter has been suggested to be involved in mucus degradation;2⁠⁠ a representative of Erysipelatoclostridium (formerly named Clostridium ramosum) has been associated with diet-induced obesity in gnotobiotic mice;44⁠ Marvinbryantia has been described as fermenting glucose to acetate in the presence of high formate concentrations.45⁠ Enrichment of Prevotellaceae_UCG-001, Oscillibacter, and Parabacteroides in the lumen of the cecum and colon agrees with a study comparing lumen-associated and mucosa-associated microbiota.33⁠ ⁠We observed an interaction between metformin treatment and each of the levels of studied factors, showing the complex landscape of metformin’s effects on the gut microbiome. Many genera were strongly affected in only one of the sexes or specific intestinal sites. For example, this analysis allowed us to confirm the promoting effect of metformin on the abundance of Lactococcus in the distal small intestine while adding information that it is mainly confined to the mucosa and that the abundance of Lactococcus is reduced in the lumen of HFD-fed males. In addition, we found that the genus is increased in the proximal small intestine of male mice only in the lumen. This study has some limitations. First, although we evaluated the appropriate sample size before conducting the experiment, the study would have benefited from a larger sample size for each group studied. Second, due to the complexity of the Sidle algorithm, we had to balance the sampling depth and the number of samples still available after sampling, resulting in a lower resolution for taxa with small sequence counts to retain the statistical power. Finally, we observed considerable variability in gut microbiome composition between cages, so future studies should carefully consider the appropriate experimental unit for such studies. In conclusion, our study describes the spatial differences of the metformin’s effects on the gut microbiome in luminal and mucosal layers of the intestine using both sexes of mice. Our results have revealed that metformin mainly exerts its microbiome-modulating effects in the small intestine, increasing the abundance of Lactococcus among other bacteria, while the impact on the microbiome of the cecum and colon is less pronounced. The effect of metformin on microbiome composition depends on sex, diet type, and intestinal layer. We furthermore emphasize the importance of including animals of both sexes in studies investigating the effects of metformin on the gut microbiome. ## Ethics statement Animal procedures were reviewed and approved by the National animal welfare and ethics committee (Permit No. 91). ## Animals, study design, and procedures The animal experiment has been described previously.46⁠ In short, C57BL/6N mice of both sexes with SPF status were purchased from the University of Tartu Laboratory Animal Centre and adapted to the local animal facility for one week. All the animals were housed under SPF conditions, 23 ± 2 ºC, with $55\%$ humidity. The light cycle was 12:12 hours, with a light period from 7:00 am to 7:00 pm. Animals were housed in individually ventilated cages (Tecniplast) of up to three same-sex animals per cage on aspen bedding mixed with ALPHA-dri. Mice were fed HFD (rodent diet with 60 kcal% fat (D12492, Research Diets)) or CD (rodent diet with 10 kcal% fat (D12450J, Research Diets)) ad libitum and had free access to drinking water. At the initiation of the study, all mice were six weeks old. The total number of animals was 72, forming 24 experimental units – the cages with animals. For further analysis, each experimental unit was represented by one animal from the corresponding cage. The sample size was determined by the resource equation method, which is suitable for complex designs.47⁠ The study had a randomized block design and included eight experimental groups – HFD_M_Met-, HFD_F_Met-, HFD_M_Met+, HFD_F_Met+, CD_M_Met-, CD_F_Met-, CD_M_Met+, and CD_F_Met+, depending on T2D status induced by high-fat diet (HFD) or control diet (CD) feeding; sex (M or F for males and females respectively); and metformin treatment status (Met+ and Met- indicating exposure to metformin treatment) (Figure 1). Type 2 diabetes induction and biochemical parameters were described previously46⁠⁠ in detail. The duration of type 2 diabetes induction with HFD feeding was 20 weeks. Diagnostic criteria for T2D included increased body weight, increased fasting plasma glucose and insulin levels in HFD-fed mice compared to CD-fed mice, and a HOMA-IR index of at least 2. Metformin was then administered in the drinking water at a concentration calculated to correspond to 50 mg/kg body mass/day for 10 weeks while continuing to be fed either an HFD or CD. During the treatment period, all bottles, including those of the control group, were changed daily, and metformin was added freshly to the drinking water of the treatment groups. ## Sample collection All the animals were sacrificed by cervical dislocation. Luminal and mucosal microbiome samples representing four different intestinal segments – proximal small intestine (duodenum-jejunum), distal small intestine (ileum), cecum, and colon were collected (Figure 1). First, intestinal segments were rinsed separately with distilled water to collect luminal contents. Second, rinsed tissue samples were put in a tissue dish containing cold PBS and cut longitudinally. Third, the remaining lumen contents were removed by repeated rinsing in separate clean tissue dishes. Finally, mucosa samples containing microbiome were obtained by scraping the inner intestinal surface of the intestinal segment with a cell scraper and collected in sterile tubes. Samples were stored at 80ºC until further analysis. ## DNA isolation and 16S rRNA gene amplification Bacterial DNA was isolated from the collected samples using FastDNA Spin Kit for Soil (MP Biomedicals) according to the manufacturer’s instructions, and the concentration was measured using the Qubit 2.0 Fluorometer (Invitrogen). The concentration of DNA before library preparation was normalized to 5 ng/µl for every sample. The V1-V2, V3-V4, and V5-V6 hypervariable regions of the 16S rRNA gene were amplified by PCR using specific primers (Supplementary Table 1) tagged with Illumina sequencing adapters and sample-specific barcodes according to Illumina’s instructions. PCR products were analyzed by $1.2\%$ agarose gel electrophoresis and purified using NucleoMag (Macherey-Nagel) magnetic beads. Purified amplicons were pooled at equimolar concentrations, and sample indexes were added by additional PCR. The quality of libraries was assessed by Agilent High Sensitivity DNA Kit (Agilent Technologies) on the Agilent 2100 Bioanalyzer (Agilent Technologies). Samples were sequenced on Illumina MiSeq (Illumina) platform with MiSeq Reagent Kit V2 (500-cycles) (Illumina) according to the manufacturer’s instructions. ## Data analysis Data analysis began with evaluating sequencing data quality and read count distribution per sample group using the FastQC (v0.11.9)48⁠⁠ and MultiQC (v1.12)⁠49⁠ tools to identify possible sample level outliers and adapter contamination. Most of the analysis and diversity index calculation was performed using QIIME2 (v2022.2)⁠50⁠ microbiome analysis environment. First, we processed the regional amplicons for each individual, using the Cutadapt plugin51⁠ to trim the forward and reverse primers for each amplicon, specifying the allowed error rate of 0.1 and allowing for indels or deletions in bases when matching with the primers. In this step, reads that did not match the primers were discarded. The demultiplexed regional sequences were denoised with the DADA252⁠⁠ plugin to generate Amplicon Sequence Variants (ASVs). The trim lengths for DADA2 were selected for each region based on the base quality drop-off threshold from the visual inspection of the sample group level sequence quality box and whiskers plots to maximize the number of merged reads (Suplementary Table 1). Subsequently, all merged reads for each region were truncated to the length of 200 base pairs. The SILVA (v128) taxonomic database53⁠ was imported into QIIME 2 using the RESCRIPt⁠54⁠ plugin; the database was filtered to remove any sequence with more than 5 degenerate nucleotides. Regional database reads were extracted using the q2-feature-classifier plugin.55⁠ The regional database reads were aligned with the representative ASV sequences, allowing a mismatch of 2 nucleotides. Regional average relative abundances were solved through the Sidle implementation of the Short Multiple Reads Framework (SMURF) algorithm.56,57⁠ The phylogenetic tree was reconstructed by inserting consensus sequences for reconstructed amplicons into the SILVA (v128) backbone using the SEPP algorithm58⁠⁠ phylogenetic reference backbone while also inserting sequences that did not align with the SILVA taxonomy reference database. To discard low information or artifact sequences, the reconstructed ASV table was frequency filtered for features observed in at least three samples and with a taxonomic classification of genus level or higher. From the resulting table, a random feature subsample of 945 sequences per sample was made to normalize for the differences in library size, which was then used to calculate Shannon diversity, Faith’s phylogenetic diversity, and Pielou’s evenness alpha diversity indices. Statistically significant differences in the alpha diversity between analyzed groups were identified using the Wilcoxon Rank Sum test and Benjamini-Hochgberg’s procedure. False discovery rate (FDR) values<0.05 were considered statistically significant. Further analysis and visualization of results were performed in the RStudio (v2021.09.0) environment. Reconstructed and classified data was then imported into the phyloseq (v1.38.0)59⁠ environment. Sample level ordination was calculated on rarefied (945 sequences, seed = 43980) genus-level aggregated data, which was then transformed with the centered log ratio method and reduced with principal components analysis for each intestinal part and layer to evaluate the beta diversity. Taxonomic distribution bar plot graphs and ordination graphs were created with the microViz (v0.9.0.9009)60⁠ and Matplotlib (3.5.2)61⁠⁠ packages, while alpha diversity box plot graphs were created with the ggplot2 (v3.3.6)62⁠⁠ package. Finally, we performed a differential abundance test with the ANCOM-BC (v1.4.0)63⁠ package, including independent variables in the formula and excluding features not observed in at least $10\%$ of all samples. To see the full list of used R packages, consult Supplementary Table 2. Median log fold change values of differentially abundant taxa of the same genera were visualized as bar plots using python libraries Matplotlib (3.5.2) and pandas (1.4.3).64 Analysis of compositions of microbiomes with bias correction (ANCOM-BC) was performed using different levels of factors: metformin treatment status, sex, and diet type in each intestinal layer and part, $$n = 12$$ per group. To investigate the interaction between metformin treatment and the studied factors, analysis of compositions of microbiomes with bias correction (ANCOM-BC) was performed separately for each of the different combinations of factor levels, contrasting Met+ and Met- samples, $$n = 3$$ per group. Features representing the same genus were combined, and medians of the LogFC of abundances were plotted in each of the analyzed contrasts (Figures 7–10 and Supplementary Figures S1–S4). Individual dots were included in the plots to show the genera consisting of multiple features and the distribution of LogFC for each of the features.⁠ ## Disclosure statement No potential conflict of interest was reported by the authors. ## Supplementary material Supplemental data for this article can be accessed online at https://doi.org/$\frac{10.1080}{19490976.2023.2188663}$ ## References 1. Shin NR, Lee JC, Lee HY, Kim MS, Whon TW, Lee MS, Bae JW.. An increase in the Akkermansia spp. population induced by metformin treatment improves glucose homeostasis in diet-induced obese mice. Gut (2014.0) 63 727-23. PMID: 23804561 2. Forslund K, Hildebrand F, Nielsen T, Falony G, Le Chatelier E, Sunagawa S, Prifti E, Vieira-Silva S, Gudmundsdottir V, Pedersen H. Disentangling the effects of type 2 diabetes and metformin on the human gut microbiota. Nature (2015.0) 528 262-266. PMID: 26633628 3. Bryrup T, Thomsen CW, Kern T, Allin KH, Brandslund I, Jørgensen NR, Vestergaard H, Hansen T, Hansen TH, Pedersen O. Metformin-induced changes of the gut microbiota in healthy young men: results of a non-blinded, one-armed intervention study. Diabetologia (2019.0) 62 1024-1035. PMID: 30904939 4. Tropini C, Earle KA, Huang KC, Sonnenburg JL. The gut microbiome: connecting spatial organization to function. Cell Host Microbe (2017.0) 21 433-442. PMID: 28407481 5. Donaldson GP, Lee SM, Mazmanian SK. Gut biogeography of the bacterial microbiota. Nat Rev Microbiol 6. Belizário JE, Napolitano M. Human microbiomes and their roles in dysbiosis, common diseases, and novel therapeutic approaches. Front Microbiol (2015.0) 6 1-16. PMID: 25653648 7. Wang J, Lang T, Shen J, Dai J, Tian L, Wang X. Core gut bacteria analysis of healthy mice. Front Microbiol (2019.0) 10 1-14. PMID: 30728808 8. Thursby E, Juge N. Introduction to the human gut microbiota. Biochem J (2017.0) 474 1823-1836. PMID: 28512250 9. Gu S, Chen D, Zhang JN, Lv X, Wang K, Duan LP, Nie Y, Wu XL. Bacterial community mapping of the mouse gastrointestinal tract. PLoS One (2013.0) 8 10. Liu Y, Zheng Z, Yu L, Wu S, Sun L, Wu S, Xu Q, Cai S, Qin N, Bao W. Examination of the temporal and spatial dynamics of the gut microbiome in newborn piglets reveals distinct microbial communities in six intestinal segments. Sci Rep [Internet] (2019.0) 9 1-8. DOI: 10.1038/s41598-019-40235-z 11. Yamamoto Y, Nakanishi Y, Murakami S, Aw W, Tsukimi T, Nozu R, Ueno M, Hioki K, Nakahigashi K, Hirayama A. A metabolomic-based evaluation of the role of commensal microbiota throughout the gastrointestinal tract in mice. Microorganisms (2018.0) 6 101. PMID: 30274293 12. Singh RP, Halaka DA, Hayouka Z, Tirosh O. High-Fat diet induced alteration of mice microbiota and the functional ability to utilize fructooligosaccharide for ethanol production. Front Cell Infect Microbiol (2020.0) 10 1-11. PMID: 32083019 13. Wang B, Kong Q, Li X, Zhao J, Zhang H, Chen W, Wang G. A high-fat diet increases gut microbiota biodiversity and energy expenditure due to nutrient difference. Nutrients (2020.0) 12 1-20 14. Jandhyala SM, Talukdar R, Subramanyam C, Vuyyuru H, Sasikala M, Reddy DN. Role of the normal gut microbiota. World J Gastroenterol (2015.0) 21 8836-8847. PMID: 26269673 15. Kirchgessner A. The role of the microbiota and potential for dietary intervention in chronic fatigue syndrome [Internet]. Elsevier Inc (2016.0). DOI: 10.1016/B978-0-12-802304-4.00021-9 16. Bauer PV, Duca FA, Waise TMZ, Rasmussen BA, Abraham MA, Dranse HJ, Puri A, O’brien CA, Lam TKT. Metformin alters upper small intestinal microbiota that impact a glucose-SGLT1-Sensing glucoregulatory pathway. Cell Metab [Internet] (2018.0) 27 101-117.e5. DOI: 10.1016/j.cmet.2017.09.019 17. Buse JB, DeFronzo RA, Rosenstock J, Kim T, Burns C, Skare S, Baron A, Fineman M. The primary glucose-lowering effect of metformin resides in the gut, not the circulation: results from short-term pharmacokinetic and 12-week dose-ranging studies. Diabetes Care (2016.0) 39 198-205. PMID: 26285584 18. Bailey CJ, Wilcock C, Scarpello JHB. Metformin and the intestine. Diabetologia (2008.0) 51 1552-1553. PMID: 18528677 19. Bravard A, Gérard C, Defois C, Benoit B, Makki K, Meugnier E, Rainteau D, Rieusset J, Godet M, Vidal H. Metformin treatment for 8 days impacts multiple intestinal parameters in high-fat high-sucrose fed mice. Sci Rep [Internet] (2021.0) 11 1-12. DOI: 10.1038/s41598-021-95117-0 20. Wu H, Esteve E, Tremaroli V, Khan MT, Caesar R, Mannerås-Holm L, Ståhlman M, Olsson LM, Serino M, Planas-Fèlix M. Metformin alters the gut microbiome of individuals with treatment-naive type 2 diabetes, contributing to the therapeutic effects of the drug. Nat Med [Internet] (2017.0) 23 850-858. DOI: 10.1038/nm.4345 21. Lee H, Lee Y, Kim J, An J, Lee S, Kong H, Song Y, Lee CK, Kim K. Modulation of the gut microbiota by metformin improves metabolic profiles in aged obese mice. Gut Microbes [Internet] (2018.0) 9 155-165. DOI: 10.1080/19490976.2017.1405209 22. Dong TS, Chang HH, Hauer M, Lagishetty V, Katzka W, Rozengurt E, Jacobs JP, Eibl G. Metformin alters the duodenal microbiome and decreases the incidence of pancreatic ductal adenocarcinoma promoted by diet-induced obesity. Am J Physiol - Gastrointest Liver Physiol (2019.0) 317 G763-72. PMID: 31545922 23. Ahmadi S, Razazan A, Nagpal R, Jain S, Wang B, Mishra SP, Wang S, Justice J, Ding J, McClain DA. Metformin reduces aging-related leaky gut and improves cognitive function by beneficially modulating gut microbiome/goblet cell/mucin axis. J Gerontol Ser A Biol Sci Med Sci (2020.0) 75 E9-21. PMID: 32129462 24. Liu Z, Liao W, Zhang Z, Sun R, Luo Y, Chen Q, Li X, Lu R, Ying Y. Metformin affects gut microbiota composition and diversity associated with amelioration of dextran sulfate sodium-induced colitis in mice. Front Pharmacol (2021.0) 12 1-9 25. Jauvain M, Courtois S, Lehours P, Bessède E. Metformin modifies the gut microbiota of mice infected with helicobacter pylori. Pharmaceuticals (2021.0) 14 26. Li Q, He R, Zhang F, Zhang J, Lian S, Liu H. Combination of oligofructose and metformin alters the gut microbiota and improves metabolic profiles, contributing to the potentiated therapeutic effects on diet-induced obese Animals. Front Endocrinol (Lausanne) (2020.0) 10 1-14 27. Lkhagva E, Hong J, Hong W, Tang W, Lee S. The regional diversity of gut microbiome along the GI tract of male C57BL/6 mice. BMC Microbiol (2021.0) 21 1-13. PMID: 33386072 28. Herp S, Durai Raj AC, Salvado Silva M, Woelfel S, Stecher B. The human symbiont Mucispirillum schaedleri: causality in health and disease. Med Microbiol Immunol [Internet] (2021.0) 210 173-179. DOI: 10.1007/s00430-021-00702-9 29. Loy A, Pfann C, Steinberger M, Hanson B, Herp S, Brugiroux S, Gomes C, Boekschoten VM, Schwab C, Urich T. Lifestyle and horizontal gene transfer- mediated evolution of mucispirillum schaedleri, a core member of the murine gut microbiota. mSystems (2017.0) 2 e00116-71. PMID: 28951891 30. Bornstein S, Moschetta M, Kawano Y, Sacco A, Huynh D, Brooks D, Manier S, Fairfield H, Falank C, Roccaro AM. Metformin affects cortical bone mass and marrow adiposity in diet-induced obesity in male Mice. Endocrinology (2017.0) 158 3369-3385. PMID: 28977604 31. Zhang Q. Effects of metformin on the gut microbiota in obesity and type 2 diabetes mellitus. Diabetes Metab Syndr Obes (2020.0) 32. Gurung M, Li Z, You H, Rodrigues R, Jump DB, Morgun A, Shulzhenko N. Role of gut microbiota in type 2 diabetes pathophysiology. EBioMedicine [Internet] (2020.0) 51 102590. DOI: 10.1016/j.ebiom.2019.11.051 33. Borgo F, Garbossa S, Riva A, Severgnini M, Luigiano C, Benetti A, Pontiroli AE, Morace G, Borghi E. Body mass index and sex affect diverse microbial niches within the gut. Front Microbiol (2018.0) 9 1-12. PMID: 29403456 34. Sedighi M, Razavi S, Navab-Moghadam F, Khamseh ME, Alaei-Shahmiri F, Mehrtash A, Amirmozafari N. Comparison of gut microbiota in adult patients with type 2 diabetes and healthy individuals. Microb Pathog [Internet] (2017.0) 111 362-369. DOI: 10.1016/j.micpath.2017.08.038 35. Gulnaz A, Nadeem J, Han JH, Lew LC, Dong SJ, Park YH, Rather IA, Hor YY. Lactobacillus Sps in reducing the risk of diabetes in high-fat diet-induced diabetic mice by modulating the gut microbiome and inhibiting key digestive enzymes associated with diabetes. Biology (Basel) (2021.0) 36. Karlsson FH, Tremaroli V, Nookaew I, Bergström G, Behre CJ, Fagerberg B, Nielsen J, Bäckhed F. Gut metagenome in European women with normal, impaired and diabetic glucose control. Nature (2013.0) 498 99-103. PMID: 23719380 37. Rooj AK, Kimura Y, Buddington RK. Metabolites produced by probiotic Lactobacilli rapidly increase glucose uptake by Caco-2 cells. BMC Microbiol (2010.0) 10. PMID: 20074323 38. Santos-Marcos JA, Haro C, Vega-Rojas A, Alcala-Diaz JF, Molina-Abril H, Leon-Acuña A, Lopez-Moreno J, Landa BB, Tena-Sempere M, Perez-Martinez P. Sex differences in the gut microbiota as potential determinants of gender predisposition to disease. Mol Nutr Food Res (2019.0) 63 1-11 39. Casimiro I, Stull ND, Tersey SA, Mirmira RG. Phenotypic sexual dimorphism in response to dietary fat manipulation in C57BL/6J mice. J Diabetes Complications [Internet] (2021.0) 35 107795. DOI: 10.1016/j.jdiacomp.2020.107795 40. Valeri F, Endres K. Frontiers in Neuroendocrinology How biological sex of the host shapes its gut microbiota. Front Neuroendocrinol [Internet] (2021.0) 61. DOI: 10.1016/j.yfrne.2021.100912 41. Kim YS, Unno T, Kim BY, Park MS. Sex differences in gut microbiota. World J Mens Health (2020.0) 38 48-60. PMID: 30929328 42. Voigt RM, Forsyth CB, Green SJ, Engen PA. Circadian rhythm and the gut microbiome [Internet] (2016.0). DOI: 10.1016/bs.irn.2016.07.002 43. Leeming ER. Effect of diet on the gut microbiota: rethinking intervention duration. Nutrients (2019.0) 11 1-28 44. Woting A, Pfeiffer N, Loh G, Klaus S, Blaut M. Clostridium ramosum promotes high-fat diet-induced obesity in gnotobiotic mouse models. MBio (2014.0) 5 1-10 45. Wolin MJ, Miller TL, Collins MD, Lawson PA. Formate-Dependent growth and homoacetogenic fermentation by a bacterium from human feces: description of Bryantella formatexigens gen. nov., sp. nov. Appl Environ Microbiol (2003.0) 69 6321-6326. PMID: 14532100 46. Silamiķele L, Silamiķelis I, Ustinova M, Kalniņa Z, Elbere I, Petrovska R, Kalniņa I, Kloviņš J. Metformin strongly affects gut microbiome composition in high-fat diet-induced type 2 diabetes mouse model of both sexes. Front Endocrinol (Lausanne) (2021.0) 12 47. Festing MFW, Overend P, Cortina Borja M, Berdoy M. The design of animal experiments: reducing the use of animals in research through better experimental design (Laboratory Animal Handbooks) (2016.0) 48. Andrews S. FastQC: a quality control tool for high throughput sequence data. (2010.0) 49. Ewels P, Magnusson M, Lundin S, Käller M. MultiQC: summarize analysis results for multiple tools and samples in a single report. Bioinformatics (2016.0) 32 3047-3048. PMID: 27312411 50. Bolyen E, Rideout JR, Dillon MR, Bokulich NA, Abnet CC, Al-Ghalith GA, Alexander H, Alm EJ, Arumugam M, Asnicar F. Reproducible, interactive, scalable and extensible microbiome data science using QIIME 2. Nat Biotechnol (2019.0) 37 852-857. PMID: 31341288 51. Martin M. Cutadapt removes adapter sequences from high-throughput sequencing reads. EMBnetJournal (2011.0) 17 10-12 52. Callahan BJ, McMurdie PJ, Rosen MJ, Han AW, Johnson AJA, Holmes SP. DADA2: high resolution sample inference from Illumina amplicon data. Nat Methods (2016.0) 13 581-583. PMID: 27214047 53. Quast C, Pruesse E, Yilmaz P, Gerken J, Schweer T, Yarza P, Peplies J, Glöckner FO. The SILVA ribosomal RNA gene database project: improved data processing and web-based tools. Nucleic Acids Res (2013.0) 41 590-596 54. Robeson MS, O’rourke DR, Kaehler BD, Ziemski M, Dillon MR, Foster JT, Bokulich NA. Rescript: reproducible sequence taxonomy reference database management. (2021.0). DOI: 10.1371/journal.pcbi.1009581 55. Bokulich NA, Kaehler BD, Rideout JR, Dillon M, Bolyen E, Knight R, Huttley GA, Gregory Caporaso J. Optimizing taxonomic classification of marker-gene amplicon sequences with QIIME 2’s q2-feature-classifier plugin. Microbiome (2018.0) 6 1-17. PMID: 29291746 56. Debelius JW, Robeson M, Hugerth LW, Boulund F, Engstrand L. A comparison of approaches to scaffolding multiple regions along the 16S rRNA gene 1 for improved resolution 2. bioRxiv [Internet] (2021.0). DOI: 10.1101/2021.03.23.436606 57. Fuks G, Elgart M, Amir A, Zeisel A, Turnbaugh PJ, Soen Y, Shental N. Combining 16S rRNA gene variable regions enables high-resolution microbial community profiling. Microbiome (2018.0) 6 1-13. PMID: 29291746 58. Mirarab S, Nguyen NW. SEPP: sATé-enabled phylogenetic placement. Pac Symp Biocomput (2012.0) 247-258. PMID: 22174280 59. McMurdie PJ, Holmes S. Phyloseq: an R package for reproducible interactive analysis and graphics of microbiome census data. PLoS One (2013.0) 8 60. Barnett D, Arts I, Penders J. microViz: an R package for microbiome data visualization and statistics. J Open Source Softw (2021.0) 6 3201 61. Hunter JD. Matplotlib: a 2D graphics environment. Comput Sci Eng (2007.0) 9 90-95 62. Wickham H. Ggplot2: elegant graphics for data analysis (2016.0) 63. Lin H, Das Peddada S. Analysis of compositions of microbiomes with bias correction. Nat Commun [Internet] (2020.0) 11 1-11. DOI: 10.1038/s41467-020-17041-7 64. McKinney W. Data structures for statistical computing in python. Proc 9th Python Sci Conf (2010.0) 1 56-61
--- title: Pruning and thresholding approach for methylation risk scores in multi-ancestry populations authors: - Junyu Chen - Evan Gatev - Todd Everson - Karen N. Conneely - Nastassja Koen - Michael P. Epstein - Michael S. Kobor - Heather J. Zar - Dan J. Stein - Anke Hüls journal: Epigenetics year: 2023 pmcid: PMC10026878 doi: 10.1080/15592294.2023.2187172 license: CC BY 4.0 --- # Pruning and thresholding approach for methylation risk scores in multi-ancestry populations ## ABSTRACT Recent efforts have focused on developing methylation risk scores (MRS), a weighted sum of the individual’s DNA methylation (DNAm) values of pre-selected CpG sites. Most of the current MRS approaches that utilize Epigenome-wide association studies (EWAS) summary statistics only include genome-wide significant CpG sites and do not consider co-methylation. New methods that relax the p-value threshold to include more CpG sites and account for the inter-correlation of DNAm might improve the predictive performance of MRS. We paired informed co-methylation pruning with P-value thresholding to generate pruning and thresholding (P+T) MRS and evaluated its performance among multi-ancestry populations. Through simulation studies and real data analyses, we demonstrated that pruning provides an improvement over simple thresholding methods for prediction of phenotypes. We demonstrated that European-derived summary statistics can be used to develop P+T MRS among other populations such as African populations. However, the prediction accuracy of P+T MRS may differ across multi-ancestry population due to environmental/cultural/social differences. ## Introduction DNA methylation (DNAm), one of the most studied epigenetic mechanisms, regulates the mode of expression of DNA segments independent of alterations of their sequence by adding a methyl group at cytosine residues, hence contributing to variation in cellular phenotypes [1]. With current advances in and reduction of cost of array-based profiling technologies, increasing numbers of large-scale epigenome-wide association studies (EWAS) have been conducted to study DNAm in association with complex human diseases as well as environmental and social factors [2,3]. EWAS have thus far been successful in identifying dozens of cytosine guanine dinucleotide sites (CpGs) associated with various diseases and exposures, which could potentially be used for disease diagnosis and prediction, development of drug targets, and monitoring of drug response [3–8]. However, differential DNAm in individual CpGs often shows a weak prediction capacity and can only explain a small fraction of phenotype variance. Polyepigenetic approaches that aggregate information on differential DNAm from multiple CpGs might produce a more accurate biomarker for clinical usage [9,10]. A well-known polygenic approach for genotype data is polygenic risk scores (PRS), which are weighted sums of risk alleles of a pre-selected number of genetic variants [11]. Recently, many efforts have focused on transferring PRS approaches to DNA methylation data to construct methylation risk scores (MRS), which are defined as weighted sums of the individuals’ DNAm values of a pre-selected number of CpGs [10]. However, there are many methodological challenges in constructing DNA methylation risk scores [10,12,13]. One of the problems is that DNAm is influenced by ancestry, which captures genetic ancestry (differences in the genome related to ancestry) as well as social determinants of health such as racism and discrimination, socioeconomic status, and environmental effects [14]. Thus, ideally, when external weights are used for the calculation of MRS, these weights should be assessed in a population with the same ancestry as the study samples. However, current epigenetic literature remains limited by the lack of diversity, with most focusing on European populations [15], therefore making it difficult to identify appropriate weights for MRS for other populations. While it is well known that PRSs are not applicable across different ancestries [16,17], little is known about the performance of MRS across multi-ancestry populations. Currently, there are two popular approaches to construct MRS. The first one is to use penalized regression models such as Elastic Net and LASSO regularization [10,18,19], which usually requires individual-level DNAm data. When only summary-level statistics are available, individual CpGs that reached genome-wide significance in an external EWAS are selected and the beta-coefficients estimated from EWAS are used as weights to generate MRS [10]. However, research in PRS has shown that the optimal p-value threshold strongly depends on the data [20], and including a larger proportion of variants could potentially capture more of the phenotype variation [21]. Moreover, most MRS from the second approach do not consider DNA co-methylation, defined as proximal CpGs with correlated DNAm across individuals [22], which could potentially bias the generation of MRS. Shah et al 2015 proposed to remove redundant CpGs by keeping the most significant CpGs in co-methylation [23,24], however, a window to define DNA co-methylation needs to be pre-defined and the effect of accounting for DNA co-methylation was not evaluated. One of the most widely used PRS approaches to deal with single nucleotide polymorphisms (SNPs) in high linkage disequilibrium (LD) and to identify p-value thresholds with the best prediction accuracy is the pruning and threshold (P+T) method [25]. In the P+T approach, the correlation square (R2) for SNPs within a close genetic distance is calculated and less significant SNPs that are correlated with an R2 greater than a particular value (LD pruning) [26] are removed. Next, several p-value thresholds are tested to maximize the prediction accuracy of the derived PRS (p-value thresholding) [26,27]. Theoretically, the P + T approach could be applied to generate MRS, however, there is no standard procedure on how to conduct pruning for DNAm data and the performance of such MRS across multi-ancestry populations remains unknown. Here, we propose to use the Co-Methylation with genomic CpG Background (CoMeBack) approach, a tool that uses a sliding window to estimate DNA co-methylation, to account for correlations of DNAm at proximal CpG sites [22], and pair it with p-value thresholding to construct P+T CoMeBack MRS. CoMeBack uses unmeasured intermittent CpGs from the human reference genome to link array probes in hope of reducing false positives while improving the identification of biologically relevant co-methylation [22]. We conducted simulation studies based on data from an adult population consisting of three groups of different ancestries (Indian, White and Black, $$n = 1$$,199) to evaluate the prediction performance of P+T CoMeBack MRS and how it changes across multi-ancestry population. Next, we applied the P+T CoMeBack approach to DNAm data from the Drakenstein Child Health Study ($$n = 260$$) [28], a multi-ancestry birth cohort from South Africa, to evaluate the performance of MRS for maternal smoking status. Our simulation study and real data application demonstrated that the P+T approach improves the predictive accuracy of MRS over methods that do not account for co-methylation and has similar performance as LASSO regression, which requires access to the raw DNAm data. We also showed that MRS built upon the data from a population of one genetic ancestry could achieve high prediction performance among populations of other genetic ancestries, but the performance might differ in the presence of environmental/cultural/social differences associated with ancestry. ## P+T CoMeBack approach for MRS P+T CoMeBack method refers to the calculation of MRS using informed co-methylation pruning (P) with CoMeBack and P-value thresholding (T). First, summary statistics from an EWAS (typically include the effect size, standard error, and P-value of each CpG site) need to be estimated in an independent dataset (training dataset) to avoid overfitting, and then applied to generate MRS in a testing dataset (the samples used to evaluate the performance of MRS). In our P+T CoMeBack method, co-methylation pruning is completed by applying CoMeBack to DNAm data of the testing dataset or a reference panel [22]. Specifically, CoMeBack chains two adjacent array probes if the following requirements are met: 1. two probes are less than 2kb apart; 2. the reference human genome annotation shows a set of unmeasured genomic CpGs between them; 3. the density of unmeasured genomic CpGs between them is at least one CpG every 400bp. Chaining of adjacent array probes continues until an array probe does not meet the requirements, which will form a unit where multiple CpGs are chained together. Correlations between DNAm levels will then be calculated for all array probes inside each unit. If all pairs of adjacent probes in a unit have a correlation square (R2) greater than 0.3, such unit will be declared as a co-methylated region (CMR). Pruning is conducted by only keeping one CpG site per CMR in the dataset, the one with the lowest (most significant) P-value in the EWAS summary statistics. P+T CoMeBack will be compared to the standard pruning approach, in which less significant CpGs that are correlated with an R2>0.3 and located within 2000bp of each other are being removed. Next, P-value thresholding step (T) is performed for the pruned set of CpG sites. Specifically, the P-value thresholding step (T) is performed by applying different P-value thresholds (e.g., P-value thresholds ε [0.05, 0.005, 5 × 10−4, 5 × 10−5, …]) and only including those CpG sites in the final MRS calculation that reached a P-value below those thresholds in the EWAS summary statistics. Finally, for each P-value threshold, MRS are calculated as a weighted sum of DNAm β values (β value = methylated allele intensity/(unmethylated allele intensity + methylated allele intensity+100), ranging from 0 representing unmethylation to 1 for complete methylation) of the selected CpGs, where the weights are the corresponding effect sizes for each CpG from the EWAS summary statistics. The squared correlation (R2) between the phenotype of interest and MRS obtained using each P-value threshold is calculated to represent the prediction accuracy. The P-value threshold that produces MRS with the highest prediction accuracy in the testing data set is selected as the optimal P-value and the corresponding MRS is used for downstream analysis. The pipeline for generating P+T MRS (P+T CoMeBack as well as P+T with the standard pruning approach) is written in an R script, which is available at GitHub (https://github.com/jche453/Pruning-Thresholding-MRS.git). In our simulation studies and real data application, we compare the P+T CoMeBack MRS approach to the standard P+T and T approach, which refers to an approach in which the MRS is calculated by only thresholding, not accounting for correlations between included CpGs (no pruning). ## Simulation studies To validate the performance of the proposed P+T MRS approach, we conducted simulation studies based on whole blood Illumina Infinium Human Methylation 450K BeadChip data from an ethnically heterogeneous discovery cohort composed of several publicly available datasets (GSE55763, GSE84727, GSE80417, GSE111629, and GSE72680) [22]. Intra-dataset normalization and batch effects correction were performed using ComeBat function in R-package sva [29], followed by merging of datasets and correction for inter-dataset batch effects using the same function. After removing XY chromosome binding, non-CpG, cross-hybridizing probes and probes that are in close distance (1bp) with common SNPs, there were 386,362 CpGs left for MRS analysis. We randomly selected 1,199 adults (898 Indians, 136 Blacks and 165 Whites) to conduct the simulation studies. CoMeBack was applied to the DNA methylation β values of the 386,362 CpGs to obtain CMR. In each simulation, 10 of the 386,362 CpGs were randomly selected to be causal, k% ($k = 30$, 50, 70 or 100) of which are in a CMR with other CpGs. At most, one CpG would be causal in each CMR. The causal CpGs were randomly assigned a ‘true’ effect size from a uniform distribution as wi∼U−0.5,0.5. We then simulated a phenotype for the j-th subject as follows:Yj=∑$i = 110$wimij+εj,εj∼N0,δ2, where mijis the DNAm β value of causal CpG site i of the j-th subject, and εj is an error term that follows a normal distribution. Differentδ2 were set to ensure that the targeted variance of phenotype explained by DNAm alone equals $10\%$, $30\%$ $50\%$ or $80\%$. We also simulated a second phenotype Yj for the j-th subject, which was directly affected by ancestry using European ancestry as reference:Yj∗=∑$i = 110$wimij+a∗ifIndian+b∗ifBlack+εj∗,εj∗∼N0,δ∗2 Effect of ancestry in our simulations is simulated as the effect of genetic ancestry assuming there were no complex social determinants involved in the causal pathway. Different δ∗2 were used so that the variance of phenotype that was explained by DNAm and ancestry together equals $20\%$, $50\%$ or 80. For our simulations, effect a was set to 0.1 and b to 0.2. In each simulation, both simulated phenotypes Yj and Yj∗ share the same epigenetic liability (∑$i = 110$wimij). In each simulation, for fair comparison, 762 Indians were randomly chosen as the training dataset so that there were at least 136 people left for each race group in the testing dataset. Associations between CpGs and each of the two simulated phenotypes were assessed by robust linear regression model using limma R package [30] in the training dataset. We calculated top 10 principal components (PCs) from DNAm of 386,362 CpGs [31] and used EpiDISH to estimate cell type proportions of each CpGs [32]. We observed that in our simulation dataset, top 10 PCs are highly correlated with cell type proportions (Supplement Figure 1a), and using either summary statistics adjusted for top 10 PCs or summary statistics adjusted for cell type proportions would lead to almost identical prediction performance of MRS (Supplement Figure 1b). Thus, to account for population stratification and cell type difference, we adjusted for the top 10 PCs in our main analyses. The summary statistics (effect size and P-values) obtained from association tests in the training data were saved and later used to construct MRS in the testing dataset. We repeated 1000 simulations per scenario to evaluate the prediction accuracy (R2), statistical power, and type 1 error rate of the P+T MRS. Linear regression analysis was used to access the association between MRS and simulated phenotypes in the test data, and statistical power is defined as the proportion of simulations where MRS were significantly associated with the simulated phenotype with at α level of 0.05. To estimate type 1 error rate, we first obtained a null association between MRS values and simulated phenotype values by permutation of MRS values. Linear regression analysis was used to access the association between permutated MRS and simulated phenotypes, and type 1 error rate is defined as the proportion of simulations where permutated MRS were significantly associated with the simulated phenotype. Figure 1.Simulation study. Prediction R2 of methylation risk scores (MRS) estimated with pruning and thresholding + Co-Methylation with genomic CpG Background ((P+T CoMeBack), P+T and thresholding (T) method in dependence of (A) the proportion of causal CpG sites in co-methylation regions (CMRs) and (B) proportion of phenotype variance explained by DNA methylation, among Indian participants. For each simulation, the discovery cohort was repeatedly and randomly split into a training set comprising 762 Indians and a testing set comprising 136 people of the same ancestry. Phenotypes were simulated without an influence of ancestry. Results are shown for (a) different proportions of causal CpGs located in CMR ($30\%$, $50\%$, $70\%$, $100\%$) and (b) different proportions of phenotype variance explained by DNA methylation ($10\%$, $30\%$, $50\%$, $80\%$). Each box represents the distribution of prediction accuracy across 1000 simulations, where the central mark is the median and the edges of the box are the 25th and 75th percentiles. We evaluated the performance of the MRS not only in scenarios of A) same ancestry in training and test data, but also B) across different ancestry groups (training data: Indian, test data: European or African) and C) in multi-ancestry populations (training data: Indian, test data: Indian, European, and African). For scenario C), we evaluated two analysis strategies: 1. Joint-analysis: MRS analyses in the whole testing dataset where subjects from all racial groups were merged; 2. Standardization: scale MRS to have a standard normal distribution within each racial group before merging all subjects for analyses. ## Application study of smoking MRS To evaluate the performance of the P+T CoMeBack approach in a real data setting, we applied the P+T CoMeBack approach to calculate a MRS for maternal smoking status during pregnancy using cord blood DNAm data from newborns in the South African Drakenstein Child Health Study (DCHS), a multi-ancestry longitudinal study investigating determinants of early child development [33]. There were 145 Black African infants and 115 Mixed ancestry infants in the DCHS. A detailed description of the enrolment process, inclusion criteria, variables measurement, and ethical approval of the study have been previously published [33,34]. Cotinine levels were measured in urine provided by mothers within four weeks of enrolment and classified as <499 ng/ml (non-smoker), or ≥500 ng/ml (active smoker) [28]. Cord blood was collected at time of delivery and used to measure DNA methylation by either MethylationEPIC BeadChips (EPIC, $$n = 145$$) or the Illumina Infinium HumanMethylation450 BeadChips (450K, $$n = 103$$) [33,34], followed by quality control and normalization to calculate β values (details have been published elsewhere) [35]. Summary statistics for the calculation of MRS were obtained from a study that meta-analysed the associations between newborn blood DNA methylation and sustained maternal smoking during pregnancy among 5,648 mother-child pairs as part of the Pregnancy and Childhood Epigenetics (PACE) Consortium (Table 1) [36]. The participants of all cohorts used in the meta-analysis except one were of European ancestry. Table 1.Overview of included EWAS, their phenotypes, training sample and methods. MRSTraining dataset publicationTraining PopulationPhenotypeMRS publicationP-value threshold/MethodNo. of CpG sites (joint-analysis)P+T CoMeBack MRSSikdar et al.2019 [36]Multi-ethnic newborns (mainly White, $$n = 5$$,648)Most cohorts ascertained sustained smoking during pregnancy by questionnaires; two cohorts incorporated cotinine-based smoking measureProposed in this article5×10−22 (Mixed)21 (43 passed P-value threshold and 22 excluded by pruning)5×10−24 (Black)20 (42 passed P-value threshold and 22 excluded by pruning)5×10−22 (Pooled)21 (43 passed P-value threshold and 22 excluded by pruning)P+T MRSSikdar et al.2019 [36]Multi-ethnic newborns (mainly White, $$n = 5$$,648)Most cohorts ascertained sustained smoking during pregnancy by questionnaires; two cohorts incorporated cotinine-based smoking measureProposed in this article5×10−10 (Mixed)198 (233 passed P-value threshold and 35 excluded by pruning)5×10−36 (Black)4 (26 passed P-value threshold and 22 excluded by pruning)5×10−24 (Pooled)8 (42 passed P-value threshold and 34 excluded by pruning)T MRSSikdar et al.2019 [36]Multi-ethnic newborns (mainly White, $$n = 5$$,648)Most cohorts ascertained sustained smoking during pregnancy by questionnaires; two cohorts incorporated cotinine-based smoking measureProposed in this article5×10−9 (Mixed)3445×10−24 (Black)425×10−16 (Pooled)72Reese MRSReese et al. 2017 [37]White newborns ($$n = 1$$,068)Sustained smoking during pregnancy obtained from combined information of cotinine-based and self-report based classificationReese et al. 2017 [37]Logistic LASSO regression28Richmond 568 MRSJoubert et al. 2016 [38]Multi-ethnic newborns ($$n = 6$$,685)Maternal smoking during pregnancy via questionnairesRichmond et al. 2018 [39]Robust linear regression; Bonferroni corrected P-value<0.05568Richmond 19 MRSJoubert et al. 2016 [38]Multi-ethnic older children (average age = 6.8 years) ($$n = 3$$,187)Maternal smoking during pregnancy via questionnairesRichmond et al. 2018Robust linear regression; Bonferroni corrected P-value<0.0519Note: These training populations overlapped with training population for summary statistics used for P+T MRS.Abbreviations: Methylation risk scores (MRS); Pruning and thresholding (P+T); Co-Methylation with genomic CpG Background (CoMeBack); Thresholding (T). In addition, we compared P+T CoMeBack MRS to three previously published MRS for maternal smoking during pregnancy (Reese MRS, Richmond 19 MRS, Richmond 568 MRS; Table 1). Reese MRS model was trained among 1,068 newborns of European ancestry in the Norwegian Mother and Child Cohort Study, while Richmond 568 MRS and Richmond 19 MRS was trained in multi-ancestry newborns ($$n = 6$$,685) and children around 6.8 years old ($$n = 3$$,187) in PACE Consortium respectively. The training population for Reese MRS and Richmond 19 MRS overlapped with the training population for summary statistics used in P+T CoMeBack MRS in our study. Reese et al. used a LASSO regression to select CpGs for Reese MRS, which is a weighted sum of DNAm β values of 28 CpGs with weights estimated from the LASSO regression [37]. Richmond 19 MRS is a weighted sum of DNAm β values of 19 CpGs that were significantly associated with prenatal smoking in an EWAS conducted in peripheral blood from children of averaged 6.8 years age (Richmond 19 MRS) [38,39]. In the same study, Richmond 568 MRS was proposed based on 568 CpGs that were significantly associated with prenatal smoking in cord blood [38,39]. We obtained the weights of reported CpGs from the mentioned studies and applied them to DNAm data in the DCHS to generate Reese MRS, Richmond 19 MRS and Richmond 568 MRS. Linear regressions were used to assess the associations between maternal smoking status and each MRS, using maternal smoking as the independent variable and MRS as the dependent variable. Each model was adjusted for ancestry (in pooled samples), cell type proportions and top 5 PCs calculated from genotypes. In order to obtain comparable beta-coefficients and standard errors across different MRS, each MRS was divided by their interquartile range (IQR) before linear regression analysis. ## Simulation results We compared the prediction performance of P+T CoMeBack MRS to the T method among 136 Indians in the test data across different simulation scenarios (Figure 1). Figure 1a shows that P+T CoMeBack MRS that account for co-methylation between CpGs have stable prediction performance when proportion of causal CpGs located in a CMR (k%) varies. P+T without CoMeBack had similar prediction performance while the T method had a slightly lower prediction performance. While the P+T CoMeBack MRS showed subtle improvement over T method when proportions of phenotype variance explained by DNA methylation is $80\%$ (Figure 1a), the difference between P+T CoMeBack MRS and the T method decreases as the proportions of phenotype variance explained by DNA methylation decreases. This is likely because as variance explained by DNAm decreases, there is less power for association testing, and it becomes increasingly difficult to distinguish real signals from statistical noise while generating the summary statistics. Next, we assessed the performance of P+T CoMeBack, P+T, and T method across different ancestries and among multi-ancestry populations. All three methods achieved a high power (>$95\%$) and a low type 1 error rate (~$5\%$) within each ancestry in most scenarios for both phenotypes except when the phenotype variance explained by DNA methylation is $10\%$ or $30\%$ (Supplement Table 1-4). Whether the simulated phenotypes were independent of ancestry or not, MRS among Whites and Blacks achieved a prediction R2 as high as among Indians, which should have the best prediction of the simulated phenotypes since weights were obtained from Indian training samples (Figure 2). Findings were similar when the phenotype variance explained by DNA methylation was reduced from $80\%$ to $10\%$, $30\%$ or $50\%$ (Supplement Figure 2). When the phenotypes are not associated with ancestry (Figure 2a), the three MRS analyses strategies (stratification, joint analysis, and standardization) lead to nearly identical results. However, when the phenotypes are ancestry-dependent, both joint analysis and standardization of MRS showed very poor prediction of the phenotypes (Figure 2b). Figure 2.Simulation study. Prediction R2 of methylation risk scores (MRS) estimated with pruning and thresholding + Co-Methylation with genomic CpG Background (P+T CoMeBack), P+T and thresholding (T) approach across different racial groups and among multi-ancestry populations. For each simulation, the discovery cohort was repeatedly and randomly split into a training set comprising 762 Indians and a testing set comprising 136 people of each ancestry group. The proportion of causal CpGs located in co-methylation regions (CMR) is $70\%$ and the proportion of phenotype variance explained by DNA methylation (and ancestry) is $80\%$. Results are shown for the prediction of simulated phenotypes (a) without an influence of ancestry and (b) influenced by ancestry. Joint-analysis refers to MRS analyses of all participants pooled from all ancestry groups and standardization refers to standardizing MRS within each ancestry group and then merging all participants before analyses. Each box represents the distribution of prediction accuracy across 1000 simulations, where the central mark is the median and the edges of the box are the 25th and 75th percentiles. ## MRS of maternal smoking status Figure 3 shows the prediction performance of MRS for maternal smoking status among DCHS newborns. As the p-value threshold decreases, the prediction accuracy of the resulting MRS increases before reaching a plateau, demonstrating the importance of P-value thresholding in MRS to control for noise. Among mixed ancestry newborns, P+T CoMeBack MRS of smoking status excluded 22 CpGs in pruning and achieved a prediction R2 of $29.5\%$ using P-value threshold of 5 × 10−22, while the best standard P+T without CoMeBack had a lower prediction R2 of $26.2\%$ using P-value threshold of 5 × 10−10 and the best T method MRS had the lowest prediction accuracy ($24.5\%$) using P-value threshold 5 × 10−9, confirming the benefits of pruning in MRS calculation (Figure 3a). All three MRS had lower prediction performance for maternal smoking among Black African infants ($10.9\%$, and $8.0\%$ respectively) (Figure 3b), which is likely due to the low prevalence of smokers among mothers of Black African infants in DCHS ($13\%$) compared to mothers of mixed ancestry infants ($49\%$) (Supplement Figure 3). Additionally, the distributions of all MRS in Black African infants and mixed ancestry infants were similar within each category of maternal smoking status (Supplement Figure 4), confirming that the difference of prediction R2 between Mixed and Black infants is less likely due to ancestry-related factors other than prevalence of maternal smoking. Joint-analysis of P+T CoMeBack MRS showed a prediction accuracy of $20.4\%$, which is between the prediction accuracy of P+T CoMeBack MRS among Black African infants and mixed ancestry infants (Figure 3c). Standardization approach did not improve the performance of MRS (Figure 3d). Figure 3.Real data application. Methylation risk scores (MRS) for the prediction of maternal smoking during pregnancy using cord blood DNA methylation data from newborns in the South African Drakenstein Child Health Study (DCHS). Prediction R2 of maternal smoking status is shown stratified for a. Mixed infants. b. Black infants. c. joint-analysis (all subjects pooled from all ancestries) d. Standardization (standardizing MRS within each ancestry and merging all subjects before analyses. Figure 4.Real data application. Comparison of MRS estimated with pruning and thresholding+ Co-Methylation with genomic CpG Background (P+T CoMeBack), P+T, Thresholding (T) and 3 other published methylation risk scores (MRS) for predicting maternal smoking status in the South African Drakenstein Child Health Study (DCHS). a. A Prediction R2 of all 6 MRS methods for Mixed infants, Black infants and pooled samples (joint-analysis). A receiver operating characteristic (ROC) curve comparing prediction performance of all 6 MRS among (b) Mixed infants, (c) Black infants and (d) pooled samples (joint-analysis). We next compared the prediction accuracy R2 and distribution of P+T CoMeBack MRS to other established MRS for maternal smoking during pregnancy and newborn DNAm (Figure 4). Overall, P+T CoMeBack and Reese MRS had stable and similar classification performance in all analyses compared to other MRS. P+T CoMeBack MRS and Reese MRS showed a similar prediction R2 among both Black and Mixed ancestry infants, which were better than other smoking MRS (Figure 4A). P+T CoMeBack MRS had the largest AUC (0.820) in the ROC curve among mixed infants (Figure 4b) but a smaller AUC than Reese MRS in Black infants and joint-analysis (Figure 4c-d). Further, all 6 MRS showed significant association with smoking status in Mixed infants, Black infants and joint-analysis (Table 2), showing the promise of using MRS to capture the overall DNAm signals in association testing. Table 2.Association between maternal smoking status and MRS in DCHS.MRSMixedBlackPooled (Joint-analysis)Beta-coefficientStandard ErrorP-valueBeta-coefficientStandard ErrorP-valueBeta-coefficientStandard ErrorP-valueP+T CoMeBack MRS0.880.125.65 ×10−110.760.185.74 ×10−50.820.103.01×10−14P+T MRS0.710.111.05 ×10−80.520.143.29 ×10−40.690.101.32×10−10T MRS0.800.137.35 ×10−90.700.192.82 ×10−40.640.092.55 ×10−11Reese MRS0.970.133.77 ×10−110.730.162.06 ×10−50.770.092.50 ×10−15Richmond 568 MRS0.850.149.84 ×10−90.540.167.82 ×10−40.650.101.57 ×10−10Richmond 19 MRS0.760.145.37E x10−70.700.193.78 ×10−40.700.112.05 ×10−9Beta-coefficients indicate change in interquartile range (IQR) of MRS between smokers and non-smokers. Abbreviations: Methylation risk scores (MRS); South African Drakenstein Child Health Study (DCHS); Pruning and thresholding (P+T); Co-Methylation with genomic CpG Background (CoMeBack); Thresholding (T). ## Discussion Based on the well-established P+T framework in PRS, we developed P+T CoMeBack MRS, which aggregates EWAS signals into a MRS [44,40,41]. The proposed P+T CoMeBack MRS approach uses CoMeBack for co-methylation pruning and evaluates multiple P-value thresholds to maximize prediction performance. Such MRS could potentially serve as a powerful dimension reduction approach for mediation and multi-omics integration analyses [44,40,41,42,43] as well as biomarkers of individual disease risk in a clinical setting [44,45,46]. Overall, our simulation studies demonstrated good performance of P+T CoMeBack MRS for predicting phenotypes of interest with good statistical power and well-controlled type 1 error. We demonstrated that the prediction accuracy of MRS reflects the variance of phenotype that is explained by DNAm. By accounting for inter-correlation between CpGs, P+T CoMeBack MRS and P+T without CoMeBack showed a slightly better performance than the standard T method. In the real data application, we observed the best prediction of maternal smoking status when using P+T CoMeBack, which confirms the usefulness of accounting for co-methylation and demonstrates the ability of CoMeBack to control for false discover of CMR and usefulness in constructing MRS [22]. However, we note that P+T CoMeBack MRS could still have poor prediction performance if the external EWAS is underpowered or subject to bias. In the prediction of maternal smoking status, P+T CoMeBack MRS showed comparable performance to Reese MRS, which was derived using the LASSO method [37]. When predictors are highly correlated, LASSO typically selects one of the correlated predictors and shrinks the effect size of the rest to zero, which might produce similar results to our pruning procedure in developing MRS. One of the advantages of P+T CoMeBack MRS is that it is based on EWAS summary statistics which are often publicly available, hence making it a valuable approach, as it is often difficult to obtain individual DNAm data from an external cohort. Additionally, P+T CoMeBack MRS can make use of meta-analysis-type summary statistics, which aggregates results from multiple studies to improve association estimates. In contrast, to construct MRS like Reese MRS, individual DNAm data are usually required to perform a LASSO regression, and these are often not accessible. Furthermore, approaches like LASSO regression require the data to be split into training and test data to avoid overfitting. Consequently, larger sample sizes are needed to develop MRS using LASSO regression than when using publicly available EWAS summary statistics. Recently, novel penalized regressions have been proposed to generate PRS with only GWAS summary statistics and publicly available reference data [47], but their applications to EWAS summary statistics for MRS have not been investigated. To develop MRS for different exposures and outcomes, we urge EWAS studies to make their epigenome-wide summary statistics publicly available. In our simulation studies, weights obtained from Indian training samples were applied to generate P+T CoMeBack MRS, thus MRS among Indian testing samples were assumed to have the best prediction of the simulated phenotypes. However, MRS among Whites and Blacks also achieved a prediction accuracy as high as among Indians for both simulated phenotypes suggesting that genetic ancestry does not contribute to difference in prediction abilities of MRS across multi-ancestry population. This is likely because we assumed all ancestries share the same causal CpGs and effect sizes. However, in the real world, this assumption could possibly be violated for many phenotypes. Unlike ancestry in our simulation studies, ancestry in the real world is complex. The meaning of ancestry could be different in different regions/nations, and ‘effect of ancestry’ involves the joint effects of ancestry-associated social determinants of health and environmental effects, and cultural context [48]. Ancestry, along with environment and social differences associated with it, could affect both MRS and phenotypes in numerous causal pathways and potentially modify the effect of MRS on the phenotypes. Thus, even if all ancestries indeed share the same causal CpGs and effect sizes, it might still not be sufficient to disentangle the relationship between ancestry, DNAm and phenotype of interest. This may greatly impact the transferability of MRS across different ancestries, which could be the reason why we observed an inconsistency of performance of P+T CoMeBack MRS in terms of their distributions and predictions across multi-ancestry population in the real data analyses. In practice, we recommend that researchers conduct MRS analyses stratified by ancestry first and evaluate the effect of ancestry on MRS analyses before pooling participants together for a joint analysis. In our real data application, summary statistics for smoking were obtained from a cohort with mainly people of European ancestry [36]. MRS of smoking among mixed ancestry infants achieved a prediction accuracy of nearly $30\%$. However, the prediction accuracy of P+T CoMeBack MRS among Black African infants was only $10.9\%$. We suspect that the difference was largely due to the prevalence of active smoking among mothers of Black African infants being lower than those of mixed ancestry infants ($13\%$ vs $49\%$), which is similar to how the prevalence of outcome affects the predictive ability of PRS [49]. To the best of our knowledge, this is the first study to propose using CoMeBack for pruning MRS among multi-ancestry populations. However, there are several potential limitations that warrant mention. First, in the simulation dataset, ancestry information was self-reported and it may not fully represent the genetic ancestry estimated from genetic data. Second, the sample size of the simulation testing dataset and real data analyses was relatively small. Third, lack of different ancestry-specific summary statistics made it impossible to compare the use of external weights from population of different ancestries (e.g., European ancestry vs other ancestries). Fourth, the prevalence of active maternal smoking was different in different ancestries and has influenced the performance of P+T CoMeBack MRS in the real data application. As a result, real data analysis of smoking MRS could not provide firm evidence about the transferability of MRS between Black African and mixed ancestry infants. Last, we mainly focused on the prediction performance of P+T CoMeBack MRS. Further studies are needed to assess the performance of P+T CoMeBack MRS in mediation analysis. In conclusion, P+T in general and P+T using CoMeBack in particular, provides an improvement for prediction of phenotype of interest, over T method that does not account for co-methylation between CpGs. In contrast to PRS, using existing summary statistics that were derived from European populations can be used to calculate MRS in other ancestries, thus reducing the ancestry/ethnicity disparity in medical research. However, caution is needed in the analyses and interpretation of MRS results across multi-ancestry populations, especially due to environmental/cultural/social differences associated with ancestry. More investigations of MRS are urged to further improve their prediction accuracy and translational values, also in combination with other clinical and non-clinical variables, especially among multi-ancestry population. With the current increase of large consortia-led EWAS for different exposures and health outcomes (e.g., the PACE consortium), we believe the predictive performance of MRS will continue to increase, and the P+T CoMeBack method has the potential to be widely used for risk prediction and association testing. ## Disclosure statement All authors declare they have no actual or potential competing financial interest. ## Data availability statement The data for simulation studies were derived from the following dataset (GSE55763, GSE84727, GSE80417, GSE111629 and GSE72680) from NCBI Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo/). The data for real data application are available from the corresponding author, [AH], upon reasonable request. ## Supplementary material Supplemental data for this article can be accessed online at https://doi.org/$\frac{10.1080}{15592294.2023.2187172}$ ## References 1. Berger SL, Kouzarides T, Shiekhattar R. An operational definition of epigenetics: figure 1. Genes Deve 23 781-14 2. Rakyan VK, Down TA, Balding DJ. Epigenome-wide association studies for common human diseases. Nat Rev Genet (2011.0) 12 529-541. PMID: 21747404 3. Wei S, Tao J, Xu J. Ten Years of EWAS. 8 e2100727. DOI: 10.1002/advs.202100727 4. Wahl S, Drong A, Lehne B. Epigenome-wide association study of body mass index, and the adverse outcomes of adiposity. Nature 541 81-86. DOI: 10.1038/nature20784 5. McCartney DL, Hillary RF, Stevenson AJ. Epigenetic prediction of complex traits and death. Genome Bio (2018.0) 19 136. DOI: 10.1186/s13059-018-1514-1 6. Heiss JA, Brenner H.. Epigenome-wide discovery and evaluation of leukocyte DNA methylation markers for the detection of colorectal cancer in a screening setting. Clin Epigenetics (2017.0) 9 24. PMID: 28270869 7. Ka G, Shaikhibrahim Z, Ocker M. Targeting epigenetic regulators for cancer therapy: modulation of bromodomain proteins, methyltransferases, demethylases, and microRnas. Expert Opin Ther Targets (2016.0) 20 783-799. DOI: 10.1517/14728222.2016.1134490 8. Krushkal J, Silvers T, Reinhold WC. Epigenome-wide DNA methylation analysis of small cell lung cancer cell lines suggests potential chemotherapy targets. Clin Epigenetics (2020.0) 12 93. DOI: 10.1186/s13148-020-00876-8 9. Guan Z, Yu H, Cuk K. Whole-blood DNA methylation markers in early detection of breast cancer: a systematic literature review. Cancer Epidemiol Biomarkers Prev (2019.0) 28 496-505. PMID: 30487132 10. Hüls A, Czamara D.. Methodological challenges in constructing DNA methylation risk scores. Epigenetics (2020.0) 15 1-11. PMID: 31318318 11. Wray NR, Lee SH, Mehta D. Research review: polygenic methods and their application to psychiatric traits. J Child Psychol Psychiatry (2014.0) 55 1068-1087. PMID: 25132410 12. Martin EM, Fry RC. Environmental influences on the epigenome: exposure- associated DNA methylation in human populations. Ann Rev Public Health (2018.0) 39 309-333. PMID: 29328878 13. Notterman DA, Mitchell C. Epigenetics and understanding the impact of social determinants of health. (2015.0) 62 1227-1240. PMID: 26318949 14. Borrell LN, Elhawary JR, Fuentes-Afflick E. Race and genetic ancestry in medicine — a time for reckoning with racism. N Engl J Med (2021.0) 384 474-480. PMID: 33406325 15. Cronjé HT, Elliott HR, Nienaber-Rousseau C. Replication and expansion of epigenome-wide association literature in a black South African population. Clin Epigenetics (2020.0) 12 6. PMID: 31910897 16. Khera AV, Chaffin M, Zekavat SM. Whole-genome sequencing to characterize monogenic and polygenic contributions in patients hospitalized with early-onset myocardial infarction. Circulation 139 1593-1602. DOI: 10.1161/circulationaha.118.035658 17. Martin AR, Kanai M, Kamatani Y. Clinical use of current polygenic risk scores may exacerbate health disparities. Nature Genet (2019.0) 51 584-591. PMID: 30926966 18. Horvath S. DNA methylation age of human tissues and cell types. Genome Bio (2013.0) 14 R115. DOI: 10.1186/gb-2013-14-10-r115 19. Thompson M, Hill BL, Rakocz N. Methylation risk scores are associated with a collection of phenotypes within electronic health record systems. NPJ Genom Med (2022.0) 7 50. PMID: 36008412 20. Goldstein BA, Yang L, Salfati E. Contemporary considerations for constructing a genetic risk score: an empirical approach. (2015.0) 39 439-445. PMID: 26198599 21. Euesden J, Lewis CM, O’reilly PF. Prsice: polygenic risk score software. Bioinformatics (2015.0) 31 1466-1468. PMID: 25550326 22. Gatev E, Gladish N, Mostafavi S. DNA methylation array data analysis for co-methylated regions. Bioinformatics 36 2675-2683. DOI: 10.1093/bioinformatics/btaa049 23. Shah S, Bonder MJ, Marioni RE. Improving phenotypic prediction by combining genetic and epigenetic associations. 97 75-85. DOI: 10.1016/j.ajhg.2015.05.014 24. Odintsova VV, Rebattu V, Fa H. Predicting complex traits and exposures from polygenic scores and blood and buccal DNA methylation profiles. Front Psychiatry (2021.0) 12 688464. PMID: 34393852 25. Wu J, Pfeiffer RM, Gail MH. Strategies for developing prediction models from genome-wide association studies. (2013.0) 37 768-777. PMID: 24166696 26. Privé F, Vilhjálmsson BJ, Aschard H. Making the most of clumping and thresholding for polygenic scores. Am J Hum Genet (2019.0) 105 1213-1221. PMID: 31761295 27. Choi SW, O’reilly PF. Prsice-2: polygenic risk score software for biobank-scale data. Gigascience (2019.0) 8. DOI: 10.1093/gigascience/giz082 28. Vanker A, Barnett W, Workman L. Early-life exposure to indoor air pollution or tobacco smoke and lower respiratory tract illness and wheezing in African infants: a longitudinal birth cohort study. (2017.0) 1 e328-336. PMID: 29167839 29. Leek JT, Johnson WE, Parker HS. The sva package for removing batch effects and other unwanted variation in high-throughput experiments. Bioinformatics 28 882-883. PMID: 22257669 30. Ritchie ME, Phipson B, Wu D. Limma powers differential expression analyses for RNA-sequencing and microarray studies. 43 e47. DOI: 10.1093/nar/gkv007 31. Barfield RT, Almli LM, Kilaru V. Accounting for population stratification in DNA methylation studies. Genet Epidemiol (2014.0) 38 231-241. DOI: 10.1002/gepi.21789 32. Teschendorff AE, Breeze CE, Zheng SC. A comparison of reference-based algorithms for correcting cell-type heterogeneity in epigenome-wide association studies. BMC Bioinf 18 105 33. Zar HJ, Barnett W, Myer L. Investigating the early-life determinants of illness in Africa: the drakenstein child health study. Thorax (2015.0) 70 592-594. PMID: 25228292 34. Stein DJ, Koen N, Donald KA. Investigating the psychosocial determinants of child health in Africa: the drakenstein child health study. 252 27-35. DOI: 10.1016/j.jneumeth.2015.03.016 35. Hüls A, Wedderburn CJ, Groenewold NA. Newborn differential DNA methylation and subcortical brain volumes as early signs of severe neurodevelopmental delay in a South African birth cohort study. World J Biol Psychiatry 1-12. DOI: 10.1080/15622975.2021.2016955 36. Sikdar S, Joehanes R, Joubert BR. Comparison of smoking-related DNA methylation between newborns from prenatal exposure and adults from personal smoking. Epigenomics (2019.0) 11 1487-1500. PMID: 31536415 37. Reese SE, Zhao S, Wu MC. DNA methylation score as a biomarker in newborns for sustained maternal smoking during pregnancy. (2017.0) 125 760-766. PMID: 27323799 38. Joubert BR, Felix Janine F, Yousefi P. DNA methylation in newborns and maternal smoking in pregnancy: genome-wide consortium meta-analysis. Am J Hum Genet (2016.0) 98 680-696. PMID: 27040690 39. Richmond RC, Suderman M, Langdon R. DNA methylation as a marker for prenatal smoke exposure in adults. Int J Epidemiol (2018.0) 47 1120-1130. PMID: 29860346 40. Yu H, Raut JR, Schöttker B. Individual and joint contributions of genetic and methylation risk scores for enhancing lung cancer risk stratification: data from a population-based cohort in Germany. 12 89. PMID: 32552915 41. Westerman K, Fernández-Sanlés A, Patil P. Epigenomic assessment of cardiovascular disease risk and interactions with traditional risk metrics. 9 e015299. DOI: 10.1161/jaha.119.015299 42. Guan Z, Raut JR, Weigl K. Individual and joint performance of DNA methylation profiles, genetic risk score and environmental risk scores for predicting breast cancer risk. (2020.0) 14 42-53. DOI: 10.1002/1878-0261.12594 43. Grant CD, Jafari N, Hou L. A longitudinal study of DNA methylation as a potential mediator of age-related diabetes risk. Geroscience (2017.0) 39 475-489. PMID: 29159506 44. García-Calzón S, Perfilyev A, Martinell M. Epigenetic markers associated with metformin response and intolerance in drug-naïve patients with type 2 diabetes. 12. DOI: 10.1126/scitranslmed.aaz1803 45. Deng Y, Wan H, Tian J. CpG-methylation-based risk score predicts progression in colorectal cancer. Epigenomics (2020.0) 12 605-615. PMID: 32180433 46. Kilanowski A, Chen J, Everson T. Methylation risk scores for childhood aeroallergen sensitization: results from the {LISA} birth cohor. (2021.0). DOI: 10.22541/au.163620398.85835627/v1 47. Pattee J, Pan W, Zhang D. Penalized regression and model selection methods for polygenic scores on summary statistics. (2020.0) 16 e1008271. PMID: 33001975 48. VanderWeele TJ, Robinson WR. On the causal interpretation of race in regressions adjusting for confounding and mediating variables. Epidemiology (2014.0) 25 473-484. PMID: 24887159 49. Gibson G, Barsh GS. On the utilization of polygenic risk scores for therapeutic targeting. (2019.0) 15 e1008060. PMID: 31022172
--- title: Circ_0078607 increases platinum drug sensitivity via miR-196b-5p/GAS7 axis in ovarian cancer authors: - Cheng Dai - Shi-Yuan Dai - Yan Gao - Ting Yan - Qi-Yin Zhou - Shi-jun Liu - Xuan Liu - Dan-Ni Deng - Dong-Hong Wang - Qing-Feng Qin - Dan Zi journal: Epigenetics year: 2023 pmcid: PMC10026884 doi: 10.1080/15592294.2023.2175565 license: CC BY 4.0 --- # Circ_0078607 increases platinum drug sensitivity via miR-196b-5p/GAS7 axis in ovarian cancer ## ABSTRACT Platinum-based chemotherapy is one of the predominant strategies for treating ovarian cancer (OC), however, platinum resistance greatly influences the therapeutic effect. Circular RNAs (circRNAs) have been found to participate in the pathogenesis of platinum resistance. Our aim was to explore the involvement of circ_0078607 in OC cell cisplatin (DDP) resistance and its potential mechanisms. Circ_0078607, miR-196b-5p, and growth arrest-specific 7 (GAS7) levels were assessed by qPCR. Circ_0078607 stability was assessed by ribonuclease R digestion and actinomycin D treatment. Cell viability of various conic of DDP treatment was measured by CCK-8. The cell proliferation was determined by CCK-8 and colony formation assay. Western blotting was performed for determining GAS7, ABCB1, CyclinD1 and Bcl-2 protein levels. The direct binding between miR-196b-5p and circ_0078607 or GAS7 was validated by dual-luciferase reporter and RIP assay. DDP resistance in vivo was evaluated in nude mice. Immunohistochemistry staining for detecting Ki67 expression in xenograft tumours. Circ_0078607 and GAS7 was down-regulated, but miR-196b-5p was up-regulated in OC samples and DDP-resistant cells. Overexpression of circ_0078607 inhibited DDP resistance, cell growth and induced apoptosis in DDP-resistant OC cells. Mechanistically, circ_0078607 sequestered miR-196b-5p to up-regulate GAS7. MiR-196b-5p mimics reversed circ_0078607 or GAS7 overexpression-mediated enhanced sensitivity. Finally, circ_0078607 improved the sensitivity of DDP in vivo. Circ_0078607 attenuates DDP resistance via miR-196b-5p/GAS7 axis, which highlights the therapeutic potential of circ_0078607 to counter DDP resistance in OC. ## Introduction Ovarian cancer (OC) is the main cause of gynaecological-malignancy-related deaths worldwide [1]. Surgical resection combined with platinum-based chemotherapy remains the conventional treatment for OC. Platinum drugs can cause DNA crosslinks and DNA double-strand breaks in infinite proliferous cancer cells [2,3]. Although the platinum therapy has significantly improved the prognosis of OC, some patients have no response to platinum or suffer platinum resistance [4]. About $80\%$ OC patients experience a recurrence within five years due to platinum resistance [5]. The lifecycle for OC patients with platinum resistance is within 12 months [6]. Currently, the molecular mechanisms of platinum resistance remain largely unknown. Therefore, uncovering the underlying mechanisms of platinum resistance is crucial to develop effective strategy for overcoming chemoresistance. Circular RNAs (circRNAs) are a kind of closed single-stranded circular RNAs without coding potential. CircRNAs exert crucial regulation in multiple physiological and pathophysiological processes [7]. More importantly, the dysregulation of circRNAs has been shown to be implicated in chemoresistance. For instance, down-regulation of circRNA Cdr1as led to platinum drug resistance by modulating the miR-1270/SCAI axis in ovarian cancer [8]. Luo et al demonstrated the increased circFoxp1 level was responsible for cisplatin resistance in patients with OC [9]. Circ_0078607, derived from SLC22A3 gene, its expression has been demonstrated to be declined in OC and enforced circ_0078607 expression delayed OC development via miR-518a-5p/Fas pathway [10]. A recent study also reported that lower circ_0078607 level correlated with an adverse prognosis of high-grade OC patients [11]. However, whether circ_0078607 is involved in platinum resistance during OC is worth to be explored. CircRNAs have been recognized to act as competitive endogenous RNAs (ceRNAs) to counteract the inhibitory functions of miRNAs in their target genes [12]. The aberrant expression of circRNA-miRNA-mRNA axis has been documented in various malignancies. CircFN1 functioned as miR-1205 sponge to promote E2F1 expression during sorafenib resistance in hepatocellular carcinoma [13]. Circ_0006528 facilitated paclitaxel resistance via sequestering miR-1299 to activate CDK8 in breast cancer [14]. Interestingly, both circ_0078607 and growth arrest-specific 7 (GAS7) were predicted to have putative binding sites in miR-196b-5p using bioinformatics analysis. MiR-196b-5p can act as a promoter of multiple tumours [15,16]. Repression of GAS7 resulted in enhanced metastatic potential of neuroblastoma [17]. In addition, miR-181a-mediated down-regulation of GAS7 contributed to gefitinib resistance in non-small cell lung cancer [18]. So far, the interaction among circ_0078607, miR-196b-5p and GAS7 during platinum resistance in OC has not been elucidated. This article discovered the declined circ_0078607 level in OC tissues and cisplatin (DDP)-resistant OC cells. Functional studies indicated that overexpression of circ_0078607 sensitized OC cells to DDP via sponging miR-196b-5p to increase GAS7 expression. Uncovering the modulatory role of circ_0078607/miR-196b-5p/GAS7 axis provides an alternative method for reversing OC chemoresistance. ## Clinical samples Forty-eight OC specimens and paired normal tissues were collected from patients diagnosed as OC through surgical resection at Guizhou Provincial People’s Hospital. All OC patients did not receive chemotherapy or radiotherapy before. This study was approved by the Ethics committee of Guizhou Provincial People’s Hospital. Written informed consents were obtained from all participants. ## Cell culture and treatment OC cell lines (A2780 and SKOV3) were obtained from ATCC (Manassas, VA, USA) and normal ovarian epithelial cells (IOSE-80) were purchased from BioVector NTCC (Beijing, China). A2780 and SKOV3 cells were added with escalating concentrations of DDP to acquire DDP-resistant cells (A2780/DDP and SKOV3/DDP). These cells were cultured in RPMI 1640 medium (Thermo Fisher, Waltham, MA, USA) with $10\%$ FBS (Thermo Fisher) under $5\%$ CO2 at 37°C. To maintain the acquired DDP resistance, the culture media was supplemented with 0.3 μg/mL DDP. ## Cell transfection For overexpression (oe), circ_0078607 or GAS7 sequences provided by GenePharma (Shanghai, China) were cloned into pcDNA3.1 to establish oe-circ_0078607 or oe-GAS7 plasmid. miR-196b-5p mimics and mimics negative control (NC) were acquired from GenePharma. All constructed plasmids or segments were transfected into A2780/DDP and SKOV3/DDP cells using Lipofectamine 2000 (Thermo Fisher). ## Quantitative polymerase chain reaction (qPCR) Total RNA was extracted from A2780/DDP and SKOV3/DDP cells using the Total RNA Extractor (Sangon, Shanghai, China). For nuclear and cytoplasmic RNA separation, the Cytoplasmic and Nuclear RNA Purification Kit (Norgen Biotek, Thorold, Canada) was adopted. Then the Prime Script RT Master Mix (Thermo Fisher) was applied to generate cDNA. qPCR was performed using Universal SYBR Green Master Mix (Roche, Basel, Switzerland). The relative gene abundance normalized to GAPDH or U6 was analysed using 2−ΔΔCt method. Primer sequences are as follows. Circ_0078607 F: 5’-GACGCATTGCTAAGTGCAATGG-3’; Circ_0078607 R: 5’- AGGTGAATGCTCCAGTCAGG-3’; SLC22A3 F: 5’- GACGTGGATGACTTGCTACG −3’; SLC22A3 R: 5’- GGCAATTCCAGGGAGAATTA −3’; miR-196b-5p F: 5’- GGCGCTAGGTAGTTTCCTGTT −3’; miR-196b-5p R: 5’- GCAGGGTCCGAGGTATTC −3’; GAS7 F: 5’- CGAGCTACGTGCAGTTGCT −3’; GAS7 R: 5’- CATGTGGGCAGTCTCTGGAG −3’; U6 F: 5’- CTCGCTTCGGCAGCACATATACT-3’; U6 R: 5’- ACGCTTCACGAATTTGCGTGTC-3’; GAPDH F: 5’- CCAGGTGGTCTCCTCTGA-3’; GAPDH R: 5’- GCTGTAGCCAAATCGTTGT-3’. ## Detection for circ_0078607 stability For evaluation the stability of circ_0078607, the cells were treated with 2 mg/mL actinomycin D for 0, 8, 16, and 24 h before RNA extraction. In addition, the isolated RNA was digested with 3 U/μg ribonuclease R (RNase R) for 30 min at 37°C. After that, circ_0078607 and SLC22A3 levels were assessed by qPCR as described above. ## Cell counting kit-8 (CCK-8) For cell viability detection, cells from different groups were exposed to various concentrations of DDP for 48 h and then incubated with 10 μL CCK-8 solution (Acmec Biochemical Co., Ltd, Shanghai, China) for 4 h. The results were detected at 450 nm on a microplate reader. As previously described, the cell viability of DDP treatment was analysed. For proliferation assay, CCK-8 (10 μL) was added to cells at the indicated time points. The following procedure was carried out as described above. ## Colony formation assay Two hundred cells seeded in 6-well plate were maintained for 14 d in culture medium. After fixation in methanol, the colonies were immersed in $0.1\%$ crystal violet solution. Finally, the colonies were photographed and quantified. ## Flow cytometry The apoptosis of cells with multiple treatments was assessed using the Annexin V-FITC apoptosis assay kit (Absin, Shanghai, China). Briefly, the collected cells were stained with 5 μL Annexin V-FITC for 15 min in the dark. Then 5 μL PI solution was added to cells at 5 min before the detection on a flow cytometer (Partec, Germany). ## Western blotting Total protein was acquired using cold RIPA lysis buffer (KEYGENE, Nanjing, China), followed by separation by SDS-PAGE and transferring onto polyvinylidene fluoride membranes. Blocking in $5\%$ skim milk for 1 h was performed and then the membranes were probed with ABCB1 (A00049, 1:200, Boster, Wuhan, China), CyclinD1 (ab16663, 1:200, Abcam, Cambridge, UK), Bcl-2 (ab59348, 1:1000, Abcam), GAS7 (ab168370, 1:1000, Abcam) and GAPDH (A00227, 1:1000, Boster) at 4°C overnight. The secondary antibody was adopted for 30 min. The bands were displayed using ECL Substrates (Applygen, Beijing, China). ## Dual-luciferase reporter assay The wild-type (WT) sequences of circ_0078607 and 3ʹ untranslated region (UTR) of GAS7 containing putative miR-196b-5p binding sites or their mutant (MUT) sequences were inserted into pmirGLO vector. Cells were transfected with the above plasmids together with miR-196b-5p mimics or mimics NC. After culture for 48 h, the relative luciferase activity was assessed using the Luc-Pair™ Duo-Luciferase Assay Kit (Yeasen, Shanghai, China). ## RNA immunoprecipitation (RIP) assay RIP assay was performed using a Magna RNA-Binding Protein Immunoprecipitation Kit (Millipore). In short, A2780/DDP and SKOV3/DDP cells were lysed in RNA immunoprecipitation lysis buffer and probed with magnetic beads conjugated with anti-Ago2 (1:50, Millipore) or IgG (Millipore). After incubation with protease K, circ_0078607 and miR-196b-5p expression was evaluated by qPCR. ## Animal experiment Twenty female BALB/c nude mice were obtained from Vital River Laboratory Animal Technology Co., Ltd (Beijing, China). A2780/DDP cells (1 × 107) stably transfected with oe- circ_0078607 or oe-vector were subcutaneously injected into the mice. Seven days after the injection, DDP (3 mg/kg) or equal volume PBS was intravenously injected into the mice once a week. Tumour volume of each group was calculated every week. Five weeks later, the mice were sacrificed and tumours were removed for further analysis. All procedures were approved by the Ethics Committee of Guizhou Provincial People’s Hospital. ## Immunohistochemistry staining The xenograft tumours were fixed in $4\%$ formalin and embedded in paraffin. After deparaffinage, the tumour sections (5 μm) were received antigen retrieval and blocked in $5\%$ goat serum. The primary antibodies against Ki67 (ab15580, 1:100, Abcam) was applied overnight at 4°C. After incubation with secondary antibody and DAB staining, the images were captured under a microscope. Using ImageJ software, the ki67 positive rate was calculated in five random fields. In each image, the ki67 positive cells and total number of cells per field were counted to determine the positive cell ratio. ## Statistical analysis Data are presented as mean± standard deviation. GraphPad Prism was adopted for statistical analysis using Student’s t test or One-Way ANOVA. A p-value less than 0.05 was defined as statistically significant. ## Down-regulation of circ_0078607 in OC specimens and DDP-resistant OC cells To explore the function of circ_0078607 in DDP resistance, we firstly determined circ_0078607 abundance in OC specimens. As assessed by qPCR, circ_0078607 level in OC samples was obviously lower than control (Figure 1a). In addition, low circ_0078607 expression indicated a decrease in overall survival (Figure 1b) and was significantly associated with clinical stage and venous invasion (Table 1). DDP-resistant OC cells (A2780/DDP and SKOV3/DDP) were more resistant to various DDP concentrations (Figure 1c). Similarly, circ_0078607 expressions were declined in OC cells (A2780 and SKOV3) relative to normal IOSE-80 cells, which was further reduced by DDP resistance (Figure 1d). As shown in Figure 1e&F, circ_0078607 was resistant to actinomycin D or RNase R exposure, while linear SLC22A3 was degraded, suggesting the ultrastability of circ_0078607. Subsequently, circ_0078607 was mostly located in the cytoplasm of A2780/DDP and SKOV3/DDP cells (Figure 1g). Therefore, down-regulation of circ_0078607 might be involved in DDP resistance in OC. Figure 1.Decreased circ_0078607 abundance in OC tissues and DDP-resistant cells. ( a) qPCR analysis of circ_0078607 level in OC clinical samples. ( b) Kaplan-Meier survival plot of OC patients with high or low circ_0078607 level. ( c) The cell viability at different concentrations of DDP treatment was detected by CCK-8. ( d) qPCR analysis of circ_0078607 expressions in various cells. ( e) Cells were exposed to actinomycin D for various time intervals, then circ_0078607 and SLC22A3 levels were assessed by qPCR. ( f) The circ_0078607 and SLC22A3 levels in response to RNase R treatment were detected by qPCR. ( g) qPCR for the nuclear and cytoplasmic expression of circ_0078607. , $P \leq 0.01$; , $P \leq 0.001.$Table 1.The clinicopathological parameters of patients with ovarian cancer. circ_0078607 expression RNAseq ParametersHigh ($$n = 24$$)Low ($$n = 24$$)P valueAge 0.7702≧601513 <60911 Clinical stage 0.0012I/II144 III/IV1020 Venous invasion 0.0189No157 Yes917 Living Status 0.0003Living821 Dead163 ## MiR-196b-5p was a target of circ_0078607 As predicted by StarBase database, circ_0078607 possessed potential binding sites in miR-196b-5p (Figure 2a). Moreover, miR-196b-5p expression was demonstrated to be elevated in OC tissues (Figure 2b). We observed a negative correlation between circ_0078607 and miR-196b-5p (Figure 2c). Consistently, miR-196b-5p expression was elevated in OC cells, and further up-regulated in DDP-resistant cells (Figure 2d). Additionally, RIP assay confirmed the direct interaction between circ_0078607 and miR-196b-5p (Figure 2e). Besides, we found a remarkable decrease in luciferase activity after co-transfection with circ_0078607-WT and miR-196b-5p mimics, while miR-196b-5p mimics did not affect the circ_0078607-MUT, which further validated circ_0078607 directly interacted with miR-196b-5p (figure 2f). Collectively, circ_0078607 could directly sponge miR-196b-5p. Figure 2.Circ_0078607 directly bond to miR-196b-5p. ( a) The binding sites between circ_0078607 and miR-196b-5p. ( b) qPCR for miR-196b-5p level in OC samples. ( c) Pearson correlation analysis of circ_0078607 and miR-196b-5p expression. ( d) qPCR for miR-196b-5p level in different cells. RIP assay (e) and dual-luciferase reporter assay (f) for the interaction between circ_0078607 and miR-196b-5p. , $P \leq 0.01$; *, $P \leq 0.001.$ ## Circ_0078607 increased DDP sensitivity by targeting miR-196b-5p Given that circ_0078607 could directly interact with miR-196b-5p, we further determined whether circ_0078607/miR-196b-5p axis was implicated with DDP resistance of OC cells. Circ_0078607 overexpression led to a reduction in miR-196b-5p expression in A2780/DDP and SKOV3/DDP cells (Figure 3a). The enforced expression of miR-196b-5p was confirmed by qPCR (Figure 3b). Moreover, circ_0078607 overexpression evidently decreased the cell viability, which was overturned by miR-196b-5p mimics (Figure 3c). In addition, the growth of DDP-resistant cells was restrained by circ_0078607 overexpression, whereas miR-196b-5p overexpression abolished these changes (Figure 3d&E). Furthermore, enforced expression of circ_0078607 enhanced the apoptotic rate, which was partially reversed by co-transfection with miR-196b-5p mimics (figure 3f). Furthermore, ABCB1, CyclinD1, and Bcl-2 levels were declined in circ_0078607-overexpressed cells, however, miR-196b-5p mimics could partly restore the decreased expression of ABCB1, CyclinD1, and Bcl-2 (Figure 3g). Therefore, miR-196b-5p participated in circ_0078607-mediated enhanced DDP sensitivity. Figure 3.Circ_0078607 raised DDP sensitivity by sponging miR-196b-5p. ( a) qPCR for circ_0078607 and miR-196b-5p expression after transfection with circ_0078607 overexpression plasmid. ( b) MiR-196b-5p expression was determined by qPCR. ( c) The cell viability at different concentrations of DDP treatment was detected by CCK-8. The proliferation was assessed by CCK-8 (d) and colony formation assay (e). ( f) Flow cytometry for evaluating the apoptosis of cells. ( g) Western blotting assay for determining ABCB1, CyclinD1 and Bcl-2 levels in A2780/DDP and SKOV3/DDP cells. , $P \leq 0.05$; , $P \leq 0.01$; , $P \leq 0.001.$ ## GAS7 was a target of miR-196b-5p Next, the potential target of miR-196b-5p was analysed through StarBase database, and GAS7 was focused on. The potential binding sites between miR-196b-5p and GAS7 were illustrated in Figure 4a. In addition, a decreased expression of GAS7 was validated in OC samples (Figure 4b). Furthermore, GAS7 was negatively correlated with miR-196b-5p expression (Figure 4c), while positively correlated with circ_0078607 expression (Figure 4d). Besides, in comparison with normal IOSE-80 cells, GAS7 was down-regulated in OC cells. More importantly, GAS7 level was even lower in DDP-resistant OC cells (Figure 4e). In addition, GAS7 expression was enhanced by circ_0078607 overexpression (figure 4f), but reduced by miR-196b-5p mimics (Figure 4g). Additionally, miR-196b-5p mimics negatively regulated the luciferase activity of GAS7-WT, rather than GAS7-MUT (Figure 4h). These data suggested that miR-196b-5p could target GAS7 and repress its expression. Figure 4.GAS7 was a down-stream target of miR-196b-5p. ( a) The binding sites of miR-196b-5p in the 3’-UTR of GAS7. ( b) qPCR for GAS7 mRNA expression in OC specimens. Pearson correlation analysis for evaluating the correlation between miR-196b-5p (c) circ_0078607 (d) and GAS7. ( e) GAS7 mRNA expression in multiple cells was detected by qPCR. ( f) qPCR for GAS7 abundance in circ_0078607-overexpressed cells. ( g) GAS7 expression after transfection with miR-196b-5p mimics was assessed by qPCR. ( h) The interaction between GAS7 and miR-196b-5p was determined by dual-luciferase reporter assay. , $P \leq 0.05$; , $P \leq 0.01$; , $P \leq 0.001.$ ## MiR-196b-5p facilitated DDP resistance in OC cells by repressing GAS7 expression To further evaluate whether miR-196b-5p modulated DDP resistance of OC cells via targeting GAS7, cells were co-transfected with miR-196b-5p mimics and pcDNA3.1 GAS7 plasmid. Overexpression of GAS7 was validated by qPCR (Figure 5a). Western blotting further demonstrated that pcDNA3.1 GAS7-mediated overexpression of GAS7 was counteracted by miR-196b-5p mimics (Figure 5b). Subsequently, functional experiments indicated that overexpression of GAS7 decreased cell viability, inhibited proliferation, and induced apoptosis, whereas miR-196b-5p mimics could abolish pcDNA3.1 GAS7-mediated the above modulations (Figure 5c-f). In addition, we found declined ABCB1, CyclinD1, and Bcl-2 levels in GAS7-overexpressed cells, which were evidently abolished by miR-196b-5p mimics (Figure 5g). Thus, miR-196b-5p conferred DDP resistance in OC cells by targeting GAS7. Figure 5.MiR-196b-5p conferred DDP resistance by targeting GAS7. ( a) qPCR for GAS7 mRNA level after transfection with GAS7 overexpression plasmid. ( b) Western blotting for evaluating GAS7 protein level in cells. ( c) CCK-8 assay for determining cell viability at different concentrations of DDP treatment. ( d) CCK-8 and (e) colony formation assays were used to detect the proliferation of cells. ( f) Apoptosis was assessed by flow cytometry. ( g) The protein levels of ABCB1, CyclinD1 and Bcl-2 were assessed by western blotting. , $P \leq 0.05$; , $P \leq 0.01$; , $P \leq 0.001.$ ## Circ_0078607 enhanced the sensitivity of DDP in vivo The regulatory functions of circ_0078607 in DDP resistance were further investigated in nude mice in vivo. In A2780/DDP cells implantation mice tumour tissues, we found that oe-circ 0078607 inhibited miR-196b-5p but promoted GAS7 expression, whereas DDP had no effect on these levels (Figure 6a-b). Furthermore, circ 0078607 overexpression or DDP administration reduced tumour volume and weight, which was more pronounced in the oe-circ 0078607 combination with DDP group (Figure 6c-e). Immunohistochemistry staining revealed a reduced Ki67 expression after overexpression of circ_0078607 or DDP treatment. Combined oe-circ_0078607 with DDP further reinforced this trend (figure 6f). Taken together, overexpression of circ_0078607 increased DDP sensitivity in vivo. Figure 6.Circ_0078607 improved DDP sensitivity in vivo. ( a) qPCR detected circ_0078607, miR-196b-5p, and GAS7 levels in tumour tissue. ( b) Western blotting for evaluating the GAS7 protein level in tumour tissue. ( c) The xenograft tumours of mice from different groups were shown. The growth curve (d) and weight (e) of subcutaneous xenograft tumours. ( f) Immunohistochemistry staining for measuring Ki67 expression in xenograft tumours (200×). , $P \leq 0.05$; *, $P \leq 0.001.$ ## Discussion CircRNAs with prevalent expression in human tissues have been identified as regulators of a wide range of human diseases, including malignancy [19], Parkinson’s disease [20], acute kidney injury [21] and diabetes mellitus [22]. Recently, circ_0078607 has been reported to slow down OC development through miR-518a-5p/Fas pathway [10]. However, the biological regulation of circ_0078607 in OC chemoresistance remains obscure. Our data revealed that circ_0078607 was down-regulated in DDP-resistant OC cells. In addition, circ_0078607 alleviated DDP resistance via regulating miR-196b-5p/GAS7 axis. Our observations suggested circ_0078607 might be an effective target for treating DDP resistant OC. During the last decades, chemotherapeutic drugs have substantially improved the survival of OC patients. DDP, one of platinum agents with broad-spectrum anticancer efficacy, has been widely used for the treatment of OC. Nevertheless, the resistance to DDP remains a key impediment for OC patients to achieve a satisfactory therapeutic efficacy [23]. To date, the roles of circRNAs in DDP resistance have been reported by limited studies [24,25]. In this study, circ_0078607 level was reduced in OC tissues and DDP-resistant OC cells. Next, rescue experiments suggested that overexpression of circ_0078607 reduced cell viability of DDP, restrained growth and induced apoptosis of A2780/DDP and SKOV3/DDP cells. Moreover, xenograft experiments indicated that ectopic expression of circ_0078607 enhanced the sensitivity of DDP in vivo. Therefore, loss of circ_0078607 might promote the progression of OC through conferring DDP resistance. Next, we investigated the regulatory mechanisms of circ_0078607 underlying DDP resistance in OC cells. CircRNAs, distributed in cytoplasm, can function as ceRNAs of miRNAs to compete their activities during the development of different cancers [26]. It has been accepted that circRNAs possess miRNA binding sites, which sequester miRNAs to restrain miRNAs-mediated regulation in their target genes. For instance, circARNT2 facilitated DDP resistance in hepatocellular carcinoma cells through sequestering miR-155-5p to raise PDK1 expression [27]. miRNAs are involved in the development and chemoresistance of malignancies [28,29]. The oncogenic functions of miR-196b-5p have been reported in a variety of cancers, such as non-small cell lung cancer [16], colorectal cancer [15], lung adenocarcinoma [30], and so on. Herein, circ_0078607 was predicted to target miR-196b-5p by bioinformatics analysis. Moreover, we demonstrated the binding between circ_0078607 and miR-196b-5p. Furthermore, up-regulation of miR-196b-5p in OC samples and DDP resistant OC cells was negatively corelated with circ_0078607 level. More importantly, miR-196b-5p mimics reversed circ_0078607-mediated sensitivity to DDP in OC cells. These findings indicated that circ_0078607 affected DDP resistance of OC cells via direct binding to miR-196b-5p. GAS7, as a growth arrest-specific gene, has been reported to hinder the progression of malignant tumours. A previous study indicated the tumour suppressive effect of GAS7 on acute myeloid leukaemia [31]. GAS7 depletion resulted in enhanced metastatic capacity of neuroblastoma [17]. Notably, GAS7 deficiency was implicated in miR-181a-mediated gefitinib resistance in non-small-cell lung cancer [18]. In our study, GAS7 was identified as a novel target of miR-196b-5p. As expected, GAS7 level was lower in OC specimens and DDP-resistant cells. circ_0078607 modulated GAS7 expression via sequestering miR-196b-5p. GAS7-mediated re-sensitivity of DDP could be antagonized by miR-196b-5p mimics. These data indicated that circ_0078607 could regulate DDP sensitivity through miR-196b-5p/GAS7 axis. In conclusion, we demonstrated that circ_0078607 and GAS7 were down-regulated, while miR-196b-5p level was elevated in OC samples and DDP resistant cells. Further results indicated that circ_0078607 sensitized OC cells to DDP via sequestering miR-196b-5p, thereby reinforcing the expression of GAS7. Our observations provide a greater understanding of the elusive mechanisms of DDP resistance, identifying a novel potential treatment option to counter DDP resistance. ## Disclosure statement The authors declare that there is no conflict of interest. ## Authors’ Contribution CD Conceptualization; Writing-original draft; Methodology; Formal analysis; SYD Supervision; YG Validation; TY Data curation; QYZ Resources; ZJL Investigation; XL Software; DND Visualization; DHW Project administration; QFQ Funding acquisition; Writing-review & editing; DZ Funding acquisition; Writing-review & editing. All authors have read and approved the final version of this manuscript to be published. ## Ethical Approval This study was approved by the Ethics committee of Guizhou Provincial People’s Hospital. Written informed consents were obtained from all participants. All procedures were approved by the Ethics Committee of Guizhou Provincial People’s Hospital. ## Consent for Publication The informed consent was obtained from study participants. ## Availability of Data and Material All data generated or analyzed during this study are included in this article. The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request. ## References 1. Chen W, Zheng R, Baade PD. Cancer statistics in China, 2015*. CA Cancer J Clin* (2016) 66 115-13. PMID: 26808342 2. Martin LP, Hamilton TC, Schilder RJ.. Platinum resistance: the role of DNA repair pathways. Clin Cancer Res (2008) 14 1291-1295. PMID: 18316546 3. Rabik CA, Dolan ME. Molecular mechanisms of resistance and toxicity associated with platinating agents. Cancer Treat Rev (2007) 33 9-23. PMID: 17084534 4. Syrios J, Banerjee S, Kaye SB. Advanced epithelial ovarian cancer: from standard chemotherapy to promising molecular pathway targets–where are we now?. Anticancer Res (2014) 34 2069-2077. PMID: 24778008 5. Ryu J, Thomas SN. Quantitative mass spectrometry-based proteomics for biomarker development in ovarian cancer. Molecules (2021) 26 9 6. Elzek MA, Rodland KD. Proteomics of ovarian cancer: functional insights and clinical applications. Cancer Metastasis Rev (2015) 34 83-96. PMID: 25736266 7. Xueyuan H, Qianru C, Zhaoyi L. Transcriptome analysis reveals that hydrogen sulfide exposure suppresses cell proliferation and induces apoptosis through ciR-PTPN23/miR-15a/E2F3 signaling in broiler thymus. Environ Pollut (2021) 284 117466. PMID: 34062439 8. Zhao Z, Ji M, Wang Q. RNA Cdr1as upregulates SCAI to suppress cisplatin resistance in ovarian cancer via miR-1270 suppression. Mol Ther Nucleic Acids (2019) 18 24-33. PMID: 31479922 9. Luo Y, Gui R. Circulating exosomal circFoxp1 confers cisplatin resistance in epithelial ovarian cancer cells. J Gynecol Oncol (2020) 31 e75. PMID: 32808501 10. Zhang N, Jin Y, Hu Q. Circular RNA hsa_circ_0078607 suppresses ovarian cancer progression by regulating miR-518a-5p/Fas signaling pathway. J Ovarian Res (2020) 13 64. PMID: 32503653 11. Zhang N, Yang Z, Jin Y. Low expression of circular RNA hsa_circ_0078607 predicts poor prognosis in high-grade serous ovarian cancer. Cancer Manag Res (2021) 13 2877-2883. PMID: 33833570 12. Zheng Q, Bao C, Guo W. Circular RNA profiling reveals an abundant circHIPK3 that regulates cell growth by sponging multiple miRNAs. Nat Commun (2016) 7 11215. PMID: 27050392 13. Yang C, Dong Z, Hong H. circFN1 mediates sorafenib resistance of hepatocellular carcinoma cells by sponging miR-1205 and regulating E2F1 expression. Mol Ther Nucleic Acids (2020) 22 421-433. PMID: 33230446 14. Liu G, Zhang Z, Song Q. Circ_0006528 contributes to paclitaxel resistance of breast cancer cells by regulating miR-1299/CDK8 axis. Onco Targets Ther (2020) 13 9497-9511. PMID: 33061434 15. Xin H, Wang C, Chi Y. MicroRNA-196b-5p promotes malignant progression of colorectal cancer by targeting ING5. Cancer Cell Int (2020) 20 119. PMID: 32308564 16. Liang G, Meng W, Huang X. miR-196b-5p-mediated downregulation of TSPAN12 and GATA6 promotes tumor progression in non-small cell lung cancer. Proc Natl Acad Sci U S A (2020) 117 4347-4357. PMID: 32041891 17. Dong Z, Yeo KS, Lopez G. GAS7 deficiency promotes metastasis in MYCN-driven neuroblastoma. Cancer Res (2021) 81 2995-3007. PMID: 33602789 18. Ping W, Gao Y, Fan X. MiR-181a contributes gefitinib resistance in non-small cell lung cancer cells by targeting GAS7. Biochem Biophys Res Commun (2018) 495 2482-2489. PMID: 29269300 19. Zhou Z, Zhang Y, Gao J. Circular RNAs act as regulators of autophagy in cancer. Mol Ther Oncolytics (2021) 21 242-254. PMID: 34095462 20. RNA-sequencing of peripheral blood circular RNAs in Parkinson disease. Medicine (Baltimore) (2021) 100 e25888. PMID: 34114985 21. Tan M, Bei R. Circ_0091702 serves as a sponge of miR-545-3p to attenuate sepsis-related acute kidney injury by upregulating THBS2. J Mol Histol (2021;) 52 717-728. PMID: 34101064 22. Samavarchi Tehrani S, Goodarzi G, Panahi G. Multiple novel functions of circular RNAs in diabetes mellitus. Arch Physiol Biochem (2021) Jun 4 1-30 23. Duan L, Yan Y, Wang G. MuiR-182-5p functions as a tumor suppressor to sensitize human ovarian cancer cells to cisplatin through direct targeting the cyclin dependent kinase 6 (CDK6). J BUON (2020) 25 2279-2286. PMID: 33277846 24. Wang Y, Li L, Zhang W. Thorac Cancer (2021) 1959-1972 25. Sun Y, Ma J, Lin J. Circular RNA circ_ASAP2 regulates drug sensitivity and functional behaviors of cisplatin-resistant gastric cancer cells by the miR-330-3p/NT5E axis. Anticancer Drugs (2021) 32 950-961. PMID: 34016832 26. Cui X, Wang J, Guo Z. Emerging function and potential diagnostic value of circular RNAs in cancer. Mol Cancer (2018) 17 123. PMID: 30111313 27. Li Y, Zhang Y, Zhang S. circRNA circARNT2 suppressed the sensitivity of hepatocellular carcinoma cells to cisplatin by targeting the miR-155-5p/PDK1 axis. Mol Ther Nucleic Acids (2021) 23 244-254. PMID: 33425483 28. Bao Y, Zhang Y, Lu Y. Overexpression of microRNA-9 enhances cisplatin sensitivity in hepatocellular carcinoma by regulating EIF5A2-mediated epithelial-mesenchymal transition. Int J Biol Sci (2020) 16 827-837. PMID: 32071552 29. Wang Y, Yang J, Chen P. MicroRNA-320a inhibits invasion and metastasis in osteosarcoma by targeting cytoplasmic polyadenylation element-binding protein 1. Cancer Med (2020) 9 2833-2845. PMID: 32064777 30. Xu Q, Xu Z. miR-196b-5p promotes proliferation, migration and invasion of lung adenocarcinoma cells via targeting RSPO2. Cancer Manag Res (2020) 12 13393-13402. PMID: 33402849 31. Wu F, Yin C, Qi J. miR-362-5p promotes cell proliferation and cell cycle progression by targeting GAS7 in acute myeloid leukemia. Hum Cell (2020) 33 405-415. PMID: 31925702

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.